AgiBot Web Caption vs Cosmos3-Super

10 clipped segments, generated 2026-06-11T20:03:26+0000. Videos are cut using each row's start_sec/end_sec before captioning.

agibot_web_000

top

Webpage Caption

Captured from the robot's left-hand camera view (hand_left), the video shows a bimanual manipulation task involving a green plastic bottle with a yellow cap on a white fabric surface, likely a couch. At the start, the robot's right hand is seen gripping the bottle's neck from above, while the robot's left hand, visible in the foreground as a black gripper, supports the base of the bottle. As the robot releases its grip, the bottle loses stability and tips over to the left, sliding slightly across the white surface. The robot's left hand then moves away and then back toward the fallen bottle. The right hand re-enters the frame from the top to grasp the yellow cap, while the left hand simultaneously moves under the base of the bottle to push it back into a vertical position. This coordinated bimanual effort successfully rights the bottle. By the end of the clip, the robot releases the bottle once more; the right hand moves upward and the left hand retreats, leaving the green bottle standing upright on the surface. A yellow and black pouch remains stationary in the background throughout the sequence. The camera motion is tied to the movement of the left arm, causing the viewpoint to shift and tilt as the left hand approaches and retreats from the bottle. The final state shows the robot's left hand in the foreground, the right hand out of view, and the bottle standing stable on the white surface.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "A robot uses its left and right hands to manipulate a green plastic bottle on a white surface. The robot initially holds the bottle with its right hand and stabilizes it with its left hand, then releases it, and finally attempts to grasp it again.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:01",
        "actions": [
          "The robot's right hand grips the top of the green bottle",
          "The robot's left hand supports the bottom of the bottle"
        ],
        "interactions": [
          "Right hand pinch grip on the bottle neck",
          "Left hand cradle support on the bottle base"
        ],
        "causal": [
          "Bimanual coordination stabilizes the bottle in an upright position"
        ]
      },
      {
        "time": "00:00:01–00:00:03",
        "actions": [
          "The robot releases the bottle",
          "The robot's left hand moves away from the bottle",
          "The bottle tilts and falls over"
        ],
        "interactions": [
          "Right hand releases grip",
          "Left hand breaks contact with the base"
        ],
        "causal": [
          "Loss of support causes the bottle to tip over onto its side"
        ]
      },
      {
        "time": "00:00:03–00:00:06",
        "actions": [
          "The robot's left hand moves toward the fallen bottle",
          "The robot's right hand moves out of the frame"
        ],
        "interactions": [
          "Left hand approaches the side of the bottle"
        ],
        "causal": [
          "The robot attempts to re-engage with the displaced object"
        ]
      },
      {
        "time": "00:00:06–00:00:11",
        "actions": [
          "The robot's right hand returns to grip the bottle neck",
          "The robot's left hand supports the bottle base again",
          "The robot lifts the bottle slightly"
        ],
        "interactions": [
          "Right hand grips the yellow cap area",
          "Left hand contacts the bottom of the bottle"
        ],
        "causal": [
          "Bimanual grip restores the bottle to a vertical orientation"
        ]
      },
      {
        "time": "00:00:11–00:00:13",
        "actions": [
          "The robot releases the bottle again",
          "The robot's left hand moves away"
        ],
        "interactions": [
          "Right hand releases the cap",
          "Left hand breaks contact with the base"
        ],
        "causal": [
          "The bottle remains standing on the surface"
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The green bottle is centered on a white couch/surface",
        "A yellow pouch is located behind the bottle",
        "The robot's left hand is in the foreground, the right hand approaches from above"
      ],
      "temporal": [
        "Initial bimanual hold -> release and fall -> re-grasping -> final release"
      ],
      "comparisons": [
        "The bottle is upright at the start and end, but horizontal in the middle of the clip"
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "Robot",
          "desc": "To practice grasping, releasing, and recovering a cylindrical object"
        }
      ],
      "theme": "Object manipulation and recovery",
      "social": "None"
    }
  },
  "prediction": {
    "next": "The robot may move the bottle to a different location or attempt to pick it up and move it entirely.",
    "outcome": "The bottle will likely remain stationary on the white surface unless acted upon again.",
    "past": "The robot likely placed the bottle on the surface or was instructed to pick it up.",
    "what_if": "If the robot had not used its left hand for support, the bottle might have tipped over more quickly during the first release."
  },
  "meta": {
    "style": "Wrist-mounted robot camera view with some fisheye distortion and motion blur."
  },
  "final_dense_caption": "Captured from the robot's left-hand camera view (hand_left), the video shows a bimanual manipulation task involving a green plastic bottle with a yellow cap on a white fabric surface, likely a couch. At the start, the robot's right hand is seen gripping the bottle's neck from above, while the robot's left hand, visible in the foreground as a black gripper, supports the base of the bottle. As the robot releases its grip, the bottle loses stability and tips over to the left, sliding slightly across the white surface. The robot's left hand then moves away and then back toward the fallen bottle. The right hand re-enters the frame from the top to grasp the yellow cap, while the left hand simultaneously moves under the base of the bottle to push it back into a vertical position. This coordinated bimanual effort successfully rights the bottle. By the end of the clip, the robot releases the bottle once more; the right hand moves upward and the left hand retreats, leaving the green bottle standing upright on the surface. A yellow and black pouch remains stationary in the background throughout the sequence. The camera motion is tied to the movement of the left arm, causing the viewpoint to shift and tilt as the left hand approaches and retreats from the bottle. The final state shows the robot's left hand in the foreground, the right hand out of view, and the bottle standing stable on the white surface.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:01:12+0000

Captured from a first-person perspective mounted on a robotic end-effector, the scene unfolds indoors on a white surface, likely a couch, under bright top-lit conditions casting soft shadows. A black and metallic robotic gripper dominates the foreground, initially positioned above a transparent green plastic bottle with a yellow cap, though the specific brand or contents remain uncertain due to unreadable label text. The gripper descends to make contact, securely grasping the bottle before lifting it slightly against gravity. Maintaining stability, the end-effector translates the bottle to the left side of the surface near the couch edge. Upon placement, the gripper releases the object, allowing it to rest upright, and subsequently retracts to its initial hovering position. Throughout the sequence, the camera moves synchronously with the manipulator, maintaining a close-up, low-angle view with a shallow depth of field that keeps the background slightly blurred. In the distance, a person wearing dark pants and sneakers with red accents stands partially visible near a white appliance, remaining mostly stationary. The task concludes with the bottle relocated from the center to the left, while the gripper returns to its starting state, demonstrating successful manipulation despite the undefined context of the operation.

Rewrite raw output
{
  "rewritten_free_form_caption": "Captured from a first-person perspective mounted on a robotic end-effector, the scene unfolds indoors on a white surface, likely a couch, under bright top-lit conditions casting soft shadows. A black and metallic robotic gripper dominates the foreground, initially positioned above a transparent green plastic bottle with a yellow cap, though the specific brand or contents remain uncertain due to unreadable label text. The gripper descends to make contact, securely grasping the bottle before lifting it slightly against gravity. Maintaining stability, the end-effector translates the bottle to the left side of the surface near the couch edge. Upon placement, the gripper releases the object, allowing it to rest upright, and subsequently retracts to its initial hovering position. Throughout the sequence, the camera moves synchronously with the manipulator, maintaining a close-up, low-angle view with a shallow depth of field that keeps the background slightly blurred. In the distance, a person wearing dark pants and sneakers with red accents stands partially visible near a white appliance, remaining mostly stationary. The task concludes with the bottle relocated from the center to the left, while the gripper returns to its starting state, demonstrating successful manipulation despite the undefined context of the operation.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:03:15+0000 · display: free_form_caption

The video captures a first-person perspective of a robotic gripper manipulating a green bottle on a white surface. The gripper moves down to grasp the bottle, lifts it, moves it to the left, places it back on the surface, and then retracts to its initial position. The background shows a person standing in the distance, observing the task. The video demonstrates the capabilities of the robotic gripper in performing precise manipulation tasks.

Raw free_form_caption
The video captures a first-person perspective of a robotic gripper manipulating a green bottle on a white surface. The gripper moves down to grasp the bottle, lifts it, moves it to the left, places it back on the surface, and then retracts to its initial position. The background shows a person standing in the distance, observing the task. The video demonstrates the capabilities of the robotic gripper in performing precise manipulation tasks.
Cosmos structured JSON
{
  "scene_imagination": "The video is captured from a first-person perspective, likely from a camera mounted on a robotic arm or end-effector. The scene takes place indoors, possibly in a living room or similar space, with a white couch or sofa visible. A green plastic bottle with a yellow cap is the primary object of interest. A robotic gripper is positioned to interact with the bottle. In the background, a person wearing dark pants and sneakers is partially visible, standing near a white appliance or container. The floor is light-colored, and there are some other objects, including a yellow and black item, near the couch.",
  "temporal_caption": "The video begins with the robotic gripper positioned above the green bottle on the white surface. The gripper then moves down to grasp the bottle. After securing the bottle, the gripper lifts it slightly and then moves it to the left, placing it on the white surface near the edge of the couch. The gripper releases the bottle and then moves back to its initial position. The person in the background remains mostly stationary throughout the sequence.",
  "audio_description": "The video does not contain any discernible audio elements. There are no sounds of movement, speech, or background noise that can be clearly identified.",
  "subjects": [
    {
      "description": "A green plastic bottle with a yellow cap, likely containing a beverage.",
      "appearance_details": "The bottle is transparent green with a yellow cap and has some text and graphics on its label.",
      "relationship": "The bottle is the primary object being manipulated by the robotic gripper.",
      "location": "Initially on the white surface, then moved to the left side of the surface.",
      "relative_size": "Medium within the frame.",
      "orientation": "Upright initially, then tilted slightly during manipulation.",
      "pose": "Standing upright on the surface.",
      "action": "The bottle is grasped by the robotic gripper, lifted, moved to the left, and placed back on the surface.",
      "state_changes": "The bottle changes position from the center of the surface to the left side."
    },
    {
      "description": "A robotic gripper with black and metallic components.",
      "appearance_details": "The gripper has a black base with metallic joints and fingers designed to grasp objects.",
      "relationship": "The gripper interacts with the green bottle, manipulating its position.",
      "location": "Positioned above and around the green bottle.",
      "relative_size": "Large within the frame, dominating the foreground.",
      "orientation": "Facing downwards towards the bottle.",
      "pose": "Extended towards the bottle, then retracting after manipulation.",
      "action": "The gripper moves down to grasp the bottle, lifts it, moves it to the left, and releases it.",
      "state_changes": "The gripper changes position from above the bottle to holding it, then moving it to the left."
    },
    {
      "description": "A person wearing dark pants and sneakers, partially visible in the background.",
      "appearance_details": "The person is wearing dark pants and sneakers with red accents.",
      "relationship": "The person is standing in the background, observing the robotic manipulation.",
      "location": "In the background, to the right of the frame.",
      "relative_size": "Small within the frame, partially visible.",
      "orientation": "Facing towards the robotic gripper and the bottle.",
      "pose": "Standing upright.",
      "action": "The person remains mostly stationary, observing the robotic manipulation.",
      "state_changes": "No significant change in position or pose."
    }
  ],
  "background_setting": "The scene takes place indoors, likely in a living room or similar space. A white couch or sofa is visible, with a light-colored blanket or cloth draped over it. The floor is light-colored, and there are some other objects, including a yellow and black item, near the couch. A white appliance or container is visible in the background, to the right of the frame.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-lit, casting soft shadows downwards.",
    "shadows": "Soft shadows are visible on the white surface and around the objects.",
    "illumination_effect": "The lighting highlights the objects clearly, with minimal glare or overexposure."
  },
  "aesthetics": {
    "composition": "The composition is centered on the robotic gripper and the green bottle, with the background elements providing context.",
    "color_scheme": "The color scheme is dominated by the green of the bottle, the white of the surface and couch, and the black of the robotic gripper.",
    "mood_atmosphere": "The mood is neutral and focused on the robotic manipulation task.",
    "patterns": "There are no prominent patterns visible in the scene."
  },
  "cinematography": {
    "camera_motion": "The camera is mounted on the robotic gripper, providing a first-person perspective. The camera moves with the gripper as it manipulates the bottle.",
    "framing": "The framing is close-up, focusing on the robotic gripper and the green bottle.",
    "camera_angle": "The camera angle is low, looking down at the surface and the objects.",
    "depth_of_field": "The depth of field is relatively shallow, with the foreground objects in sharp focus and the background slightly blurred.",
    "focus": "The focus is on the robotic gripper and the green bottle, with the background elements slightly out of focus.",
    "lens_focal_length": "The lens appears to be a wide-angle lens, capturing a broad view of the immediate surroundings."
  },
  "style_medium": "The video is a real-world recording of a robotic manipulation task, likely captured using a camera mounted on the robotic gripper.",
  "artistic_style": "The video has a documentary or instructional style, focusing on the robotic manipulation task.",
  "context": "The video appears to be a demonstration or recording of a robotic manipulation task, where the robotic gripper is used to move a green bottle from one position to another on a white surface.",
  "actions": [
    {
      "time": "0-2 seconds",
      "description": "The robotic gripper moves down towards the green bottle on the white surface.",
      "actor": "Robotic gripper",
      "object": "Green bottle",
      "contact_event": "The gripper makes contact with the bottle and begins to grasp it.",
      "state_changes": "The gripper moves from above the bottle to grasping it."
    },
    {
      "time": "2-5 seconds",
      "description": "The robotic gripper lifts the green bottle slightly and moves it to the left.",
      "actor": "Robotic gripper",
      "object": "Green bottle",
      "contact_event": "The gripper holds the bottle securely as it moves it to the left.",
      "state_changes": "The bottle is moved from the center of the surface to the left side."
    },
    {
      "time": "5-7 seconds",
      "description": "The robotic gripper places the green bottle back on the white surface and releases it.",
      "actor": "Robotic gripper",
      "object": "Green bottle",
      "contact_event": "The gripper releases the bottle, allowing it to rest on the surface.",
      "state_changes": "The bottle is placed back on the surface, and the gripper retracts."
    },
    {
      "time": "7-9 seconds",
      "description": "The robotic gripper moves back to its initial position above the surface.",
      "actor": "Robotic gripper",
      "object": "None",
      "contact_event": "The gripper retracts and moves back to its starting position.",
      "state_changes": "The gripper returns to its initial position."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "Unreadable text on the green bottle label",
      "category": "physical_in_scene",
      "appearance": "Small, printed text on the green bottle label.",
      "spatial": "On the green bottle, visible throughout the video.",
      "spatial_temporal": "Static on the bottle, visible throughout the video.",
      "context": "The text is part of the bottle's label, likely indicating the brand or contents."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0-2 seconds",
      "description": "The robotic gripper moves down towards the green bottle on the white surface.",
      "key_changes": "The gripper moves from above the bottle to grasping it.",
      "camera": "The camera is mounted on the gripper, moving down with it.",
      "robot_motion": "The gripper moves down to grasp the bottle.",
      "object_motion": "The bottle remains stationary until grasped.",
      "physical_logic": "The gripper makes contact with the bottle and begins to grasp it."
    },
    {
      "segment_index": 1,
      "time_range": "2-5 seconds",
      "description": "The robotic gripper lifts the green bottle slightly and moves it to the left.",
      "key_changes": "The bottle is moved from the center of the surface to the left side.",
      "camera": "The camera moves with the gripper as it manipulates the bottle.",
      "robot_motion": "The gripper lifts the bottle and moves it to the left.",
      "object_motion": "The bottle is lifted and moved to the left.",
      "physical_logic": "The gripper holds the bottle securely as it moves it to the left."
    },
    {
      "segment_index": 2,
      "time_range": "5-7 seconds",
      "description": "The robotic gripper places the green bottle back on the white surface and releases it.",
      "key_changes": "The bottle is placed back on the surface, and the gripper retracts.",
      "camera": "The camera moves with the gripper as it places the bottle and retracts.",
      "robot_motion": "The gripper releases the bottle and retracts.",
      "object_motion": "The bottle is placed back on the surface.",
      "physical_logic": "The gripper releases the bottle, allowing it to rest on the surface."
    },
    {
      "segment_index": 3,
      "time_range": "7-9 seconds",
      "description": "The robotic gripper moves back to its initial position above the surface.",
      "key_changes": "The gripper returns to its initial position.",
      "camera": "The camera moves with the gripper as it retracts.",
      "robot_motion": "The gripper retracts and moves back to its starting position.",
      "object_motion": "No significant object motion.",
      "physical_logic": "The gripper completes its task and returns to its initial position."
    }
  ],
  "transitions": [
    {
      "transition": "The gripper moves down to grasp the bottle.",
      "description": "The gripper transitions from a position above the bottle to grasping it.",
      "type": "Continuous motion"
    },
    {
      "transition": "The gripper lifts and moves the bottle to the left.",
      "description": "The gripper transitions from grasping the bottle to moving it to the left.",
      "type": "Continuous motion"
    },
    {
      "transition": "The gripper places the bottle back on the surface and releases it.",
      "description": "The gripper transitions from holding the bottle to releasing it.",
      "type": "Continuous motion"
    },
    {
      "transition": "The gripper moves back to its initial position.",
      "description": "The gripper transitions from releasing the bottle to returning to its starting position.",
      "type": "Continuous motion"
    }
  ],
  "resolution": {
    "W": 640,
    "H": 528
  },
  "aspect_ratio": "16,9",
  "duration": "8.9",
  "fps": 30.0,
  "robotics": {
    "embodiment": "The robotic arm is equipped with a gripper that can grasp and manipulate objects.",
    "end_effector": "The end-effector is a black and metallic gripper designed to grasp objects.",
    "manipulated_objects": "The green bottle is the primary object being manipulated.",
    "task_goal": "The task goal is to move the green bottle from one position to another on the white surface.",
    "initial_state": "The green bottle is stationary on the white surface.",
    "terminal_state": "The green bottle is placed back on the white surface, and the gripper returns to its initial position.",
    "affordances": "The gripper can grasp and move objects, demonstrating its ability to manipulate objects in the environment.",
    "failure_modes": "The gripper may fail to grasp the bottle securely, or it may drop the bottle during manipulation."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The gripper makes contact with the bottle and grasps it securely.",
    "support_and_stability": "The gripper provides support to the bottle during manipulation, ensuring stability.",
    "force_gravity_friction": "The gripper applies force to lift and move the bottle, overcoming gravity and friction.",
    "causal_chain": "The gripper moves down to grasp the bottle, lifts it, moves it to the left, places it back on the surface, and releases it.",
    "object_permanence": "The bottle remains a consistent object throughout the manipulation task.",
    "next_action_prediction": "The next action would likely involve the gripper moving to a new position or preparing for another manipulation task."
  },
  "uncertainties": [
    {
      "uncertainty": "The exact brand or contents of the green bottle are not clearly visible.",
      "reason": "The text on the bottle label is unreadable."
    },
    {
      "uncertainty": "The purpose or context of the robotic manipulation task is not explicitly stated.",
      "reason": "The video focuses on the manipulation task without providing additional context."
    }
  ],
  "free_form_caption": "The video captures a first-person perspective of a robotic gripper manipulating a green bottle on a white surface. The gripper moves down to grasp the bottle, lifts it, moves it to the left, places it back on the surface, and then retracts to its initial position. The background shows a person standing in the distance, observing the task. The video demonstrates the capabilities of the robotic gripper in performing precise manipulation tasks."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms100
cosmos_terms124
intersection38
jaccard0.204

agibot_web_001

top

Webpage Caption

Captured from the robot's top_head camera view, the scene shows a red-clothed table laden with various instant noodle products, including bags and stacked cups. At the start, the robot's left hand is visible in the lower-left foreground, remaining completely stationary throughout the clip. The right hand is positioned to the right, approaching a stack of noodle cups. The right hand moves forward and descends, with its gripper closing around the rim of the top cup, which has a bright green lid. As the right hand lifts the cup vertically, the remaining cups in the stack shift slightly due to the change in weight and friction. After securing the grip, the right hand moves the cup laterally and backward, pulling it away from the other products. Throughout the motion, the camera view shifts slightly, reflecting the robot's bodily movement. By the end of the sequence, the right hand is holding the green-lidded cup in the upper-right quadrant of the frame, having successfully removed it from the stack. The left hand remains in its initial position, and the other noodle products on the table remain largely undisturbed, though the stack from which the cup was taken is now one unit shorter. The lighting is bright and overhead, typical of a convenience store or lab setting, and some motion blur is visible as the right arm moves quickly.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "An AgiBot robot uses its right hand to pick up a cup of instant noodles from a stack on a red table, while its left hand remains stationary in the foreground. The robot successfully lifts the top cup, causing the stack to shift slightly, and moves the cup away from the group.",
    "genre": "robot retail-object handling"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:01",
        "actions": [
          "right hand approaches the top of a noodle cup stack"
        ],
        "interactions": [
          "right hand moves toward the green-lidded noodle cup"
        ],
        "causal": [
          "the right hand's approach prepares for grasping"
        ]
      },
      {
        "time": "00:00:01–00:00:03",
        "actions": [
          "right hand closes its gripper around the rim of the top noodle cup",
          "right hand lifts the cup vertically"
        ],
        "interactions": [
          "gripper makes contact with the plastic rim of the green cup"
        ],
        "causal": [
          "lifting the top cup causes the remaining cups in the stack to settle and shift slightly"
        ]
      },
      {
        "time": "00:00:03–00:00:06",
        "actions": [
          "right hand moves the grasped cup laterally and slightly backward",
          "left hand remains static"
        ],
        "interactions": [
          "the cup is held securely in the right gripper"
        ],
        "causal": [
          "the movement clears the cup from the original stack"
        ]
      },
      {
        "time": "00:00:06–00:00:11",
        "actions": [
          "right hand continues to move the cup across the workspace",
          "camera perspective shifts slightly as the robot body moves"
        ],
        "interactions": [
          "the cup remains held in the right hand"
        ],
        "causal": [
          "the robot is relocating the item to a different area of the table"
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The red table is the primary workspace.",
        "The left hand is positioned in the lower-left foreground.",
        "The noodle cup stacks are located in the center and right-center of the frame.",
        "The right hand operates in the center-right area."
      ],
      "temporal": [
        "The approach happens first, followed by the grasp, the lift, and finally the relocation."
      ],
      "comparisons": [
        "The right hand is active and moving, while the left hand is passive and stationary."
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "right hand",
          "desc": "To pick up and move a specific cup of instant noodles from a stack."
        }
      ],
      "theme": "Object manipulation and sorting in a retail environment.",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The robot will likely place the noodle cup in a designated area or hand it to a person.",
    "outcome": "The cup will be successfully relocated without being dropped.",
    "past": "The robot likely identified the target cup using its vision system before initiating the reach.",
    "what_if": "If the gripper had not closed tightly, the cup might have slipped and knocked over the rest of the stack."
  },
  "meta": {
    "style": "top-head robot view with some motion blur during fast movements"
  },
  "final_dense_caption": "Captured from the robot's top_head camera view, the scene shows a red-clothed table laden with various instant noodle products, including bags and stacked cups. At the start, the robot's left hand is visible in the lower-left foreground, remaining completely stationary throughout the clip. The right hand is positioned to the right, approaching a stack of noodle cups. The right hand moves forward and descends, with its gripper closing around the rim of the top cup, which has a bright green lid. As the right hand lifts the cup vertically, the remaining cups in the stack shift slightly due to the change in weight and friction. After securing the grip, the right hand moves the cup laterally and backward, pulling it away from the other products. Throughout the motion, the camera view shifts slightly, reflecting the robot's bodily movement. By the end of the sequence, the right hand is holding the green-lidded cup in the upper-right quadrant of the frame, having successfully removed it from the stack. The left hand remains in its initial position, and the other noodle products on the table remain largely undisturbed, though the stack from which the cup was taken is now one unit shorter. The lighting is bright and overhead, typical of a convenience store or lab setting, and some motion blur is visible as the right arm moves quickly.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:01:24+0000

A stationary high-angle camera provides a top-down view of a workspace illuminated by bright indoor lighting with soft shadows cast across the surface. The scene centers on a table covered with a red tablecloth, scattered with various packaged food items including bags of snacks and yellow price tags. Two black robot arms equipped with mechanical grippers are positioned on opposite sides of the frame, facing the center. Initially, the left robot arm extends from the left side towards the center to make contact with stacks of colorful instant noodle cups. Subsequently, the right robot arm moves from the right side towards the center to interact with the same objects. Both arms grasp the cups, applying force to overcome gravity and friction while supporting the objects against the table surface. Throughout the sequence, the robot arms manipulate and rearrange the vertical stacks of cups, shifting their configurations. The cups remain on the table throughout the interaction, maintaining object permanence. By the conclusion of the twenty-second segment, the robot arms have completed the rearrangement task, leaving the stacks in new positions on the red cloth. The deep depth of field keeps all objects in sharp focus as the mechanical grippers execute precise movements to reorganize the food items.

Rewrite raw output
{
  "rewritten_free_form_caption": "A stationary high-angle camera provides a top-down view of a workspace illuminated by bright indoor lighting with soft shadows cast across the surface. The scene centers on a table covered with a red tablecloth, scattered with various packaged food items including bags of snacks and yellow price tags. Two black robot arms equipped with mechanical grippers are positioned on opposite sides of the frame, facing the center. Initially, the left robot arm extends from the left side towards the center to make contact with stacks of colorful instant noodle cups. Subsequently, the right robot arm moves from the right side towards the center to interact with the same objects. Both arms grasp the cups, applying force to overcome gravity and friction while supporting the objects against the table surface. Throughout the sequence, the robot arms manipulate and rearrange the vertical stacks of cups, shifting their configurations. The cups remain on the table throughout the interaction, maintaining object permanence. By the conclusion of the twenty-second segment, the robot arms have completed the rearrangement task, leaving the stacks in new positions on the red cloth. The deep depth of field keeps all objects in sharp focus as the mechanical grippers execute precise movements to reorganize the food items.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:03:32+0000 · display: free_form_caption

The video shows two robot arms manipulating stacks of instant noodle cups on a red tablecloth. The robot arms move the cups, rearranging them into different stacks. The camera remains stationary, providing a top-down view of the scene.

Raw free_form_caption
The video shows two robot arms manipulating stacks of instant noodle cups on a red tablecloth. The robot arms move the cups, rearranging them into different stacks. The camera remains stationary, providing a top-down view of the scene.
Cosmos structured JSON
{
  "scene_imagination": "Two robot arms with black grippers are positioned on either side of a red tablecloth-covered surface, manipulating stacks of colorful instant noodle cups. The table is filled with various packaged food items, including stacks of cups and bags of snacks. Yellow price tags are visible on the table. The camera is positioned above, providing a top-down view of the scene.",
  "temporal_caption": "The video shows two robot arms interacting with stacks of instant noodle cups on a red tablecloth. The robot arms move the cups, rearranging them into different stacks. The camera remains stationary, providing a top-down view of the scene.",
  "audio_description": "The video does not contain any audio.",
  "subjects": [
    {
      "description": "Robot Arm 1, positioned on the left side of the frame, with a black gripper.",
      "appearance_details": "The robot arm is black with a mechanical gripper at the end.",
      "relationship": "Interacting with the stacks of instant noodle cups.",
      "location": "Left side of the frame.",
      "relative_size": "Large within the frame.",
      "orientation": "Facing the center of the table.",
      "pose": "Extended towards the center of the table.",
      "action": "Moving and manipulating the stacks of instant noodle cups.",
      "state_changes": "The robot arm moves from the left side of the table to the center, interacting with the cups."
    },
    {
      "description": "Robot Arm 2, positioned on the right side of the frame, with a black gripper.",
      "appearance_details": "The robot arm is black with a mechanical gripper at the end.",
      "relationship": "Interacting with the stacks of instant noodle cups.",
      "location": "Right side of the frame.",
      "relative_size": "Large within the frame.",
      "orientation": "Facing the center of the table.",
      "pose": "Extended towards the center of the table.",
      "action": "Moving and manipulating the stacks of instant noodle cups.",
      "state_changes": "The robot arm moves from the right side of the table to the center, interacting with the cups."
    },
    {
      "description": "Stacks of instant noodle cups, arranged in various configurations on the red tablecloth.",
      "appearance_details": "The cups are colorful, with various designs and labels.",
      "relationship": "Being manipulated by the robot arms.",
      "location": "Center of the table.",
      "relative_size": "Medium within the frame.",
      "orientation": "Stacked vertically.",
      "pose": "Standing upright on the table.",
      "action": "Being moved and rearranged by the robot arms.",
      "state_changes": "The stacks of cups are moved and rearranged by the robot arms."
    }
  ],
  "background_setting": "The scene takes place on a table covered with a red tablecloth. Various packaged food items, including stacks of instant noodle cups and bags of snacks, are arranged on the table. Yellow price tags are visible on the table.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-down lighting.",
    "shadows": "Soft shadows cast by the objects on the table.",
    "illumination_effect": "Even illumination across the scene."
  },
  "aesthetics": {
    "composition": "Centered composition, with the robot arms and stacks of cups as the main focus.",
    "color_scheme": "Colorful, with the red tablecloth and various colors of the food packaging.",
    "mood_atmosphere": "Functional and mechanical.",
    "patterns": "The stacks of cups create a repeating pattern on the table."
  },
  "cinematography": {
    "camera_motion": "The camera remains stationary throughout the video.",
    "framing": "Top-down view of the table.",
    "camera_angle": "High angle, looking down at the table.",
    "depth_of_field": "Deep depth of field, with all objects in focus.",
    "focus": "Sharp focus on the robot arms and stacks of cups.",
    "lens_focal_length": "Standard lens."
  },
  "style_medium": "Robotics video.",
  "artistic_style": "Realistic.",
  "context": "The video shows a demonstration of robot arms manipulating stacks of instant noodle cups on a table.",
  "actions": [
    {
      "time": "0.0-5.0s",
      "description": "Robot Arm 1 moves from the left side of the table towards the center, interacting with the stacks of instant noodle cups.",
      "actor": "Robot Arm 1",
      "object": "Stacks of instant noodle cups",
      "contact_event": "The robot arm makes contact with the cups.",
      "state_changes": "The robot arm moves the cups."
    },
    {
      "time": "5.0-10.0s",
      "description": "Robot Arm 2 moves from the right side of the table towards the center, interacting with the stacks of instant noodle cups.",
      "actor": "Robot Arm 2",
      "object": "Stacks of instant noodle cups",
      "contact_event": "The robot arm makes contact with the cups.",
      "state_changes": "The robot arm moves the cups."
    },
    {
      "time": "10.0-15.0s",
      "description": "Both robot arms continue to manipulate the stacks of instant noodle cups, rearranging them on the table.",
      "actor": "Robot Arm 1 and Robot Arm 2",
      "object": "Stacks of instant noodle cups",
      "contact_event": "The robot arms make contact with the cups.",
      "state_changes": "The stacks of cups are rearranged."
    },
    {
      "time": "15.0-20.0s",
      "description": "The robot arms continue to manipulate the stacks of instant noodle cups, completing the rearrangement.",
      "actor": "Robot Arm 1 and Robot Arm 2",
      "object": "Stacks of instant noodle cups",
      "contact_event": "The robot arms make contact with the cups.",
      "state_changes": "The stacks of cups are rearranged."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "9.9元3桶",
      "category": "physical_in_scene",
      "appearance": "Yellow price tag with black text.",
      "spatial": "Bottom right of the frame.",
      "spatial_temporal": "Static throughout the video.",
      "context": "Price tag for the instant noodle cups."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0.0-5.0s",
      "description": "Robot Arm 1 moves from the left side of the table towards the center, interacting with the stacks of instant noodle cups.",
      "key_changes": "Robot Arm 1 moves the cups.",
      "camera": "Static top-down view.",
      "robot_motion": "Robot Arm 1 moves from left to center.",
      "object_motion": "The cups are moved by Robot Arm 1."
    },
    {
      "segment_index": 1,
      "time_range": "5.0-10.0s",
      "description": "Robot Arm 2 moves from the right side of the table towards the center, interacting with the stacks of instant noodle cups.",
      "key_changes": "Robot Arm 2 moves the cups.",
      "camera": "Static top-down view.",
      "robot_motion": "Robot Arm 2 moves from right to center.",
      "object_motion": "The cups are moved by Robot Arm 2."
    },
    {
      "segment_index": 2,
      "time_range": "10.0-15.0s",
      "description": "Both robot arms continue to manipulate the stacks of instant noodle cups, rearranging them on the table.",
      "key_changes": "The stacks of cups are rearranged.",
      "camera": "Static top-down view.",
      "robot_motion": "Both robot arms move the cups.",
      "object_motion": "The cups are rearranged."
    },
    {
      "segment_index": 3,
      "time_range": "15.0-20.0s",
      "description": "The robot arms continue to manipulate the stacks of instant noodle cups, completing the rearrangement.",
      "key_changes": "The stacks of cups are rearranged.",
      "camera": "Static top-down view.",
      "robot_motion": "Both robot arms move the cups.",
      "object_motion": "The cups are rearranged."
    }
  ],
  "transitions": [
    "No significant transitions.",
    "No significant transitions.",
    "No significant transitions.",
    "No significant transitions."
  ],
  "resolution": {
    "W": 640,
    "H": 400
  },
  "aspect_ratio": "4,3",
  "duration": "20.00s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "Two robot arms with black grippers.",
    "end_effector": "Mechanical grippers.",
    "manipulated_objects": "Stacks of instant noodle cups.",
    "task_goal": "Rearranging the stacks of instant noodle cups.",
    "initial_state": "The robot arms are positioned on either side of the table, with the stacks of cups arranged in various configurations.",
    "terminal_state": "The robot arms have rearranged the stacks of cups.",
    "affordances": "The robot arms are capable of grasping and moving the cups.",
    "failure_modes": "None observed."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robot arms make contact with the cups and grasp them.",
    "support_and_stability": "The cups are supported by the table and the robot arms.",
    "force_gravity_friction": "The robot arms apply force to move the cups, overcoming gravity and friction.",
    "causal_chain": "The robot arms move the cups, causing them to be rearranged.",
    "object_permanence": "The cups remain on the table throughout the video.",
    "next_action_prediction": "The robot arms will continue to manipulate the cups."
  },
  "uncertainties": [],
  "free_form_caption": "The video shows two robot arms manipulating stacks of instant noodle cups on a red tablecloth. The robot arms move the cups, rearranging them into different stacks. The camera remains stationary, providing a top-down view of the scene."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms107
cosmos_terms115
intersection28
jaccard0.144

agibot_web_002

top

Webpage Caption

Captured from the robot's top_head camera view, the sequence begins with the robot positioned in an indoor environment featuring beige chairs and a white bucket on the floor. The bucket contains several bottles, including one with a green cap. At the start, the robot's left hand, a black multi-fingered gripper, reaches forward and dips into the bucket, making contact with the bottles. The right hand remains open and stationary to the right of the bucket. As the robot moves, the camera viewpoint shifts, transitioning from the floor level up toward a white tabletop. The left hand retracts from the bucket and moves across the frame toward a small, dark rectangular object resting on the white table. Simultaneously, the right hand moves inward from the right side of the frame. Both hands converge on the object in a coordinated bimanual approach. By the end of the clip, the left hand is positioned to the left of the object and the right hand to the right, both with their grippers open and ready to grasp. The object remains stationary on the table. The lighting is bright and even, and the movement of the robot's body causes a slight panning effect in the top_head view. The final state shows both hands poised around the target object, having successfully transitioned from the floor-based task to the tabletop workspace.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "An AgiBot robot uses its bimanual grippers to interact with a white bucket containing bottles and then moves its hands toward a small object on a white table.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:02",
        "actions": [
          "The robot's left hand reaches into a white bucket",
          "The robot's right hand remains open and stationary to the right of the bucket"
        ],
        "interactions": [
          "Left hand fingertips contact the contents of the bucket"
        ],
        "causal": [
          "The left hand's movement causes slight displacement of the bottles inside the bucket"
        ]
      },
      {
        "time": "00:00:02–00:00:05",
        "actions": [
          "The left hand retracts from the bucket",
          "The robot's body/camera shifts position, moving the view toward a white table"
        ],
        "interactions": [
          "Left hand releases contact with the bucket contents"
        ],
        "causal": [
          "The camera movement changes the workspace from the floor/bucket area to the tabletop"
        ]
      },
      {
        "time": "00:00:05–00:00:08",
        "actions": [
          "The left hand moves toward a small rectangular object on the white table",
          "The right hand moves toward the same object"
        ],
        "interactions": [
          "Both hands approach the object without making contact"
        ],
        "causal": [
          "The coordinated movement of both hands centers the object in the workspace"
        ]
      },
      {
        "time": "00:00:08–00:00:11",
        "actions": [
          "The left hand opens its gripper",
          "The right hand opens its gripper"
        ],
        "interactions": [
          "Hands maintain a close distance to the object but do not grasp it"
        ],
        "causal": [
          "The grippers prepare for a potential pinch or lift action"
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The white bucket is on the floor between the robot's hands initially",
        "The white table is positioned in front of the robot in the second half of the clip",
        "The small object is centered on the white table"
      ],
      "temporal": [
        "Bucket interaction precedes the tabletop approach"
      ],
      "comparisons": [
        "The left hand is more active in the first phase, while both hands coordinate in the second phase"
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "Robot",
          "desc": "To transition from exploring/interacting with a container on the floor to manipulating a specific object on a table."
        }
      ],
      "theme": "bimanual object approach",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The robot will likely use one or both hands to grasp and lift the small rectangular object from the table.",
    "outcome": "The object will be successfully lifted or moved across the table surface.",
    "past": "The robot was likely positioned in front of the chairs and bucket, preparing to clear the area or retrieve an item.",
    "what_if": "If the robot's coordination is off, the grippers might collide with each other or push the object off the table instead of grasping it."
  },
  "meta": {
    "style": "top-head robot view with slight motion blur during transitions"
  },
  "final_dense_caption": "Captured from the robot's top_head camera view, the sequence begins with the robot positioned in an indoor environment featuring beige chairs and a white bucket on the floor. The bucket contains several bottles, including one with a green cap. At the start, the robot's left hand, a black multi-fingered gripper, reaches forward and dips into the bucket, making contact with the bottles. The right hand remains open and stationary to the right of the bucket. As the robot moves, the camera viewpoint shifts, transitioning from the floor level up toward a white tabletop. The left hand retracts from the bucket and moves across the frame toward a small, dark rectangular object resting on the white table. Simultaneously, the right hand moves inward from the right side of the frame. Both hands converge on the object in a coordinated bimanual approach. By the end of the clip, the left hand is positioned to the left of the object and the right hand to the right, both with their grippers open and ready to grasp. The object remains stationary on the table. The lighting is bright and even, and the movement of the robot's body causes a slight panning effect in the top_head view. The final state shows both hands poised around the target object, having successfully transitioned from the floor-based task to the tabletop workspace.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:01:36+0000

Viewed from a stable top-head first-person perspective, the brightly lit indoor environment features a white floor and beige chairs in the background. Two black robotic arms with visible mechanical joints, metallic components, and wires extend into the frame, each ending in a complex multi-fingered gripper. A white bucket with a black rim sits stationary on the floor to the left, containing various items including a green bottle with a yellow cap, while a white table with a smooth surface occupies the right side as a workspace. Initially, the left arm grasps a black object near the bucket, while the right arm remains positioned near the table. As the sequence progresses, the left arm manipulates the black object, moving it away from the bucket and transporting it towards the table surface. Simultaneously, the right arm shifts closer to the table, aligning with the left arm. Soft shadows cast by the top-lit illumination highlight the details of the arms and objects. By the end of the segment, both arms are positioned near the table, having completed the transfer of the black object from the bucket vicinity to the workspace area. The camera maintains a consistent deep depth of field, keeping the robot's actions and immediate surroundings in focus throughout the manipulation task. The physical logic demonstrates the arms applying force to overcome gravity and friction while maintaining object permanence during the movement.

Rewrite raw output
{
  "rewritten_free_form_caption": "Viewed from a stable top-head first-person perspective, the brightly lit indoor environment features a white floor and beige chairs in the background. Two black robotic arms with visible mechanical joints, metallic components, and wires extend into the frame, each ending in a complex multi-fingered gripper. A white bucket with a black rim sits stationary on the floor to the left, containing various items including a green bottle with a yellow cap, while a white table with a smooth surface occupies the right side as a workspace. Initially, the left arm grasps a black object near the bucket, while the right arm remains positioned near the table. As the sequence progresses, the left arm manipulates the black object, moving it away from the bucket and transporting it towards the table surface. Simultaneously, the right arm shifts closer to the table, aligning with the left arm. Soft shadows cast by the top-lit illumination highlight the details of the arms and objects. By the end of the segment, both arms are positioned near the table, having completed the transfer of the black object from the bucket vicinity to the workspace area. The camera maintains a consistent deep depth of field, keeping the robot's actions and immediate surroundings in focus throughout the manipulation task. The physical logic demonstrates the arms applying force to overcome gravity and friction while maintaining object permanence during the movement.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:03:53+0000 · display: free_form_caption

The video shows a robot's first-person view as it manipulates objects in a room. The robot's left arm holds a black object near a white bucket filled with various items, while the right arm is positioned near a white table. The left arm moves the black object away from the bucket and towards the table, while the right arm moves closer to the table. Both arms are then positioned near the table, preparing to manipulate objects on it. The scene is well-lit, with a focus on the robot's actions and the objects in the environment.

Raw free_form_caption
The video shows a robot's first-person view as it manipulates objects in a room. The robot's left arm holds a black object near a white bucket filled with various items, while the right arm is positioned near a white table. The left arm moves the black object away from the bucket and towards the table, while the right arm moves closer to the table. Both arms are then positioned near the table, preparing to manipulate objects on it. The scene is well-lit, with a focus on the robot's actions and the objects in the environment.
Cosmos structured JSON
{
  "scene_imagination": "The video depicts a first-person view from a robot's perspective, showing its two robotic arms manipulating objects in a room. The setting includes a white table, a white bucket containing various items, and beige chairs in the background. The robot's arms are black with mechanical joints and grippers. Initially, the left arm is holding a black object near the bucket, while the right arm is positioned near the table. The robot appears to be moving objects around, possibly sorting or organizing them. The lighting is bright, illuminating the scene clearly, and the camera remains relatively stable, focusing on the robot's actions and the immediate surroundings.",
  "temporal_caption": "The video begins with the robot's left arm holding a black object near a white bucket filled with various items. The right arm is positioned near a white table. As the video progresses, the left arm moves the black object away from the bucket and towards the table. The right arm then moves closer to the table, and both arms appear to be preparing to manipulate objects on the table. The camera angle remains consistent, providing a clear view of the robot's actions and the surrounding environment.",
  "audio_description": "The video does not contain any discernible audio elements. The focus is on the visual actions of the robot and the objects in the scene.",
  "subjects": [
    {
      "description": "A robotic arm with a black, mechanical design, featuring multiple joints and a gripper at the end.",
      "appearance_details": "The arm is primarily black with some metallic components and wires visible. The gripper has a complex structure with multiple fingers.",
      "relationship": "Interacting with objects in the scene, specifically a black object and items in a white bucket.",
      "location": "Left side of the frame, moving towards the center.",
      "relative_size": "Large within the frame.",
      "orientation": "Facing towards the center of the scene.",
      "pose": "Extended towards the objects, with the gripper open or closed depending on the action.",
      "action": "Manipulating a black object and moving it towards the table.",
      "state_changes": "The arm moves from holding the black object near the bucket to moving it towards the table."
    },
    {
      "description": "Another robotic arm similar in design to the first, with a black, mechanical structure and a gripper.",
      "appearance_details": "This arm is also primarily black with metallic components and wires. The gripper has a similar complex structure to the first arm.",
      "relationship": "Positioned near the table, possibly preparing to interact with objects on it.",
      "location": "Right side of the frame, moving towards the center.",
      "relative_size": "Large within the frame.",
      "orientation": "Facing towards the center of the scene.",
      "pose": "Extended towards the table, with the gripper open or closed depending on the action.",
      "action": "Moving towards the table, possibly to manipulate objects on it.",
      "state_changes": "The arm moves from a position near the table to a more central position."
    },
    {
      "description": "A white bucket containing various items, including a green bottle and other objects.",
      "appearance_details": "The bucket is white with a black rim and contains a mix of items, including a green bottle with a yellow cap.",
      "relationship": "Located near the robot's left arm, serving as a container for the objects being manipulated.",
      "location": "Center of the frame, slightly to the left.",
      "relative_size": "Medium within the frame.",
      "orientation": "Upright, with the opening facing upwards.",
      "pose": "Stationary on the floor.",
      "action": "Holding the objects being manipulated by the robot.",
      "state_changes": "No significant change in the bucket's position or state."
    },
    {
      "description": "A white table with a smooth surface, serving as a workspace for the robot.",
      "appearance_details": "The table is white with a flat surface, providing a clear area for the robot to manipulate objects.",
      "relationship": "Located near the robot's right arm, serving as a workspace for the robot.",
      "location": "Right side of the frame, moving towards the center.",
      "relative_size": "Large within the frame.",
      "orientation": "Flat, with the surface facing upwards.",
      "pose": "Stationary on the floor.",
      "action": "Serving as a workspace for the robot.",
      "state_changes": "No significant change in the table's position or state."
    }
  ],
  "background_setting": "The scene takes place in a room with a white floor and beige chairs in the background. The room appears to be a workspace or laboratory setting, with a focus on the robot's actions and the objects it is manipulating. The lighting is bright, illuminating the scene clearly.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-lit, providing even illumination across the scene.",
    "shadows": "Soft shadows cast by the robot's arms and objects on the floor and table.",
    "illumination_effect": "The lighting highlights the details of the robot's arms and the objects in the scene, creating a clear and well-lit environment."
  },
  "aesthetics": {
    "composition": "The scene is composed with the robot's arms and the objects they are manipulating as the central focus, with the background elements providing context.",
    "color_scheme": "The color scheme is primarily black, white, and beige, with some green and yellow accents from the objects in the bucket.",
    "mood_atmosphere": "The mood is focused and task-oriented, with the robot's actions being the main point of interest.",
    "patterns": "The mechanical design of the robot's arms creates a pattern of lines and joints, adding visual interest to the scene."
  },
  "cinematography": {
    "camera_motion": "The camera remains relatively stable, providing a consistent view of the robot's actions and the surrounding environment.",
    "framing": "The camera is framed to capture the robot's arms and the objects they are manipulating, with the background elements providing context.",
    "camera_angle": "The camera angle is from the robot's perspective, providing a first-person view of the scene.",
    "depth_of_field": "The depth of field is relatively deep, with both the robot's arms and the background elements in focus.",
    "focus": "The focus is on the robot's arms and the objects they are manipulating, with the background elements slightly less sharp.",
    "lens_focal_length": "The lens appears to be a standard focal length, providing a natural and undistorted view of the scene."
  },
  "style_medium": "The video is a real-world robot manipulation footage, capturing the robot's actions and the surrounding environment in a clear and detailed manner.",
  "artistic_style": "The artistic style is realistic and functional, focusing on the robot's actions and the objects it is manipulating.",
  "context": "The video appears to be a demonstration of a robot's manipulation capabilities, showing how it can interact with objects in a controlled environment.",
  "actions": [
    {
      "time": "0-5 seconds",
      "description": "The robot's left arm holds a black object near a white bucket filled with various items. The right arm is positioned near a white table.",
      "actor": "Robot's left arm.",
      "object": "Black object.",
      "contact_event": "The left arm is holding the black object.",
      "state_changes": "The left arm moves the black object away from the bucket."
    },
    {
      "time": "5-10 seconds",
      "description": "The left arm continues to move the black object towards the table. The right arm moves closer to the table.",
      "actor": "Robot's left arm.",
      "object": "Black object.",
      "contact_event": "The left arm moves the black object towards the table.",
      "state_changes": "The left arm moves the black object towards the table."
    },
    {
      "time": "10-15 seconds",
      "description": "Both arms are now positioned near the table, preparing to manipulate objects on it.",
      "actor": "Robot's arms.",
      "object": "Table.",
      "contact_event": "The arms are positioned near the table.",
      "state_changes": "The arms are positioned near the table."
    }
  ],
  "text_and_signage_elements": [],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0-5 seconds",
      "description": "The robot's left arm holds a black object near a white bucket filled with various items. The right arm is positioned near a white table.",
      "key_changes": "The left arm moves the black object away from the bucket.",
      "camera": "The camera remains relatively stable, providing a consistent view of the robot's actions and the surrounding environment.",
      "robot_motion": "The left arm moves the black object away from the bucket.",
      "object_motion": "The black object is moved away from the bucket.",
      "physical_logic": "The robot's left arm is manipulating the black object, moving it away from the bucket."
    },
    {
      "segment_index": 1,
      "time_range": "5-10 seconds",
      "description": "The left arm continues to move the black object towards the table. The right arm moves closer to the table.",
      "key_changes": "The left arm moves the black object towards the table.",
      "camera": "The camera remains relatively stable, providing a consistent view of the robot's actions and the surrounding environment.",
      "robot_motion": "The left arm moves the black object towards the table.",
      "object_motion": "The black object is moved towards the table.",
      "physical_logic": "The robot's left arm is manipulating the black object, moving it towards the table."
    },
    {
      "segment_index": 2,
      "time_range": "10-15 seconds",
      "description": "Both arms are now positioned near the table, preparing to manipulate objects on it.",
      "key_changes": "The arms are positioned near the table.",
      "camera": "The camera remains relatively stable, providing a consistent view of the robot's actions and the surrounding environment.",
      "robot_motion": "The arms are positioned near the table.",
      "object_motion": "The arms are positioned near the table.",
      "physical_logic": "The robot's arms are preparing to manipulate objects on the table."
    }
  ],
  "transitions": [
    {
      "transition": "The robot's left arm moves the black object away from the bucket.",
      "time": "0-5 seconds"
    },
    {
      "transition": "The left arm continues to move the black object towards the table.",
      "time": "5-10 seconds"
    },
    {
      "transition": "Both arms are now positioned near the table.",
      "time": "10-15 seconds"
    }
  ],
  "resolution": {
    "W": 640,
    "H": 400
  },
  "aspect_ratio": "16,9",
  "duration": "14.900s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "The robot has two arms with mechanical joints and grippers, designed for manipulating objects.",
    "end_effector": "The end effectors are grippers with multiple fingers, capable of grasping and manipulating objects.",
    "manipulated_objects": "The robot manipulates a black object and interacts with items in a white bucket and on a white table.",
    "task_goal": "The robot appears to be sorting or organizing objects, moving them from the bucket to the table.",
    "initial_state": "The robot's left arm is holding a black object near a white bucket filled with various items. The right arm is positioned near a white table.",
    "terminal_state": "Both arms are positioned near the table, preparing to manipulate objects on it.",
    "affordances": "The robot's arms have the ability to grasp and manipulate objects, move them from one location to another, and interact with the environment.",
    "failure_modes": "The robot may fail to grasp or manipulate objects correctly, or may not be able to move them to the desired location."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robot's grippers make contact with the objects, allowing it to grasp and manipulate them.",
    "support_and_stability": "The robot's arms provide support and stability, allowing it to move objects without dropping them.",
    "force_gravity_friction": "The robot's arms apply force to move objects, overcoming gravity and friction.",
    "causal_chain": "The robot's actions cause the objects to move from one location to another, changing their state and position.",
    "object_permanence": "The objects remain in the scene, even as they are moved by the robot.",
    "next_action_prediction": "The robot will likely continue to manipulate objects on the table, possibly sorting or organizing them."
  },
  "uncertainties": [],
  "free_form_caption": "The video shows a robot's first-person view as it manipulates objects in a room. The robot's left arm holds a black object near a white bucket filled with various items, while the right arm is positioned near a white table. The left arm moves the black object away from the bucket and towards the table, while the right arm moves closer to the table. Both arms are then positioned near the table, preparing to manipulate objects on it. The scene is well-lit, with a focus on the robot's actions and the objects in the environment."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms95
cosmos_terms120
intersection38
jaccard0.215

agibot_web_003

top

Webpage Caption

Captured from the robot's top_head camera view, the scene shows a retail environment with wooden shelves stocked with various snack packages. At the start, the robot's right hand, a black gripper attached to a white arm, is positioned in the center-right of the frame, approaching a white package with red accents on the top shelf. The left hand is visible on the far left, remaining stationary and not participating in the task. As the right hand closes its gripper around the package, it fails to achieve a secure pinch. The package is momentarily displaced and then slips through the gripper's fingers. Consequently, the package falls vertically, landing on the grey tiled floor among other fallen items, including a yellow and green package. After the object is dropped, the right hand continues its motion, moving upward and slightly toward the center of the frame, effectively retracting from the shelf area. By the end of the clip, the right hand is positioned higher in the frame, empty, while the left hand remains in its original static position. The surrounding packages on the shelves remain largely undisturbed, though the target package now rests on the floor. The camera remains relatively stable, though there is slight movement corresponding to the robot's arm motion, providing a first-person perspective of the failed manipulation attempt.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "An AgiBot robot uses its right hand to attempt to grasp a white and red snack package from a retail shelf, but fails to secure it, causing the package to fall to the floor. The left hand remains stationary and partially visible on the left side of the frame.",
    "genre": "robot retail-object handling"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:01",
        "actions": [
          "right hand approaches a white and red package on the top shelf"
        ],
        "interactions": [
          "right hand makes initial contact with the package"
        ],
        "causal": [
          "the package is pushed slightly as the gripper closes"
        ]
      },
      {
        "time": "00:00:01–00:00:03",
        "actions": [
          "right hand attempts to lift the package"
        ],
        "interactions": [
          "the package slips from the gripper's grasp"
        ],
        "causal": [
          "the lack of a secure grip causes the package to fall"
        ]
      },
      {
        "time": "00:00:03–00:00:05",
        "actions": [
          "right hand moves away from the shelf",
          "right hand begins to retract"
        ],
        "interactions": [
          "the package lands on the floor"
        ],
        "causal": [
          "gravity pulls the released package to the ground"
        ]
      },
      {
        "time": "00:00:05–00:00:07",
        "actions": [
          "right hand moves upward and toward the center"
        ],
        "interactions": [
          "no further contact with objects"
        ],
        "causal": [
          "the robot resets its arm position"
        ]
      }
    ],
    "relations": {
      "spatial": [
        "the right hand is the primary actor in the center-right foreground",
        "the left hand is stationary on the far left",
        "the shelves are in the background",
        "the fallen package is on the floor below the shelf"
      ],
      "temporal": [
        "the approach happens first, followed by the failed grasp, then the fall, and finally the arm retraction"
      ],
      "comparisons": [
        "the right hand is dynamic while the left hand is static"
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "right hand",
          "desc": "to pick up a specific snack package from the shelf"
        }
      ],
      "theme": "failed object manipulation",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The robot may attempt to pick up the package from the floor or try to grab another item from the shelf.",
    "outcome": "The package remains on the floor unless the robot recovers it.",
    "past": "The robot likely identified the target package and moved its arm into position to grasp it.",
    "what_if": "If the gripper had a higher friction surface or a more precise closure, the package would have been successfully lifted."
  },
  "meta": {
    "style": "top-head robot view"
  },
  "final_dense_caption": "Captured from the robot's top_head camera view, the scene shows a retail environment with wooden shelves stocked with various snack packages. At the start, the robot's right hand, a black gripper attached to a white arm, is positioned in the center-right of the frame, approaching a white package with red accents on the top shelf. The left hand is visible on the far left, remaining stationary and not participating in the task. As the right hand closes its gripper around the package, it fails to achieve a secure pinch. The package is momentarily displaced and then slips through the gripper's fingers. Consequently, the package falls vertically, landing on the grey tiled floor among other fallen items, including a yellow and green package. After the object is dropped, the right hand continues its motion, moving upward and slightly toward the center of the frame, effectively retracting from the shelf area. By the end of the clip, the right hand is positioned higher in the frame, empty, while the left hand remains in its original static position. The surrounding packages on the shelves remain largely undisturbed, though the target package now rests on the floor. The camera remains relatively stable, though there is slight movement corresponding to the robot's arm motion, providing a first-person perspective of the failed manipulation attempt.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:01:48+0000

From a static top-down perspective, a high-angle medium shot captures a retail environment featuring a wooden shelf densely stocked with colorful packaged food items, primarily instant noodles and snacks, against a speckled gray tile floor. Bright, even indoor lighting illuminates the scene, highlighting the varied packaging colors including red, white, blue, yellow, and green. An articulated robotic arm with a white cylindrical body, black joints, and a black multi-fingered gripper extends from the bottom right of the frame towards the center shelf. The robot's base is partially visible at the bottom center. The arm reaches upward and slightly leftwards, making contact with a specific red and white packaged item on the middle shelf. The gripper closes around the package, successfully grasping it and lifting it slightly against gravity before holding it briefly. Subsequently, the gripper opens to release the package back onto the shelf, returning it to its original position among the other static items. Finally, the robotic arm retracts slightly, moving away from the shelf towards a resting position. Throughout the sequence, the camera remains fixed with a wide-angle lens and deep depth of field, keeping both the foreground robotic arm and background shelf in sharp focus. The physical interaction demonstrates successful grasp and release mechanics without observed failure modes, concluding with the arm in a retracted state and the package stable on the shelf.

Rewrite raw output
{
  "rewritten_free_form_caption": "From a static top-down perspective, a high-angle medium shot captures a retail environment featuring a wooden shelf densely stocked with colorful packaged food items, primarily instant noodles and snacks, against a speckled gray tile floor. Bright, even indoor lighting illuminates the scene, highlighting the varied packaging colors including red, white, blue, yellow, and green. An articulated robotic arm with a white cylindrical body, black joints, and a black multi-fingered gripper extends from the bottom right of the frame towards the center shelf. The robot's base is partially visible at the bottom center. The arm reaches upward and slightly leftwards, making contact with a specific red and white packaged item on the middle shelf. The gripper closes around the package, successfully grasping it and lifting it slightly against gravity before holding it briefly. Subsequently, the gripper opens to release the package back onto the shelf, returning it to its original position among the other static items. Finally, the robotic arm retracts slightly, moving away from the shelf towards a resting position. Throughout the sequence, the camera remains fixed with a wide-angle lens and deep depth of field, keeping both the foreground robotic arm and background shelf in sharp focus. The physical interaction demonstrates successful grasp and release mechanics without observed failure modes, concluding with the arm in a retracted state and the package stable on the shelf.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:04:08+0000 · display: free_form_caption

A top-down view shows a robotic arm with a black gripper extending from the bottom right to grasp a red and white packaged item from a supermarket shelf. The arm lifts the package slightly, then releases it back onto the shelf before retracting slightly. The scene is brightly lit, with various colorful packages visible on the wooden shelf and a speckled gray tile floor below.

Raw free_form_caption
A top-down view shows a robotic arm with a black gripper extending from the bottom right to grasp a red and white packaged item from a supermarket shelf. The arm lifts the package slightly, then releases it back onto the shelf before retracting slightly. The scene is brightly lit, with various colorful packages visible on the wooden shelf and a speckled gray tile floor below.
Cosmos structured JSON
{
  "scene_imagination": "A top-down view of a supermarket shelf stocked with various packaged food items, primarily instant noodles and snacks in colorful bags. A robotic arm with a black gripper extends from the bottom right of the frame, reaching toward the shelf. The floor is a speckled gray tile. The robot's base is partially visible at the bottom center. The lighting is bright and even, typical of a retail environment.",
  "temporal_caption": "The robotic arm extends upward, grasps a red and white packaged item from the middle shelf, lifts it slightly, then releases it back onto the shelf before retracting slightly.",
  "audio_description": "Ambient supermarket noise, possibly faint background music or HVAC hum, with subtle mechanical sounds from the robotic arm's movement.",
  "subjects": [
    {
      "description": "A robotic arm with a white cylindrical body and a black, multi-fingered gripper at the end.",
      "appearance_details": "White body with black joints and a black gripper mechanism. The arm is articulated with visible segments.",
      "relationship": "Interacting with the packaged items on the shelf.",
      "location": "Extending from the bottom right towards the center of the frame.",
      "relative_size": "Large within the frame, dominating the right side.",
      "orientation": "Extending upwards and slightly leftwards towards the shelf.",
      "pose": "Articulated, reaching out.",
      "action": "Reaching for, grasping, and releasing a packaged item on the shelf.",
      "state_changes": "Starts extended, moves to grasp an item, then retracts slightly."
    },
    {
      "description": "A collection of packaged food items, primarily instant noodles and snacks, arranged on a wooden shelf.",
      "appearance_details": "Various colors and designs on the packaging, including red, white, blue, yellow, and green. Some packages are upright, others are lying flat.",
      "relationship": "Being manipulated by the robotic arm.",
      "location": "Center of the frame, on the shelf.",
      "relative_size": "Medium to small individual items, forming a dense cluster.",
      "orientation": "Various orientations, some upright, some flat.",
      "pose": "Static, except for the one being manipulated.",
      "action": "One item is lifted and released by the robotic arm.",
      "state_changes": "One red and white package is briefly lifted and then placed back."
    }
  ],
  "background_setting": "A supermarket or convenience store aisle. The background consists of a wooden shelf filled with various packaged food items. The floor is a speckled gray tile. The lighting is bright and even, typical of a retail environment.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-down, even illumination.",
    "shadows": "Soft shadows cast by the packages and the robotic arm.",
    "illumination_effect": "Evenly illuminates the scene, highlighting the colors of the packaging."
  },
  "aesthetics": {
    "composition": "Centered on the interaction between the robotic arm and the shelf.",
    "color_scheme": "Colorful packaging against a neutral gray floor and wooden shelf.",
    "mood_atmosphere": "Functional, mechanical, retail environment.",
    "patterns": "Speckled pattern on the floor tiles."
  },
  "cinematography": {
    "camera_motion": "Static.",
    "framing": "Top-down view, medium shot.",
    "camera_angle": "High angle, looking down at the shelf and robotic arm.",
    "depth_of_field": "Deep, with both the foreground (robotic arm) and background (shelf) in focus.",
    "focus": "Sharp focus on the robotic arm and the items on the shelf.",
    "lens_focal_length": "Wide-angle, capturing a broad view of the shelf and the robotic arm."
  },
  "style_medium": "Real-world video.",
  "artistic_style": "Documentary, functional.",
  "context": "A demonstration of a robotic arm performing a manipulation task in a retail environment, likely for inventory management or restocking.",
  "actions": [
    {
      "time": "0.0 - 2.0 seconds",
      "description": "The robotic arm extends towards the shelf and grasps a red and white packaged item.",
      "actor": "Robotic arm",
      "object": "Red and white packaged item",
      "contact_event": "Gripper makes contact with the package and closes around it.",
      "state_change": "The package is lifted slightly from the shelf."
    },
    {
      "time": "2.0 - 4.0 seconds",
      "description": "The robotic arm holds the package briefly, then releases it back onto the shelf.",
      "actor": "Robotic arm",
      "object": "Red and white packaged item",
      "contact_event": "Gripper opens, releasing the package.",
      "state_change": "The package returns to its original position on the shelf."
    },
    {
      "time": "4.0 - 6.6 seconds",
      "description": "The robotic arm retracts slightly, moving away from the shelf.",
      "actor": "Robotic arm",
      "object": "None",
      "contact_event": "No contact.",
      "state_change": "The arm moves back to a resting position."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "Various text on packaging",
      "category": "physical_in_scene",
      "appearance": "Small, printed text in various colors and fonts on the packages.",
      "spatial": "Distributed across the packages on the shelf.",
      "spatial_temporal": "Static throughout the video.",
      "context": "Product labels and branding."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0.0 - 2.0 seconds",
      "description": "The robotic arm reaches out and grasps a package from the shelf.",
      "key_changes": "The arm moves from a resting position to grasping the package.",
      "camera": "Static top-down view.",
      "robot_motion": "Extending and grasping.",
      "object_motion": "Package is lifted slightly.",
      "physical_logic": "The gripper closes around the package, lifting it."
    },
    {
      "segment_index": 1,
      "time_range": "2.0 - 4.0 seconds",
      "description": "The robotic arm releases the package back onto the shelf.",
      "key_changes": "The package returns to its original position.",
      "camera": "Static top-down view.",
      "robot_motion": "Holding and releasing.",
      "object_motion": "Package is placed back.",
      "physical_logic": "The gripper opens, releasing the package."
    },
    {
      "segment_index": 2,
      "time_range": "4.0 - 6.6 seconds",
      "description": "The robotic arm retracts slightly.",
      "key_changes": "The arm moves back to a resting position.",
      "camera": "Static top-down view.",
      "robot_motion": "Retracting.",
      "object_motion": "None.",
      "physical_logic": "The arm moves back to its initial position."
    }
  ],
  "transitions": [
    "Continuous action with no cuts or scene changes.",
    "Continuous action with no cuts or scene changes.",
    "Continuous action with no cuts or scene changes."
  ],
  "resolution": {
    "W": 640,
    "H": 400
  },
  "aspect_ratio": "16,9",
  "duration": "6.667",
  "fps": 30.0,
  "robotics": {
    "embodiment": "Articulated robotic arm with a black gripper.",
    "end_effector": "Black, multi-fingered gripper.",
    "manipulated_objects": "Red and white packaged item.",
    "task_goal": "Grasp and release a package on the shelf.",
    "initial_state": "Arm extended towards the shelf.",
    "terminal_state": "Arm retracted slightly.",
    "affordances": "The gripper is capable of grasping and releasing objects.",
    "failure_modes": "None observed in this segment."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The gripper successfully grasps the package, lifts it slightly, and then releases it.",
    "support_and_stability": "The arm is supported by its base, and the package is stable on the shelf.",
    "force_gravity_friction": "The gripper applies enough force to lift the package against gravity, then releases it.",
    "causal_chain": "The arm extends, grasps the package, lifts it, releases it, and retracts.",
    "object_permanence": "The package remains on the shelf throughout the sequence.",
    "next_action_prediction": "The arm may continue to interact with other packages on the shelf."
  },
  "uncertainties": [],
  "free_form_caption": "A top-down view shows a robotic arm with a black gripper extending from the bottom right to grasp a red and white packaged item from a supermarket shelf. The arm lifts the package slightly, then releases it back onto the shelf before retracting slightly. The scene is brightly lit, with various colorful packages visible on the wooden shelf and a speckled gray tile floor below."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms113
cosmos_terms131
intersection41
jaccard0.202

agibot_web_004

top

Webpage Caption

This video is captured from the robot's right-hand camera view, denoted as hand_right, providing an embodied perspective of a manipulation task on a green table. At the start, the workspace contains a yellow ball, a white spoon, and a grey square frame. The robot's left hand enters the frame from the top-left, descending to grasp the yellow ball. As contact occurs, the left hand pinches the ball and lifts it vertically, then slides it across the table to place it atop a white cloth. Subsequently, the left hand moves to the white spoon, grasping it by the handle and relocating it near the ball. Simultaneously, the right hand, which is the camera mount, moves forward. The right hand's gripper descends and makes contact with the top edge of the grey square frame. The right hand then performs a pinch-and-lift motion, raising the frame from the table surface and tilting it as the camera viewpoint shifts upward and backward. By the end of the clip, the left hand has retreated from the center of the workspace, the yellow ball and white spoon are clustered together on the left side of the table, and the right hand is holding the grey frame aloft, tilted at an angle. The background features a white wall and a blue banner with gold Chinese characters. The motion is characterized by purposeful, sequential movements, with the right hand's movement causing a dynamic shift in the camera's perspective as it lifts the object.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "A robot performs a sequence of manipulation tasks on a green table, involving a yellow ball, a white spoon, and a grey square frame. The robot's left hand is seen picking up and moving the ball and spoon, while the right hand (which carries the camera) interacts with the grey frame.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:01",
        "actions": [
          "left hand descends toward the yellow ball"
        ],
        "interactions": [
          "left hand gripper approaches the ball"
        ],
        "causal": [
          "the left hand is positioning itself to grasp the ball"
        ]
      },
      {
        "time": "00:00:01–00:00:02",
        "actions": [
          "left hand lifts the yellow ball"
        ],
        "interactions": [
          "left hand grasps and lifts the ball from the table"
        ],
        "causal": [
          "the ball is displaced vertically from the table surface"
        ]
      },
      {
        "time": "00:00:02–00:00:03",
        "actions": [
          "left hand moves the ball toward a white cloth/object"
        ],
        "interactions": [
          "left hand releases the ball onto a white cloth"
        ],
        "causal": [
          "the ball settles on top of the white cloth"
        ]
      },
      {
        "time": "00:00:03–00:00:05",
        "actions": [
          "left hand approaches the white spoon",
          "right hand moves toward the grey frame"
        ],
        "interactions": [
          "left hand grasps the spoon",
          "right hand gripper aligns with the grey frame"
        ],
        "causal": [
          "the spoon is lifted; the right hand prepares to grasp the frame"
        ]
      },
      {
        "time": "00:00:05–00:00:07",
        "actions": [
          "left hand moves the spoon",
          "right hand grasps the grey frame"
        ],
        "interactions": [
          "left hand releases the spoon near the ball",
          "right hand fingertips pinch the top edge of the grey frame"
        ],
        "causal": [
          "the spoon is placed on the table; the frame is lifted and tilted"
        ]
      },
      {
        "time": "00:00:07–00:00:10",
        "actions": [
          "right hand lifts and rotates the grey frame",
          "left hand moves away"
        ],
        "interactions": [
          "right hand holds the frame while moving the camera viewpoint"
        ],
        "causal": [
          "the frame is displaced from its original position and tilted upward"
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The yellow ball, white spoon, and grey frame are arranged on a green table surface.",
        "The robot's right hand is in the foreground, providing the POV.",
        "The left hand operates in the mid-ground."
      ],
      "temporal": [
        "The ball is moved first, followed by the spoon, and finally the grey frame."
      ],
      "comparisons": [
        "The left hand performs more frequent, smaller movements, while the right hand performs a single, larger lift of the frame."
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "left hand",
          "desc": "To relocate the ball and spoon to a specific area of the table."
        },
        {
          "id": "right hand",
          "desc": "To pick up and manipulate the grey square frame."
        }
      ],
      "theme": "Object sorting and manipulation",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The right hand will likely place the grey frame in a new location or release it.",
    "outcome": "The objects will be rearranged on the table surface.",
    "past": "The robot was likely initialized in a home position before reaching for the ball.",
    "what_if": "If the right hand had not grasped the frame securely, the frame might have slid across the table instead of being lifted."
  },
  "meta": {
    "style": "wrist-mounted robot camera, hand_right POV, lab capture with some motion blur"
  },
  "final_dense_caption": "This video is captured from the robot's right-hand camera view, denoted as hand_right, providing an embodied perspective of a manipulation task on a green table. At the start, the workspace contains a yellow ball, a white spoon, and a grey square frame. The robot's left hand enters the frame from the top-left, descending to grasp the yellow ball. As contact occurs, the left hand pinches the ball and lifts it vertically, then slides it across the table to place it atop a white cloth. Subsequently, the left hand moves to the white spoon, grasping it by the handle and relocating it near the ball. Simultaneously, the right hand, which is the camera mount, moves forward. The right hand's gripper descends and makes contact with the top edge of the grey square frame. The right hand then performs a pinch-and-lift motion, raising the frame from the table surface and tilting it as the camera viewpoint shifts upward and backward. By the end of the clip, the left hand has retreated from the center of the workspace, the yellow ball and white spoon are clustered together on the left side of the table, and the right hand is holding the grey frame aloft, tilted at an angle. The background features a white wall and a blue banner with gold Chinese characters. The motion is characterized by purposeful, sequential movements, with the right hand's movement causing a dynamic shift in the camera's perspective as it lifts the object.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:02:00+0000

From a first-person perspective mounted on a robotic arm, the viewpoint looks down at a green table surface illuminated by bright, even top-lit indoor lighting that casts soft shadows. The robot's embodiment consists of a white and gray arm terminating in a black gripper detailed with silver screws and a small camera mounted on top. In the background, a white wall displays text alongside a blue banner featuring Chinese characters and images of people. The camera pans and tilts continuously with the arm's motion, maintaining a high-angle close-up with deep focus on the foreground manipulation. Initially, the black gripper grasps a yellow toy resembling a banana with a red and yellow top, lifting it from the center of the table and placing it onto a slightly crumpled white cloth positioned on the left side. Subsequently, the arm moves to grasp a white plastic spoon, lifting it from the center and setting it down on the table surface to the left. Finally, the gripper reaches for a gray rectangular frame with a hollow center located on the right side, lifting it and placing it in the center of the table. Throughout the sequence, the robot applies force to overcome gravity and friction, ensuring objects remain stable on the table after placement. The continuous action concludes with the yellow toy, white spoon, and gray frame resting on the green surface, demonstrating successful manipulation without visible failure modes. The objects maintain their new positions, confirming object permanence as the arm repositions itself between tasks.

Rewrite raw output
{
  "rewritten_free_form_caption": "From a first-person perspective mounted on a robotic arm, the viewpoint looks down at a green table surface illuminated by bright, even top-lit indoor lighting that casts soft shadows. The robot's embodiment consists of a white and gray arm terminating in a black gripper detailed with silver screws and a small camera mounted on top. In the background, a white wall displays text alongside a blue banner featuring Chinese characters and images of people. The camera pans and tilts continuously with the arm's motion, maintaining a high-angle close-up with deep focus on the foreground manipulation. Initially, the black gripper grasps a yellow toy resembling a banana with a red and yellow top, lifting it from the center of the table and placing it onto a slightly crumpled white cloth positioned on the left side. Subsequently, the arm moves to grasp a white plastic spoon, lifting it from the center and setting it down on the table surface to the left. Finally, the gripper reaches for a gray rectangular frame with a hollow center located on the right side, lifting it and placing it in the center of the table. Throughout the sequence, the robot applies force to overcome gravity and friction, ensuring objects remain stable on the table after placement. The continuous action concludes with the yellow toy, white spoon, and gray frame resting on the green surface, demonstrating successful manipulation without visible failure modes. The objects maintain their new positions, confirming object permanence as the arm repositions itself between tasks.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:04:26+0000 · display: free_form_caption

A first-person view from a robotic arm's perspective, showing the arm manipulating various objects on a green table. The arm picks up a yellow toy, places it on a white cloth, then picks up a white plastic spoon and places it on the table, and finally picks up a gray rectangular frame and places it on the table. The background features a white wall and a blue banner with Chinese characters.

Raw free_form_caption
A first-person view from a robotic arm's perspective, showing the arm manipulating various objects on a green table. The arm picks up a yellow toy, places it on a white cloth, then picks up a white plastic spoon and places it on the table, and finally picks up a gray rectangular frame and places it on the table. The background features a white wall and a blue banner with Chinese characters.
Cosmos structured JSON
{
  "scene_imagination": "A first-person view from a robotic arm's perspective, looking down at a green table surface. The robot's black gripper is visible in the foreground, manipulating various objects. On the table, there is a yellow toy resembling a banana, a white cloth, a white plastic spoon, a gray rectangular frame, and a red and yellow toy. In the background, there is a white wall with some text and a blue banner with Chinese characters and images of people. The lighting is bright and even, illuminating the scene clearly.",
  "temporal_caption": "The robot arm moves its gripper to pick up a yellow toy, places it on a white cloth, then moves to pick up a white plastic spoon, places it on the table, and finally moves to pick up a gray rectangular frame.",
  "audio_description": "The video does not contain any audio.",
  "subjects": [
    {
      "description": "A robotic arm with a black gripper, viewed from a first-person perspective.",
      "appearance_details": "The gripper is black with silver screws and a small camera mounted on top. The arm is white and gray.",
      "relationship": "Interacting with objects on the table.",
      "location": "Foreground, moving across the table.",
      "relative_size": "Large within frame.",
      "orientation": "Facing the table.",
      "pose": "Moving and manipulating objects.",
      "action": "Picking up and placing objects.",
      "state_changes": "Moves from left to right, picking up a yellow toy, then a white spoon, and finally a gray frame."
    },
    {
      "description": "A yellow toy resembling a banana.",
      "appearance_details": "Yellow, curved shape, with a red and yellow top.",
      "relationship": "Being manipulated by the robot arm.",
      "location": "On the table, initially near the center, then moved to the left.",
      "relative_size": "Small.",
      "orientation": "Lying flat on the table.",
      "pose": "Static until moved by the robot.",
      "action": "Being picked up and placed on a white cloth.",
      "state_changes": "Moved from the center of the table to the left side."
    },
    {
      "description": "A white cloth.",
      "appearance_details": "White, rectangular shape, slightly crumpled.",
      "relationship": "Being used as a surface for placing the yellow toy.",
      "location": "On the table, near the center.",
      "relative_size": "Small.",
      "orientation": "Lying flat on the table.",
      "pose": "Static.",
      "action": "Receiving the yellow toy.",
      "state_changes": "No significant change."
    },
    {
      "description": "A white plastic spoon.",
      "appearance_details": "White, curved shape, with a handle.",
      "relationship": "Being manipulated by the robot arm.",
      "location": "On the table, initially near the center, then moved to the left.",
      "relative_size": "Small.",
      "orientation": "Lying flat on the table.",
      "pose": "Static until moved by the robot.",
      "action": "Being picked up and placed on the table.",
      "state_changes": "Moved from the center of the table to the left side."
    },
    {
      "description": "A gray rectangular frame.",
      "appearance_details": "Gray, rectangular shape, with a hollow center.",
      "relationship": "Being manipulated by the robot arm.",
      "location": "On the table, initially near the right side, then moved to the center.",
      "relative_size": "Small.",
      "orientation": "Lying flat on the table.",
      "pose": "Static until moved by the robot.",
      "action": "Being picked up and placed on the table.",
      "state_changes": "Moved from the right side of the table to the center."
    }
  ],
  "background_setting": "An indoor setting with a green table surface. In the background, there is a white wall with some text and a blue banner with Chinese characters and images of people. The lighting is bright and even.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-lit.",
    "shadows": "Soft shadows on the table.",
    "illumination_effect": "Even illumination across the scene."
  },
  "aesthetics": {
    "composition": "First-person perspective, centered on the table and the robot's gripper.",
    "color_scheme": "Green table, black gripper, white and yellow objects, blue banner.",
    "mood_atmosphere": "Neutral, focused on the task.",
    "patterns": "No prominent patterns."
  },
  "cinematography": {
    "camera_motion": "The camera moves with the robot's arm, panning and tilting to follow the action.",
    "framing": "Close-up on the robot's gripper and the objects on the table.",
    "camera_angle": "High angle, looking down at the table.",
    "depth_of_field": "Deep focus, keeping both the foreground and background in focus.",
    "focus": "Sharp focus on the robot's gripper and the objects on the table.",
    "lens_focal_length": "Wide-angle lens."
  },
  "style_medium": "Robotics video.",
  "artistic_style": "Realistic, documentary-style.",
  "context": "A demonstration of a robotic arm manipulating objects on a table.",
  "actions": [
    {
      "time": "0-5s",
      "description": "The robot arm moves its gripper to pick up the yellow toy and places it on the white cloth.",
      "actor": "Robot arm",
      "object": "Yellow toy",
      "contact_event": "Gripper grasps the yellow toy, lifts it, and places it on the white cloth.",
      "state_changes": "Yellow toy is moved from the center to the left side of the table."
    },
    {
      "time": "5-10s",
      "description": "The robot arm moves its gripper to pick up the white plastic spoon and places it on the table.",
      "actor": "Robot arm",
      "object": "White plastic spoon",
      "contact_event": "Gripper grasps the white plastic spoon, lifts it, and places it on the table.",
      "state_changes": "White plastic spoon is moved from the center to the left side of the table."
    },
    {
      "time": "10-17s",
      "description": "The robot arm moves its gripper to pick up the gray rectangular frame and places it on the table.",
      "actor": "Robot arm",
      "object": "Gray rectangular frame",
      "contact_event": "Gripper grasps the gray rectangular frame, lifts it, and places it on the table.",
      "state_changes": "Gray rectangular frame is moved from the right side to the center of the table."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "2025",
      "category": "scene_sign",
      "appearance": "White text on a blue background.",
      "spatial": "Bottom right corner of the blue banner.",
      "spatial_temporal": "Static throughout the video.",
      "context": "Part of the banner in the background."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0-5s",
      "description": "The robot arm picks up the yellow toy and places it on the white cloth.",
      "key_changes": "Yellow toy is moved from the center to the left side of the table.",
      "camera": "High angle, looking down at the table.",
      "robot_motion": "Moving from left to right.",
      "object_motion": "Yellow toy is moved from the center to the left side of the table.",
      "physical_logic": "The robot arm grasps the yellow toy, lifts it, and places it on the white cloth."
    },
    {
      "segment_index": 1,
      "time_range": "5-10s",
      "description": "The robot arm picks up the white plastic spoon and places it on the table.",
      "key_changes": "White plastic spoon is moved from the center to the left side of the table.",
      "camera": "High angle, looking down at the table.",
      "robot_motion": "Moving from left to right.",
      "object_motion": "White plastic spoon is moved from the center to the left side of the table.",
      "physical_logic": "The robot arm grasps the white plastic spoon, lifts it, and places it on the table."
    },
    {
      "segment_index": 2,
      "time_range": "10-17s",
      "description": "The robot arm picks up the gray rectangular frame and places it on the table.",
      "key_changes": "Gray rectangular frame is moved from the right side to the center of the table.",
      "camera": "High angle, looking down at the table.",
      "robot_motion": "Moving from right to left.",
      "object_motion": "Gray rectangular frame is moved from the right side to the center of the table.",
      "physical_logic": "The robot arm grasps the gray rectangular frame, lifts it, and places it on the table."
    }
  ],
  "transitions": [
    "Continuous action with no cuts or transitions."
  ],
  "resolution": {
    "W": 640,
    "H": 528
  },
  "aspect_ratio": "16,9",
  "duration": "17.667s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "A robotic arm with a black gripper, viewed from a first-person perspective.",
    "end_effector": "Black gripper with silver screws and a small camera mounted on top.",
    "manipulated_objects": [
      "Yellow toy",
      "White plastic spoon",
      "Gray rectangular frame"
    ],
    "task_goal": "Manipulating objects on a table.",
    "initial_state": "The robot arm is positioned over the table, ready to manipulate objects.",
    "terminal_state": "The robot arm has placed the yellow toy, white plastic spoon, and gray rectangular frame on the table.",
    "affordances": "The robot arm is capable of grasping, lifting, and placing objects.",
    "failure_modes": "No visible failure modes."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robot arm uses its gripper to grasp and manipulate objects.",
    "support_and_stability": "The objects are placed on the table, providing support and stability.",
    "force_gravity_friction": "The robot arm applies force to lift and place objects, overcoming gravity and friction.",
    "causal_chain": "The robot arm grasps an object, lifts it, and places it on the table.",
    "object_permanence": "The objects remain on the table after being placed.",
    "next_action_prediction": "The robot arm will likely continue to manipulate other objects on the table."
  },
  "uncertainties": [],
  "free_form_caption": "A first-person view from a robotic arm's perspective, showing the arm manipulating various objects on a green table. The arm picks up a yellow toy, places it on a white cloth, then picks up a white plastic spoon and places it on the table, and finally picks up a gray rectangular frame and places it on the table. The background features a white wall and a blue banner with Chinese characters."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms103
cosmos_terms127
intersection34
jaccard0.173

agibot_web_005

top

Webpage Caption

Captured from the robot's left-hand camera view (hand_left), the sequence begins with a view of a blue plastic bucket on a grey floor, containing various small items including a piece of white bubble wrap. The robot's right hand enters the frame from above, descending to pinch and lift the bubble wrap from the bucket. As the right hand lifts the object, the robot's body shifts, causing the camera to pan and tilt toward a white table. The right hand then releases the bubble wrap onto the table surface. Following this, the right hand moves toward a small, rigid brown rectangular object resting on the table. The gripper makes contact with the brown object, pinching it securely and lifting it upward. The right hand then slides the object across the white table surface toward the left side of the frame. Throughout the clip, the robot's left hand is not actively manipulating objects but serves as the mounting point for the camera, visible as the black structural components in the lower foreground. The environment includes a brown sofa in the background and a person's legs and feet visible to the left. By the end of the clip, the right hand is holding the brown object above the table, while the bubble wrap remains stationary on the table surface and the blue bucket remains on the floor, now partially empty.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "The robot uses its right hand to pick up a piece of bubble wrap from a blue bucket and then moves it to a white table, where it picks up a small brown object and moves it across the surface.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:02",
        "actions": [
          "right hand descends toward a blue bucket"
        ],
        "interactions": [
          "right hand contacts bubble wrap"
        ],
        "causal": [
          "the right hand grips the bubble wrap, lifting it from the bucket"
        ]
      },
      {
        "time": "00:00:02–00:00:04",
        "actions": [
          "right hand moves bubble wrap toward a white table",
          "robot body rotates/shifts viewpoint"
        ],
        "interactions": [
          "bubble wrap is released onto the white table"
        ],
        "causal": [
          "the bubble wrap settles on the table surface"
        ]
      },
      {
        "time": "00:00:04–00:00:07",
        "actions": [
          "right hand approaches a small brown rectangular object on the table"
        ],
        "interactions": [
          "right hand grips the brown object"
        ],
        "causal": [
          "the brown object is lifted from the table"
        ]
      },
      {
        "time": "00:00:07–00:00:11",
        "actions": [
          "right hand moves the brown object across the table surface"
        ],
        "interactions": [
          "right hand maintains grip on the brown object"
        ],
        "causal": [
          "the object is displaced to a new position on the table"
        ]
      }
    ],
    "relations": {
      "spatial": [
        "blue bucket is on the floor in the foreground/center",
        "white table is to the right and then becomes the central focus",
        "brown object is on the white table",
        "robot's left hand is visible as the camera mount/base in the lower foreground"
      ],
      "temporal": [
        "bubble wrap is moved first, followed by the brown object"
      ],
      "comparisons": [
        "the bubble wrap is larger and more deformable than the rigid brown object"
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "right hand",
          "desc": "to transfer items from a container to a table and rearrange items on the table"
        }
      ],
      "theme": "object sorting and relocation",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The robot may place the brown object in a specific location or pick up another item from the table.",
    "outcome": "The brown object will be successfully relocated.",
    "past": "The robot was likely positioned in front of the blue bucket preparing to clear it.",
    "what_if": "If the grip on the bubble wrap had failed, the robot would have had to re-attempt the grasp from the bucket."
  },
  "meta": {
    "style": "wrist-mounted robot camera with wide-angle distortion, captured in a lab or office environment"
  },
  "final_dense_caption": "Captured from the robot's left-hand camera view (hand_left), the sequence begins with a view of a blue plastic bucket on a grey floor, containing various small items including a piece of white bubble wrap. The robot's right hand enters the frame from above, descending to pinch and lift the bubble wrap from the bucket. As the right hand lifts the object, the robot's body shifts, causing the camera to pan and tilt toward a white table. The right hand then releases the bubble wrap onto the table surface. Following this, the right hand moves toward a small, rigid brown rectangular object resting on the table. The gripper makes contact with the brown object, pinching it securely and lifting it upward. The right hand then slides the object across the white table surface toward the left side of the frame. Throughout the clip, the robot's left hand is not actively manipulating objects but serves as the mounting point for the camera, visible as the black structural components in the lower foreground. The environment includes a brown sofa in the background and a person's legs and feet visible to the left. By the end of the clip, the right hand is holding the brown object above the table, while the bubble wrap remains stationary on the table surface and the blue bucket remains on the floor, now partially empty.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:02:12+0000

6. **Final Polish (checking word count more carefully):** From a first-person perspective likely mounted on a robot or head, the scene unfolds indoors under artificial top lighting within a room featuring a tiled floor and a background couch. A robotic arm, characterized by a black body with metallic joints and a gripper mechanism, dominates the foreground. Initially, the arm holds a transparent plastic water bottle with a label and cap vertically. The arm manipulates the bottle, releasing it into a cylindrical blue and white trash bin situated on the floor, where gravity causes the bottle to fall and settle inside, increasing the bin's contents. Following this release, the camera executes a panning motion upward and to the right, shifting the focus from the floor to a white table positioned towards the right side of the frame. On this table rests a rectangular wooden box with a wooden texture and a small object inside. As the view stabilizes on the table, the robotic arm transitions from its previous position and moves towards the wooden box. The arm maintains stability throughout the sequence, driven by its control system to approach the stationary box. The sequence concludes with the plastic bottle secured within the trash bin and the robotic arm in motion, extending towards the wooden box on the table, suggesting an impending interaction. The environment remains consistent with deep focus keeping both foreground manipulations and background elements sharp throughout the twenty-second duration.

Rewrite raw output
{
  "rewritten_free_form_caption": "6. **Final Polish (checking word count more carefully):** From a first-person perspective likely mounted on a robot or head, the scene unfolds indoors under artificial top lighting within a room featuring a tiled floor and a background couch. A robotic arm, characterized by a black body with metallic joints and a gripper mechanism, dominates the foreground. Initially, the arm holds a transparent plastic water bottle with a label and cap vertically. The arm manipulates the bottle, releasing it into a cylindrical blue and white trash bin situated on the floor, where gravity causes the bottle to fall and settle inside, increasing the bin's contents. Following this release, the camera executes a panning motion upward and to the right, shifting the focus from the floor to a white table positioned towards the right side of the frame. On this table rests a rectangular wooden box with a wooden texture and a small object inside. As the view stabilizes on the table, the robotic arm transitions from its previous position and moves towards the wooden box. The arm maintains stability throughout the sequence, driven by its control system to approach the stationary box. The sequence concludes with the plastic bottle secured within the trash bin and the robotic arm in motion, extending towards the wooden box on the table, suggesting an impending interaction. The environment remains consistent with deep focus keeping both foreground manipulations and background elements sharp throughout the twenty-second duration.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:04:45+0000 · display: free_form_caption

The video shows a robotic arm performing a sequence of actions in an indoor setting. Initially, the arm is seen holding a plastic water bottle and dropping it into a trash bin. The camera then pans to reveal a white table with a wooden box and a small object on it. The robotic arm moves towards the wooden box, indicating a continuation of its task.

Raw free_form_caption
The video shows a robotic arm performing a sequence of actions in an indoor setting. Initially, the arm is seen holding a plastic water bottle and dropping it into a trash bin. The camera then pans to reveal a white table with a wooden box and a small object on it. The robotic arm moves towards the wooden box, indicating a continuation of its task.
Cosmos structured JSON
{
  "scene_imagination": "The video is captured from a first-person perspective, likely mounted on a robot or a head-mounted camera, showing a sequence of actions in an indoor setting. Initially, the view focuses on a blue and white trash bin on the floor, with a person's legs visible in the background. A robotic arm, equipped with a gripper, is seen manipulating a plastic water bottle, dropping it into the trash bin. The camera then pans upward and to the right, revealing a white table with a wooden box and a small object on it. The robotic arm continues its movement, reaching towards the wooden box on the table.",
  "temporal_caption": "The video begins with a close-up view of a trash bin on the floor. A robotic arm is seen holding a plastic water bottle and drops it into the bin. The camera then pans upward and to the right, showing a white table with a wooden box and a small object on it. The robotic arm moves towards the wooden box on the table.",
  "audio_description": "The video does not contain any discernible audio elements.",
  "subjects": [
    {
      "description": "A robotic arm with a gripper, primarily black with metallic components.",
      "appearance_details": "The robotic arm has a black body with metallic joints and a gripper mechanism at the end.",
      "relationship": "Interacting with objects in the environment.",
      "location": "Foreground, moving from left to right.",
      "relative_size": "Large within the frame.",
      "orientation": "Facing forward, moving towards the right.",
      "pose": "Extended, reaching towards objects.",
      "action": "The robotic arm holds a plastic water bottle and drops it into a trash bin. It then moves towards a wooden box on a table.",
      "state_changes": "The robotic arm changes position from holding the bottle to moving towards the table."
    },
    {
      "description": "A blue and white trash bin on the floor.",
      "appearance_details": "The trash bin is cylindrical with a blue body and a white lid. It contains some trash.",
      "relationship": "Receiving the plastic water bottle from the robotic arm.",
      "location": "Center of the frame, on the floor.",
      "relative_size": "Medium within the frame.",
      "orientation": "Upright.",
      "pose": "Stationary.",
      "action": "The trash bin receives the plastic water bottle.",
      "state_changes": "The trash bin's contents increase as the plastic water bottle is dropped into it."
    },
    {
      "description": "A plastic water bottle.",
      "appearance_details": "The bottle is transparent with a label and a cap.",
      "relationship": "Being manipulated by the robotic arm and dropped into the trash bin.",
      "location": "Initially held by the robotic arm, then dropped into the trash bin.",
      "relative_size": "Small within the frame.",
      "orientation": "Held vertically by the robotic arm.",
      "pose": "Held by the robotic arm, then falling into the trash bin.",
      "action": "The bottle is dropped into the trash bin.",
      "state_changes": "The bottle changes position from being held by the robotic arm to being inside the trash bin."
    },
    {
      "description": "A wooden box on a white table.",
      "appearance_details": "The box is rectangular with a wooden texture and a small object inside.",
      "relationship": "Being approached by the robotic arm.",
      "location": "On the white table, towards the right side of the frame.",
      "relative_size": "Medium within the frame.",
      "orientation": "Upright on the table.",
      "pose": "Stationary.",
      "action": "The box remains stationary as the robotic arm approaches it.",
      "state_changes": "No significant change."
    }
  ],
  "background_setting": "The video is set indoors, likely in a room with a tiled floor. There is a couch in the background and a white table with a wooden box and a small object on it. The lighting is artificial, and the environment appears to be a living room or a similar space.",
  "lighting": {
    "conditions": "Artificial indoor lighting.",
    "direction": "Top-lit, casting shadows downwards.",
    "shadows": "Soft shadows on the floor and table.",
    "illumination_effect": "Even illumination across the scene, with some reflections on the plastic bottle and metallic components of the robotic arm."
  },
  "aesthetics": {
    "composition": "The camera is positioned at a low angle, focusing on the robotic arm and the objects it interacts with. The framing shifts as the camera pans to follow the robotic arm's movement.",
    "color_scheme": "A mix of neutral tones (black, white, gray) with pops of color from the blue trash bin and the wooden box.",
    "mood_atmosphere": "Functional and focused, emphasizing the robotic arm's actions.",
    "patterns": "The tiled floor has a geometric pattern."
  },
  "cinematography": {
    "camera_motion": "The camera starts with a static shot of the trash bin and then pans upward and to the right, following the robotic arm's movement.",
    "framing": "Close-up on the robotic arm and the objects it interacts with.",
    "camera_angle": "Low angle, looking slightly upwards.",
    "depth_of_field": "Deep focus, keeping both the foreground and background in focus.",
    "focus": "Sharp focus on the robotic arm and the objects it interacts with.",
    "lens_focal_length": "Wide-angle lens, capturing a broad view of the scene."
  },
  "style_medium": "Real-world video footage.",
  "artistic_style": "Documentary-style, capturing the robotic arm's actions in a realistic setting.",
  "context": "The video demonstrates a robotic arm performing a task, specifically dropping a plastic water bottle into a trash bin and then moving towards a wooden box on a table. This could be part of a demonstration or a test of the robotic arm's capabilities.",
  "actions": [
    {
      "time": "0s-5s",
      "description": "The robotic arm holds a plastic water bottle and drops it into the trash bin.",
      "actor": "Robotic arm",
      "object": "Plastic water bottle",
      "contact_event": "The robotic arm releases the bottle into the trash bin.",
      "state_change": "The bottle moves from the robotic arm to the trash bin."
    },
    {
      "time": "5s-10s",
      "description": "The camera pans upward and to the right, revealing a white table with a wooden box and a small object on it.",
      "actor": "Camera",
      "object": "White table, wooden box, small object",
      "contact_event": "The camera movement reveals new objects in the scene.",
      "state_change": "The view changes from the trash bin to the table."
    },
    {
      "time": "10s-20s",
      "description": "The robotic arm moves towards the wooden box on the table.",
      "actor": "Robotic arm",
      "object": "Wooden box",
      "contact_event": "The robotic arm approaches the wooden box.",
      "state_change": "The robotic arm changes position, moving towards the table."
    }
  ],
  "text_and_signage_elements": [],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0s-5s",
      "description": "The robotic arm drops a plastic water bottle into a trash bin.",
      "key_changes": "The bottle is released into the trash bin.",
      "camera": "Static shot of the trash bin.",
      "robot_motion": "The robotic arm moves downward to drop the bottle.",
      "object_motion": "The bottle falls into the trash bin.",
      "physical_logic": "Gravity causes the bottle to fall into the trash bin."
    },
    {
      "segment_index": 1,
      "time_range": "5s-10s",
      "description": "The camera pans upward and to the right, revealing a white table with a wooden box and a small object on it.",
      "key_changes": "The view changes from the trash bin to the table.",
      "camera": "Panning motion, moving upward and to the right.",
      "robot_motion": "The robotic arm remains stationary.",
      "object_motion": "No significant object motion.",
      "physical_logic": "The camera movement reveals new objects in the scene."
    },
    {
      "segment_index": 2,
      "time_range": "10s-20s",
      "description": "The robotic arm moves towards the wooden box on the table.",
      "key_changes": "The robotic arm changes position, moving towards the table.",
      "camera": "Static shot of the table.",
      "robot_motion": "The robotic arm moves towards the wooden box.",
      "object_motion": "No significant object motion.",
      "physical_logic": "The robotic arm's movement is driven by its programming or control system."
    }
  ],
  "transitions": [
    {
      "transition": "The camera pans upward and to the right, shifting the focus from the trash bin to the table.",
      "time": "5s-10s",
      "type": "Camera pan"
    },
    {
      "transition": "The robotic arm moves towards the wooden box on the table.",
      "time": "10s-20s",
      "type": "Object motion"
    }
  ],
  "resolution": {
    "W": 640,
    "H": 528
  },
  "aspect_ratio": "16,9",
  "duration": "20.000s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "A robotic arm with a gripper, primarily black with metallic components.",
    "end_effector": "A gripper mechanism at the end of the robotic arm.",
    "manipulated_objects": [
      "Plastic water bottle",
      "Wooden box"
    ],
    "task_goal": "The robotic arm is performing a task of dropping a plastic water bottle into a trash bin and then moving towards a wooden box on a table.",
    "initial_state": "The robotic arm is holding a plastic water bottle.",
    "terminal_state": "The robotic arm has dropped the bottle into the trash bin and is moving towards the wooden box on the table.",
    "affordances": "The robotic arm is capable of grasping and releasing objects, and it can move in multiple directions.",
    "failure_modes": "The robotic arm may fail to grasp or release objects properly, or it may move incorrectly."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robotic arm successfully grasps and releases the plastic water bottle.",
    "support_and_stability": "The robotic arm maintains stability while performing its tasks.",
    "force_gravity_friction": "Gravity causes the bottle to fall into the trash bin.",
    "causal_chain": "The robotic arm holds the bottle, releases it, and then moves towards the wooden box.",
    "object_permanence": "The objects in the scene remain present throughout the video.",
    "next_action_prediction": "The robotic arm will likely interact with the wooden box or the small object on the table."
  },
  "uncertainties": [],
  "free_form_caption": "The video shows a robotic arm performing a sequence of actions in an indoor setting. Initially, the arm is seen holding a plastic water bottle and dropping it into a trash bin. The camera then pans to reveal a white table with a wooden box and a small object on it. The robotic arm moves towards the wooden box, indicating a continuation of its task."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms97
cosmos_terms124
intersection32
jaccard0.169

agibot_web_006

top

Webpage Caption

Captured from the robot's right-hand camera view (hand_right), the video shows a manipulation task on a white tabletop. At the start, the right hand is seen gripping a small, brown rectangular object. The left hand is not visible throughout the clip. The right hand then opens its gripper, releasing the brown object onto the white surface, where it remains stationary. Following this, the robot's arm moves the right hand toward a taller object: a product in clear plastic packaging with a bright orange top. As the camera moves with the arm, the right hand approaches the base of the orange package. The gripper then closes around the lower portion of the packaging, establishing a firm pinch grip. Once contact is secure, the right hand lifts the orange package vertically and shifts it slightly to the right, moving it away from the brown object. By the end of the clip, the right hand is holding the orange package aloft, while the brown object remains on the table to the left. The background includes a dark surface and a small white bowl containing green items, which remain undisturbed. The viewpoint is dynamic, shifting in response to the right arm's movement, and exhibits typical wide-angle distortion from a wrist-mounted lens.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "A robot uses its right hand to pick up a packaged item from a white surface and move it to a different location, while another small brown rectangular object remains on the table.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:02",
        "actions": [
          "The right hand gripper closes around a small brown rectangular object."
        ],
        "interactions": [
          "Right hand makes contact with the brown object, pinching it."
        ],
        "causal": [
          "The gripper's closure secures the brown object."
        ]
      },
      {
        "time": "00:00:02–00:00:04",
        "actions": [
          "The right hand opens the gripper, releasing the brown object."
        ],
        "interactions": [
          "Right hand releases the brown object onto the white surface."
        ],
        "causal": [
          "The object remains stationary on the table after release."
        ]
      },
      {
        "time": "00:00:04–00:00:07",
        "actions": [
          "The right hand moves toward a tall, orange-topped packaged item."
        ],
        "interactions": [
          "The right hand approaches the base of the orange package."
        ],
        "causal": [
          "The camera view shifts as the arm moves toward the target."
        ]
      },
      {
        "time": "00:00:07–00:00:11",
        "actions": [
          "The right hand grips the orange package and lifts it upward and slightly to the right."
        ],
        "interactions": [
          "Right hand fingertips pinch the lower part of the plastic packaging."
        ],
        "causal": [
          "The orange package is lifted off the white surface, moving out of the center frame."
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The brown object is to the left of the orange package.",
        "The white surface serves as the primary workspace.",
        "The right hand is the central actor in the foreground."
      ],
      "temporal": [
        "The robot first interacts with the brown object before moving to the orange package."
      ],
      "comparisons": [
        "The orange package is significantly taller than the brown rectangular object."
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "right hand",
          "desc": "To manipulate and relocate specific objects on the table."
        }
      ],
      "theme": "object sorting or relocation",
      "social": null
    }
  },
  "prediction": {
    "next": "The robot will likely place the orange package in a designated bin or area.",
    "outcome": "The orange package will be successfully relocated.",
    "past": "The robot likely approached the table and identified the objects to be moved.",
    "what_if": "If the gripper had failed to secure the orange package, it might have tipped over due to its height."
  },
  "meta": {
    "style": "wrist-mounted robot camera with slight fisheye distortion"
  },
  "final_dense_caption": "Captured from the robot's right-hand camera view (hand_right), the video shows a manipulation task on a white tabletop. At the start, the right hand is seen gripping a small, brown rectangular object. The left hand is not visible throughout the clip. The right hand then opens its gripper, releasing the brown object onto the white surface, where it remains stationary. Following this, the robot's arm moves the right hand toward a taller object: a product in clear plastic packaging with a bright orange top. As the camera moves with the arm, the right hand approaches the base of the orange package. The gripper then closes around the lower portion of the packaging, establishing a firm pinch grip. Once contact is secure, the right hand lifts the orange package vertically and shifts it slightly to the right, moving it away from the brown object. By the end of the clip, the right hand is holding the orange package aloft, while the brown object remains on the table to the left. The background includes a dark surface and a small white bowl containing green items, which remain undisturbed. The viewpoint is dynamic, shifting in response to the right arm's movement, and exhibits typical wide-angle distortion from a wrist-mounted lens.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:02:25+0000

Captured from a first-person perspective of a black robot arm equipped with a two-pronged black gripper, the scene unfolds within an indoor workspace featuring a white table surface under bright top-lit conditions that cast soft shadows. The camera maintains a top-down angle with shallow depth of field, closely tracking the arm's movements against a background of a dark floor and a white machine with a circular component. On the table, a brown tape dispenser sits stationary on the left, while a clear plastic bag with an orange and yellow label and a white plastic fork rest on the right. The robot arm initially moves toward the tape dispenser but does not make contact, shifting focus instead to the plastic bag. The gripper closes to grasp the bag, lifting it slightly before placing it back down, demonstrating force application against gravity and friction. Subsequently, the arm targets the white plastic fork, picking it up and holding it briefly before returning it to the table surface. Throughout these interactions, the objects remain present, adhering to object permanence despite temporary displacement. The sequence concludes with the robot arm returning to a resting position after completing the tasks. Uncertainty remains regarding the exact sequence of actions depending on the robot's programming and object positions, as movements may be affected by external factors such as obstacles or malfunctions. The mechanical structure supports the arm, allowing dynamic poses as it interacts with the various items on the table.

Rewrite raw output
{
  "rewritten_free_form_caption": "Captured from a first-person perspective of a black robot arm equipped with a two-pronged black gripper, the scene unfolds within an indoor workspace featuring a white table surface under bright top-lit conditions that cast soft shadows. The camera maintains a top-down angle with shallow depth of field, closely tracking the arm's movements against a background of a dark floor and a white machine with a circular component. On the table, a brown tape dispenser sits stationary on the left, while a clear plastic bag with an orange and yellow label and a white plastic fork rest on the right. The robot arm initially moves toward the tape dispenser but does not make contact, shifting focus instead to the plastic bag. The gripper closes to grasp the bag, lifting it slightly before placing it back down, demonstrating force application against gravity and friction. Subsequently, the arm targets the white plastic fork, picking it up and holding it briefly before returning it to the table surface. Throughout these interactions, the objects remain present, adhering to object permanence despite temporary displacement. The sequence concludes with the robot arm returning to a resting position after completing the tasks. Uncertainty remains regarding the exact sequence of actions depending on the robot's programming and object positions, as movements may be affected by external factors such as obstacles or malfunctions. The mechanical structure supports the arm, allowing dynamic poses as it interacts with the various items on the table.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:05:08+0000 · display: free_form_caption

The video shows a robot arm performing a series of tasks in a controlled environment. The robot arm moves towards a tape dispenser, then a plastic bag, and finally a white plastic fork. It picks up the plastic bag and the fork, lifting them slightly before placing them back down. The camera follows the robot arm's movements closely, providing a clear view of the objects on the table. The lighting is bright, illuminating the workspace clearly. The robot arm is black with a gripper attached to it, and the objects on the table include a brown tape dispenser, a clear plastic bag with an orange and yellow label, and a white plastic fork. The background includes a dark floor and some equipment, such as a white machine with a circular component.

Raw free_form_caption
The video shows a robot arm performing a series of tasks in a controlled environment. The robot arm moves towards a tape dispenser, then a plastic bag, and finally a white plastic fork. It picks up the plastic bag and the fork, lifting them slightly before placing them back down. The camera follows the robot arm's movements closely, providing a clear view of the objects on the table. The lighting is bright, illuminating the workspace clearly. The robot arm is black with a gripper attached to it, and the objects on the table include a brown tape dispenser, a clear plastic bag with an orange and yellow label, and a white plastic fork. The background includes a dark floor and some equipment, such as a white machine with a circular component.
Cosmos structured JSON
{
  "scene_imagination": "The video is captured from a first-person perspective of a robot arm equipped with a black gripper. The setting is an indoor workspace with a white table surface. On the table, there is a brown tape dispenser, a clear plastic bag with an orange and yellow label, and a white plastic fork. The robot arm moves around the table, interacting with these objects. The background includes a dark floor and some equipment, such as a white machine with a circular component. The lighting is bright, illuminating the workspace clearly.",
  "temporal_caption": "The robot arm starts by moving towards the tape dispenser, then shifts its focus to the plastic bag. It picks up the plastic bag, lifts it slightly, and then places it back on the table. The arm then moves towards the white plastic fork, picks it up, and holds it briefly before placing it back down. Throughout the video, the camera follows the movements of the robot arm closely, providing a clear view of the objects on the table.",
  "audio_description": "The video does not contain any audio elements. The focus is solely on the visual actions of the robot arm and the objects on the table.",
  "subjects": [
    {
      "description": "A robot arm with a black gripper attached to it. The gripper has two prongs that can open and close to pick up objects.",
      "appearance_details": "The robot arm is black with metallic components. The gripper has a rectangular shape with two prongs that can move independently.",
      "relationship": "The robot arm is the main subject of the video, interacting with various objects on the table.",
      "location": "Center of the frame, moving around the table.",
      "relative_size": "Large within the frame, taking up a significant portion of the view.",
      "orientation": "Facing the table, with the gripper pointing downwards.",
      "pose": "The robot arm is in a dynamic pose, moving its gripper to interact with objects.",
      "action": "The robot arm moves towards the tape dispenser, then the plastic bag, and finally the white plastic fork. It picks up the plastic bag and the fork, lifting them slightly before placing them back down.",
      "state_changes": "The robot arm changes its position and orientation as it moves around the table and interacts with different objects."
    },
    {
      "description": "A brown tape dispenser with a roll of tape inside.",
      "appearance_details": "The tape dispenser is rectangular with a brown body and a clear plastic cover over the tape roll.",
      "relationship": "The tape dispenser is one of the objects on the table that the robot arm interacts with.",
      "location": "On the left side of the table.",
      "relative_size": "Small within the frame, taking up a small portion of the view.",
      "orientation": "Facing upwards, with the tape roll visible.",
      "pose": "The tape dispenser is stationary on the table.",
      "action": "The tape dispenser remains stationary throughout the video.",
      "state_changes": "No significant state changes."
    },
    {
      "description": "A clear plastic bag with an orange and yellow label.",
      "appearance_details": "The plastic bag is transparent with a colorful label featuring a cartoon character. It contains some items inside.",
      "relationship": "The plastic bag is one of the objects on the table that the robot arm interacts with.",
      "location": "On the right side of the table.",
      "relative_size": "Medium within the frame, taking up a moderate portion of the view.",
      "orientation": "Lying flat on the table.",
      "pose": "The plastic bag is stationary on the table.",
      "action": "The robot arm picks up the plastic bag, lifts it slightly, and then places it back down.",
      "state_changes": "The plastic bag is lifted and then placed back down by the robot arm."
    },
    {
      "description": "A white plastic fork.",
      "appearance_details": "The fork is white with four prongs and a handle.",
      "relationship": "The fork is one of the objects on the table that the robot arm interacts with.",
      "location": "On the table, near the plastic bag.",
      "relative_size": "Small within the frame, taking up a small portion of the view.",
      "orientation": "Lying flat on the table.",
      "pose": "The fork is stationary on the table.",
      "action": "The robot arm picks up the fork, holds it briefly, and then places it back down.",
      "state_changes": "The fork is lifted and then placed back down by the robot arm."
    }
  ],
  "background_setting": "The video is set in an indoor workspace with a white table surface. The background includes a dark floor and some equipment, such as a white machine with a circular component. The lighting is bright, illuminating the workspace clearly.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-lit, casting shadows downwards.",
    "shadows": "Soft shadows are cast by the objects on the table.",
    "illumination_effect": "The lighting provides clear visibility of the objects and the robot arm's movements."
  },
  "aesthetics": {
    "composition": "The camera is positioned above the table, providing a top-down view of the workspace. The robot arm and the objects are centered in the frame.",
    "color_scheme": "The color scheme is predominantly white and black, with some colorful elements from the plastic bag and the tape dispenser.",
    "mood_atmosphere": "The mood is focused and task-oriented, with the robot arm performing precise movements.",
    "patterns": "The video does not feature any prominent patterns."
  },
  "cinematography": {
    "camera_motion": "The camera follows the movements of the robot arm closely, providing a clear view of the objects on the table.",
    "framing": "The camera is framed to capture the robot arm and the objects on the table, with a top-down perspective.",
    "camera_angle": "Top-down angle, providing a clear view of the workspace.",
    "depth_of_field": "The depth of field is shallow, with the robot arm and the objects in focus.",
    "focus": "The focus is on the robot arm and the objects it interacts with.",
    "lens_focal_length": "The lens focal length is not specified, but it provides a clear and detailed view of the workspace."
  },
  "style_medium": "The video is a real-world robot manipulation footage, captured from a first-person perspective.",
  "artistic_style": "The video has a technical and precise style, focusing on the robot's movements and interactions with objects.",
  "context": "The video is part of a dataset for physical-AI video generation, specifically from the `agibot_web_caption_viewer` tool. The context is a robot arm performing tasks in a controlled environment.",
  "actions": [
    {
      "time": "0-5 seconds",
      "description": "The robot arm moves towards the tape dispenser on the left side of the table.",
      "actor": "Robot arm",
      "object": "Tape dispenser",
      "contact_event": "The robot arm approaches the tape dispenser but does not make contact.",
      "state_changes": "The robot arm changes its position to move towards the tape dispenser."
    },
    {
      "time": "5-10 seconds",
      "description": "The robot arm shifts its focus to the plastic bag on the right side of the table.",
      "actor": "Robot arm",
      "object": "Plastic bag",
      "contact_event": "The robot arm picks up the plastic bag, lifting it slightly.",
      "state_changes": "The plastic bag is lifted and then placed back down by the robot arm."
    },
    {
      "time": "10-15 seconds",
      "description": "The robot arm moves towards the white plastic fork on the table.",
      "actor": "Robot arm",
      "object": "White plastic fork",
      "contact_event": "The robot arm picks up the fork, holds it briefly, and then places it back down.",
      "state_changes": "The fork is lifted and then placed back down by the robot arm."
    },
    {
      "time": "15-20 seconds",
      "description": "The robot arm completes its interactions with the objects and returns to a resting position.",
      "actor": "Robot arm",
      "object": "None",
      "contact_event": "The robot arm does not make contact with any objects.",
      "state_changes": "The robot arm changes its position to return to a resting state."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "BabyCare",
      "category": "physical_in_scene",
      "appearance": "The text is printed on the plastic bag label, featuring a colorful design with a cartoon character.",
      "spatial": "On the plastic bag, located on the right side of the table.",
      "temporal": "Visible throughout the video.",
      "context": "The text is part of the label on the plastic bag, indicating the brand or product name."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0-5 seconds",
      "description": "The robot arm moves towards the tape dispenser on the left side of the table.",
      "key_changes": "The robot arm changes its position to move towards the tape dispenser.",
      "camera": "The camera follows the robot arm's movement.",
      "robot_motion": "The robot arm moves towards the tape dispenser.",
      "object_motion": "The tape dispenser remains stationary.",
      "physical_logic": "The robot arm is moving to interact with the tape dispenser."
    },
    {
      "segment_index": 1,
      "time_range": "5-10 seconds",
      "description": "The robot arm shifts its focus to the plastic bag on the right side of the table.",
      "key_changes": "The robot arm picks up the plastic bag, lifting it slightly.",
      "camera": "The camera follows the robot arm's movement.",
      "robot_motion": "The robot arm moves towards the plastic bag.",
      "object_motion": "The plastic bag is lifted and then placed back down.",
      "physical_logic": "The robot arm is interacting with the plastic bag."
    },
    {
      "segment_index": 2,
      "time_range": "10-15 seconds",
      "description": "The robot arm moves towards the white plastic fork on the table.",
      "key_changes": "The robot arm picks up the fork, holds it briefly, and then places it back down.",
      "camera": "The camera follows the robot arm's movement.",
      "robot_motion": "The robot arm moves towards the fork.",
      "object_motion": "The fork is lifted and then placed back down.",
      "physical_logic": "The robot arm is interacting with the fork."
    },
    {
      "segment_index": 3,
      "time_range": "15-20 seconds",
      "description": "The robot arm completes its interactions with the objects and returns to a resting position.",
      "key_changes": "The robot arm changes its position to return to a resting state.",
      "camera": "The camera remains stationary.",
      "robot_motion": "The robot arm returns to a resting position.",
      "object_motion": "The objects remain stationary.",
      "physical_logic": "The robot arm has completed its tasks."
    }
  ],
  "transitions": [
    "The robot arm transitions from interacting with the tape dispenser to the plastic bag.",
    "The robot arm transitions from interacting with the plastic bag to the white plastic fork.",
    "The robot arm transitions from interacting with the fork to returning to a resting position."
  ],
  "resolution": {
    "W": 640,
    "H": 528
  },
  "aspect_ratio": "16,9",
  "duration": "20.00s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "The robot arm is a black mechanical device with a gripper attached to it.",
    "end_effector": "The end effector is a black gripper with two prongs that can open and close.",
    "manipulated_objects": [
      "Tape dispenser",
      "Plastic bag",
      "White plastic fork"
    ],
    "task_goal": "The robot arm is performing a series of tasks, including picking up and placing down objects on the table.",
    "initial_state": "The robot arm is positioned above the table, ready to interact with objects.",
    "terminal_state": "The robot arm has completed its tasks and returned to a resting position.",
    "affordances": "The robot arm has the ability to move its gripper to pick up and place down objects.",
    "failure_modes": "The robot arm may fail to pick up or place down objects if it encounters obstacles or malfunctions."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robot arm uses its gripper to make contact with objects and grasp them.",
    "support_and_stability": "The robot arm is supported by its mechanical structure, allowing it to move and interact with objects.",
    "force_gravity_friction": "The robot arm applies force to pick up objects, and gravity and friction affect the movement and placement of objects.",
    "causal_chain": "The robot arm's movements cause it to interact with objects, leading to changes in their positions.",
    "object_permanence": "The objects on the table remain present throughout the video, even when they are picked up and placed back down.",
    "next_action_prediction": "The robot arm may continue to interact with other objects on the table or return to a resting position."
  },
  "uncertainties": [
    "The exact sequence of actions may vary depending on the robot's programming and the objects' positions.",
    "The robot's movements may be affected by external factors such as obstacles or malfunctions."
  ],
  "free_form_caption": "The video shows a robot arm performing a series of tasks in a controlled environment. The robot arm moves towards a tape dispenser, then a plastic bag, and finally a white plastic fork. It picks up the plastic bag and the fork, lifting them slightly before placing them back down. The camera follows the robot arm's movements closely, providing a clear view of the objects on the table. The lighting is bright, illuminating the workspace clearly. The robot arm is black with a gripper attached to it, and the objects on the table include a brown tape dispenser, a clear plastic bag with an orange and yellow label, and a white plastic fork. The background includes a dark floor and some equipment, such as a white machine with a circular component."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms99
cosmos_terms136
intersection37
jaccard0.187

agibot_web_007

top

Webpage Caption

This video is captured from the robot's right-hand camera view, denoted as hand_right. At the start, the scene is a retail environment with wire shelves stocked with various packaged goods, including blue and yellow bags of wet wipes. The robot's right hand, appearing as a black gripper in the immediate foreground, is positioned directly in front of a blue package of wet wipes. The left hand is visible in the background, positioned slightly to the left and remaining stationary throughout the clip. As the sequence begins, the right hand closes its gripper around the blue package, applying a pinch-like grip to the plastic material. Then, the right hand lifts the package upward and pulls it backward away from the shelf. This motion causes the blue package to slide out from among its neighbors, creating a visible gap on the shelf. As the right hand retracts, the camera POV shifts, moving away from the shelf and causing the surrounding environment to blur slightly due to the arm's movement. By the end of the clip, the right hand is holding the blue package securely, having successfully removed it from the shelf, while the left hand remains in its inactive position in the background. The other items on the shelf, including the yellow packages below and the green packages above, remain stable and undisturbed. The final state shows the right hand holding the object in the foreground, with the retail aisle visible behind it.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "An AgiBot robot uses its right hand to grasp and remove a blue package of wet wipes from a retail shelf, while its left hand remains visible but inactive in the background.",
    "genre": "robot retail-object handling"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:02",
        "actions": [
          "The right hand approaches a blue package of wet wipes on a wire shelf."
        ],
        "interactions": [
          "The right gripper makes contact with the blue package."
        ],
        "causal": [
          "The gripper's closure begins to secure the package."
        ]
      },
      {
        "time": "00:00:02–00:00:05",
        "actions": [
          "The right hand lifts the blue package upward and slightly backward."
        ],
        "interactions": [
          "The package is pulled away from the other blue packages on the shelf."
        ],
        "causal": [
          "The package is displaced from its original position on the shelf."
        ]
      },
      {
        "time": "00:00:05–00:00:08",
        "actions": [
          "The right hand continues to pull the package back, moving the camera POV away from the shelf."
        ],
        "interactions": [
          "The package remains firmly held in the right gripper."
        ],
        "causal": [
          "The viewpoint shifts as the robot arm retracts."
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The blue package is located on a wire shelf among other similar blue and yellow packages.",
        "The right hand is in the immediate foreground, while the left hand is visible in the mid-ground to the left.",
        "The shelf is positioned in front of the robot."
      ],
      "temporal": [
        "The grasp occurs first, followed by the lifting and retraction of the object."
      ],
      "comparisons": [
        "The right hand is actively manipulating the object, whereas the left hand remains stationary."
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "right hand",
          "desc": "To pick up a specific blue package of wet wipes from the store shelf."
        }
      ],
      "theme": "Object retrieval in a retail environment",
      "social": "None"
    }
  },
  "prediction": {
    "next": "The robot will likely move the package toward a shopping basket or a designated drop-off point.",
    "outcome": "The blue package will be successfully removed from the shelf.",
    "past": "The robot navigated to the shelf and identified the target blue package.",
    "what_if": "If the gripper had slipped, the package would have fallen back onto the shelf or the floor."
  },
  "meta": {
    "style": "wrist-mounted robot camera with slight fisheye distortion and motion blur"
  },
  "final_dense_caption": "This video is captured from the robot's right-hand camera view, denoted as hand_right. At the start, the scene is a retail environment with wire shelves stocked with various packaged goods, including blue and yellow bags of wet wipes. The robot's right hand, appearing as a black gripper in the immediate foreground, is positioned directly in front of a blue package of wet wipes. The left hand is visible in the background, positioned slightly to the left and remaining stationary throughout the clip. As the sequence begins, the right hand closes its gripper around the blue package, applying a pinch-like grip to the plastic material. Then, the right hand lifts the package upward and pulls it backward away from the shelf. This motion causes the blue package to slide out from among its neighbors, creating a visible gap on the shelf. As the right hand retracts, the camera POV shifts, moving away from the shelf and causing the surrounding environment to blur slightly due to the arm's movement. By the end of the clip, the right hand is holding the blue package securely, having successfully removed it from the shelf, while the left hand remains in its inactive position in the background. The other items on the shelf, including the yellow packages below and the green packages above, remain stable and undisturbed. The final state shows the right hand holding the object in the foreground, with the retail aisle visible behind it.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:02:37+0000

6. **Final Polish (checking word count more precisely):** From a first-person perspective mounted on a robot's right hand, the view looks down a brightly lit supermarket aisle featuring light-colored tile flooring and fluorescent ceiling lights. In the immediate foreground, a black gripper with silver bolts firmly holds a blue and white packaged item, maintaining a stable grasp throughout the sequence. To the left, a white robot arm with black joints extends forward towards shelves stocked with various products, including green and yellow boxes alongside blue packaged goods. The camera, utilizing a wide-angle lens with deep focus, captures the scene with sharp clarity on both the robotic components and the background shelving. Initially, the robot arm extends towards the shelves while the gripper maintains its hold on the package. Subsequently, the viewpoint pans slightly to the right, revealing more of the aisle and the arm's position, before the arm retracts slightly and the camera pans back to the left. The physical logic indicates the arm is supported by the robot's body and subject to gravity and friction during these movements. Although the arm reaches toward the shelves, the exact task remains unclear, as it is uncertain whether the robot is picking up or placing items. Similarly, the specific reason for the camera panning is not defined. The sequence concludes with the arm in a slightly retracted state while the gripper continues to secure the blue and white item, leaving the final operational goal ambiguous within the context of the supermarket environment.

Rewrite raw output
{
  "rewritten_free_form_caption": "6. **Final Polish (checking word count more precisely):** From a first-person perspective mounted on a robot's right hand, the view looks down a brightly lit supermarket aisle featuring light-colored tile flooring and fluorescent ceiling lights. In the immediate foreground, a black gripper with silver bolts firmly holds a blue and white packaged item, maintaining a stable grasp throughout the sequence. To the left, a white robot arm with black joints extends forward towards shelves stocked with various products, including green and yellow boxes alongside blue packaged goods. The camera, utilizing a wide-angle lens with deep focus, captures the scene with sharp clarity on both the robotic components and the background shelving. Initially, the robot arm extends towards the shelves while the gripper maintains its hold on the package. Subsequently, the viewpoint pans slightly to the right, revealing more of the aisle and the arm's position, before the arm retracts slightly and the camera pans back to the left. The physical logic indicates the arm is supported by the robot's body and subject to gravity and friction during these movements. Although the arm reaches toward the shelves, the exact task remains unclear, as it is uncertain whether the robot is picking up or placing items. Similarly, the specific reason for the camera panning is not defined. The sequence concludes with the arm in a slightly retracted state while the gripper continues to secure the blue and white item, leaving the final operational goal ambiguous within the context of the supermarket environment.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:05:24+0000 · display: free_form_caption

A first-person perspective from a robot's gripper looking down an aisle in a brightly lit supermarket. The robot's black gripper is holding a blue and white packaged item, and the white robot arm is extending towards the shelves. The camera pans slightly to the right and then back to the left as the robot's arm retracts slightly.

Raw free_form_caption
A first-person perspective from a robot's gripper looking down an aisle in a brightly lit supermarket. The robot's black gripper is holding a blue and white packaged item, and the white robot arm is extending towards the shelves. The camera pans slightly to the right and then back to the left as the robot's arm retracts slightly.
Cosmos structured JSON
{
  "scene_imagination": "A first-person perspective from a robot's gripper looking down an aisle in a brightly lit supermarket. The robot's black gripper is in the foreground, holding a blue and white packaged item. To the left, a white robot arm with black joints is visible, extending towards the shelves. The shelves are stocked with various products, including green and yellow boxes and blue packaged goods. The floor is a light-colored tile, and the ceiling has fluorescent lights. The camera moves slightly as the robot adjusts its position.",
  "temporal_caption": "The robot's gripper holds a blue and white packaged item while the robot arm extends towards the shelves. The camera pans slightly to the right, revealing more of the shelves and the robot arm. The robot arm then retracts slightly, and the camera pans back to the left.",
  "audio_description": "The sound of the robot's motors and the ambient noise of the supermarket, including the hum of fluorescent lights and the occasional sound of other shoppers.",
  "subjects": [
    {
      "description": "A robot's black gripper holding a blue and white packaged item.",
      "appearance_details": "The gripper is black with silver bolts and is holding a blue and white packaged item.",
      "relationship": "The gripper is attached to the robot's arm and is holding the packaged item.",
      "location": "Foreground, bottom center.",
      "relative_size": "Large within frame.",
      "orientation": "Facing forward.",
      "pose": "Holding the packaged item.",
      "action": "Holding the packaged item.",
      "state_changes": "No significant change."
    },
    {
      "description": "A white robot arm with black joints.",
      "appearance_details": "The robot arm is white with black joints and has a black gripper at the end.",
      "relationship": "The robot arm is attached to the robot's body and is holding the packaged item.",
      "location": "Left side of the frame.",
      "relative_size": "Medium within frame.",
      "orientation": "Facing forward.",
      "pose": "Extended towards the shelves.",
      "action": "Extending towards the shelves.",
      "state_changes": "The robot arm extends and then retracts slightly."
    },
    {
      "description": "A blue and white packaged item.",
      "appearance_details": "The packaged item is blue and white with some text and graphics on it.",
      "relationship": "The packaged item is held by the robot's gripper.",
      "location": "Foreground, bottom center.",
      "relative_size": "Medium within frame.",
      "orientation": "Facing forward.",
      "pose": "Held by the robot's gripper.",
      "action": "Being held by the robot's gripper.",
      "state_changes": "No significant change."
    }
  ],
  "background_setting": "A brightly lit supermarket aisle with shelves stocked with various products, including green and yellow boxes and blue packaged goods. The floor is a light-colored tile, and the ceiling has fluorescent lights.",
  "lighting": {
    "conditions": "Bright indoor lighting.",
    "direction": "Top-lit from fluorescent lights.",
    "shadows": "Soft shadows on the floor and shelves.",
    "illumination_effect": "Even illumination across the scene."
  },
  "aesthetics": {
    "composition": "First-person perspective from the robot's gripper, with the robot arm on the left and the shelves on the right.",
    "color_scheme": "Bright and colorful, with dominant colors of blue, green, yellow, and white.",
    "mood_atmosphere": "Functional and industrial.",
    "patterns": "Repeating patterns of products on the shelves."
  },
  "cinematography": {
    "camera_motion": "The camera pans slightly to the right and then back to the left.",
    "framing": "Close-up on the robot's gripper and arm, with the shelves in the background.",
    "camera_angle": "First-person perspective.",
    "depth_of_field": "Deep focus, with both the foreground and background in focus.",
    "focus": "Sharp focus on the robot's gripper and arm, with the shelves in the background also in focus.",
    "lens_focal_length": "Wide-angle lens."
  },
  "style_medium": "Robotics video.",
  "artistic_style": "Realistic and functional.",
  "context": "A robot is performing a task in a supermarket, likely picking up or placing items on the shelves.",
  "actions": [
    {
      "time": "0-5 seconds",
      "description": "The robot's gripper holds a blue and white packaged item while the robot arm extends towards the shelves.",
      "actor": "Robot",
      "object": "Robot's gripper and arm",
      "contact_event": "The robot's arm extends towards the shelves.",
      "state_changes": "The robot's arm extends."
    },
    {
      "time": "5-10 seconds",
      "description": "The camera pans slightly to the right, revealing more of the shelves and the robot arm.",
      "actor": "Camera",
      "object": "Shelves and robot arm",
      "contact_event": "The camera pans to the right.",
      "state_changes": "The camera pans to the right."
    },
    {
      "time": "10-15 seconds",
      "description": "The robot arm retracts slightly, and the camera pans back to the left.",
      "actor": "Robot",
      "object": "Robot's arm",
      "contact_event": "The robot's arm retracts.",
      "state_changes": "The robot's arm retracts."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "10.90",
      "category": "scene_sign",
      "appearance": "White text on a blue background.",
      "spatial": "Right side of the frame, on the shelves.",
      "spatial_temporal": "Static on the shelves.",
      "context": "Price tag for the blue packaged goods."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0-5 seconds",
      "description": "The robot's gripper holds a blue and white packaged item while the robot arm extends towards the shelves.",
      "key_changes": "The robot's arm extends.",
      "camera": "Static.",
      "robot_motion": "The robot's arm extends.",
      "object_motion": "No significant object motion.",
      "physical_logic": "The robot's arm extends to reach the shelves."
    },
    {
      "segment_index": 1,
      "time_range": "5-10 seconds",
      "description": "The camera pans slightly to the right, revealing more of the shelves and the robot arm.",
      "key_changes": "The camera pans to the right.",
      "camera": "Panning to the right.",
      "robot_motion": "No significant robot motion.",
      "object_motion": "No significant object motion.",
      "physical_logic": "The camera pans to the right."
    },
    {
      "segment_index": 2,
      "time_range": "10-15 seconds",
      "description": "The robot arm retracts slightly, and the camera pans back to the left.",
      "key_changes": "The robot's arm retracts.",
      "camera": "Panning to the left.",
      "robot_motion": "The robot's arm retracts.",
      "object_motion": "No significant object motion.",
      "physical_logic": "The robot's arm retracts."
    }
  ],
  "transitions": [
    "The camera pans slightly to the right and then back to the left.",
    "The robot's arm extends and then retracts slightly."
  ],
  "resolution": {
    "W": 640,
    "H": 528
  },
  "aspect_ratio": "16,9",
  "duration": "20.00s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "A robot with a white arm and a black gripper.",
    "end_effector": "A black gripper holding a blue and white packaged item.",
    "manipulated_objects": "A blue and white packaged item.",
    "task_goal": "Picking up or placing items on the shelves.",
    "initial_state": "The robot's gripper is holding a blue and white packaged item.",
    "terminal_state": "The robot's arm retracts slightly.",
    "affordances": "The robot's arm can extend and retract.",
    "failure_modes": "The robot's arm may not reach the shelves or may drop the packaged item."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robot's gripper is holding a blue and white packaged item.",
    "support_and_stability": "The robot's arm is supported by its body.",
    "force_gravity_friction": "The robot's arm is subject to gravity and friction.",
    "causal_chain": "The robot's arm extends to reach the shelves, the camera pans to the right, the robot's arm retracts slightly, and the camera pans back to the left.",
    "object_permanence": "The blue and white packaged item is held by the robot's gripper throughout the video.",
    "next_action_prediction": "The robot's arm may extend again to reach the shelves."
  },
  "uncertainties": [
    "The exact task the robot is performing is unclear.",
    "The reason for the camera panning is unclear."
  ],
  "free_form_caption": "A first-person perspective from a robot's gripper looking down an aisle in a brightly lit supermarket. The robot's black gripper is holding a blue and white packaged item, and the white robot arm is extending towards the shelves. The camera pans slightly to the right and then back to the left as the robot's arm retracts slightly."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms108
cosmos_terms136
intersection41
jaccard0.202

agibot_web_008

top

Webpage Caption

Captured from the robot's top_head camera view, the scene opens with a robot performing a bimanual manipulation task on a white table covered with a red-and-white patterned tablecloth. In the background, a red movie poster with blurred faces is visible. At the start, the robot's left hand is holding a green and white packaged object in an upright, tilted position, while the right hand is positioned to the right, preparing to assist. As the sequence progresses, the right hand moves inward to grasp the package, and together, the two hands coordinate to rotate and push the package down until it lies flat on the tablecloth. The movement is somewhat hesitant, with the package sliding slightly across the fabric. Once the package is stable on the surface, the right hand releases its grip and shifts its trajectory toward a small, orange, cylindrical object located a few centimeters to the right. Meanwhile, the left hand continues to nudge the green package into a final position before eventually releasing it. By the end of the clip, the left hand has retracted slightly, and the right hand is in the process of making contact with the orange object, which is displaced slightly by the gripper's approach. The final state shows the green package lying flat in the center of the workspace, the orange object slightly shifted to the right, and both robot hands positioned near these objects, having completed the transition from bimanual package handling to single-hand object targeting.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "A robot uses both hands to manipulate a green and white packaged object on a red-and-white patterned tablecloth, attempting to reposition or open it while interacting with a small orange object nearby.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:01",
        "actions": [
          "The robot's left hand holds a green packaged object upright.",
          "The robot's right hand approaches the object from the right."
        ],
        "interactions": [
          "Left hand maintains a grip on the green package."
        ],
        "causal": [
          "The grip of the left hand keeps the package elevated and stable."
        ]
      },
      {
        "time": "00:00:01–00:00:03",
        "actions": [
          "The robot's right hand makes contact with the package.",
          "The left hand begins to tilt the package downward."
        ],
        "interactions": [
          "Bimanual contact with the green package."
        ],
        "causal": [
          "The coordinated movement of both hands causes the package to rotate and descend toward the table."
        ]
      },
      {
        "time": "00:00:03–00:00:06",
        "actions": [
          "The robot's hands shift their grip on the package.",
          "The package is pushed flat against the tablecloth."
        ],
        "interactions": [
          "The package slides across the red-and-white surface."
        ],
        "causal": [
          "Downward and lateral pressure from the grippers forces the package to lie flat."
        ]
      },
      {
        "time": "00:00:06–00:00:11",
        "actions": [
          "The robot's right hand releases the package and moves toward a small orange object.",
          "The left hand maintains contact with the package, adjusting its position."
        ],
        "interactions": [
          "Right hand moves away from the package; left hand continues to manipulate the package."
        ],
        "causal": [
          "The release of the right hand allows it to target a different object in the workspace."
        ]
      },
      {
        "time": "00:00:11–00:00:14",
        "actions": [
          "The robot's right hand attempts to grasp or push the small orange object.",
          "The left hand releases the green package."
        ],
        "interactions": [
          "Right hand contacts the orange object; left hand separates from the package."
        ],
        "causal": [
          "The right hand's movement displaces the orange object slightly."
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The green package is centrally located on the red-and-white tablecloth.",
        "The orange object is to the right of the package.",
        "The robot's hands enter from the left and right sides of the frame.",
        "A movie poster is visible in the background."
      ],
      "temporal": [
        "Bimanual manipulation of the package precedes the independent movement of the right hand toward the orange object."
      ],
      "comparisons": [
        "The green package is significantly larger than the small orange object."
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "Robot hands",
          "desc": "To flatten the green package on the table and then interact with the orange object."
        }
      ],
      "theme": "Object manipulation and sorting",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The robot will likely pick up the orange object with its right hand.",
    "outcome": "The orange object will be lifted from the table.",
    "past": "The robot likely picked up the green package from a standing or stacked position.",
    "what_if": "If the robot had not used both hands, the package might have tipped over unpredictably instead of being laid flat."
  },
  "meta": {
    "style": "top-head robot view, lab capture, stable embodied camera"
  },
  "final_dense_caption": "Captured from the robot's top_head camera view, the scene opens with a robot performing a bimanual manipulation task on a white table covered with a red-and-white patterned tablecloth. In the background, a red movie poster with blurred faces is visible. At the start, the robot's left hand is holding a green and white packaged object in an upright, tilted position, while the right hand is positioned to the right, preparing to assist. As the sequence progresses, the right hand moves inward to grasp the package, and together, the two hands coordinate to rotate and push the package down until it lies flat on the tablecloth. The movement is somewhat hesitant, with the package sliding slightly across the fabric. Once the package is stable on the surface, the right hand releases its grip and shifts its trajectory toward a small, orange, cylindrical object located a few centimeters to the right. Meanwhile, the left hand continues to nudge the green package into a final position before eventually releasing it. By the end of the clip, the left hand has retracted slightly, and the right hand is in the process of making contact with the orange object, which is displaced slightly by the gripper's approach. The final state shows the green package lying flat in the center of the workspace, the orange object slightly shifted to the right, and both robot hands positioned near these objects, having completed the transition from bimanual package handling to single-hand object targeting.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:02:49+0000

Revision:* From a stationary, slightly elevated top-head viewpoint, the scene captures a brightly lit indoor environment featuring a table covered with a red and white patterned tablecloth. Two robotic arms with sleek black and white segments, visible joints, and cables dominate the foreground, positioned on either side of the table. Their black, durable-looking grippers firmly grasp a flexible green and white object, which appears to be a piece of fabric or clothing, though its exact nature remains uncertain. Behind the table, a large poster with images of people and text hangs on a light-colored wall alongside a smaller poster to the right, though the specific text is not clearly legible. The lighting is bright and even from above, casting minimal shadows and highlighting the details of the manipulation. Initially, the arms hold the object while adjusting their grip to prepare for movement. Subsequently, the arms work together to pull and adjust the object, unfolding it across the table surface. The object changes shape and position gradually as the arms apply appropriate force to overcome gravity and friction, maintaining a firm grip to ensure precise manipulation. Throughout the continuous shot, the depth of field remains deep, keeping both foreground and background in sharp focus. The robotic arms continue to manipulate the object, making final adjustments to ensure it is properly positioned and spread out on the table. By the end of the segment, the object is fully unfolded and spread out, with the arms maintaining their grip to complete the positioning task. The physical interaction demonstrates stable support and precise manipulation without abrupt changes or interruptions, reflecting the continuous nature of the task as the object transitions from a held state to a spread configuration on the patterned surface.

Rewrite raw output
{
  "rewritten_free_form_caption": "Revision:* From a stationary, slightly elevated top-head viewpoint, the scene captures a brightly lit indoor environment featuring a table covered with a red and white patterned tablecloth. Two robotic arms with sleek black and white segments, visible joints, and cables dominate the foreground, positioned on either side of the table. Their black, durable-looking grippers firmly grasp a flexible green and white object, which appears to be a piece of fabric or clothing, though its exact nature remains uncertain. Behind the table, a large poster with images of people and text hangs on a light-colored wall alongside a smaller poster to the right, though the specific text is not clearly legible. The lighting is bright and even from above, casting minimal shadows and highlighting the details of the manipulation. Initially, the arms hold the object while adjusting their grip to prepare for movement. Subsequently, the arms work together to pull and adjust the object, unfolding it across the table surface. The object changes shape and position gradually as the arms apply appropriate force to overcome gravity and friction, maintaining a firm grip to ensure precise manipulation. Throughout the continuous shot, the depth of field remains deep, keeping both foreground and background in sharp focus. The robotic arms continue to manipulate the object, making final adjustments to ensure it is properly positioned and spread out on the table. By the end of the segment, the object is fully unfolded and spread out, with the arms maintaining their grip to complete the positioning task. The physical interaction demonstrates stable support and precise manipulation without abrupt changes or interruptions, reflecting the continuous nature of the task as the object transitions from a held state to a spread configuration on the patterned surface.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:05:44+0000 · display: free_form_caption

The video shows two robotic arms manipulating a green and white object on a table covered with a red and white patterned tablecloth. The arms work together to pull, adjust, and unfold the object, spreading it out on the table. The background features a large poster with images of people and text, as well as another smaller poster on the right side. The lighting is bright and even, illuminating the scene clearly. The video captures the precise movements of the robotic arms, highlighting their capabilities in handling and adjusting objects.

Raw free_form_caption
The video shows two robotic arms manipulating a green and white object on a table covered with a red and white patterned tablecloth. The arms work together to pull, adjust, and unfold the object, spreading it out on the table. The background features a large poster with images of people and text, as well as another smaller poster on the right side. The lighting is bright and even, illuminating the scene clearly. The video captures the precise movements of the robotic arms, highlighting their capabilities in handling and adjusting objects.
Cosmos structured JSON
{
  "scene_imagination": "Two robotic arms are positioned on either side of a table covered with a red and white patterned tablecloth. The arms are manipulating a green and white object, which appears to be a folded piece of fabric or clothing. The background features a large poster with images of people and text, as well as another smaller poster on the right side. The lighting is bright and even, illuminating the scene clearly.",
  "temporal_caption": "The video begins with the two robotic arms holding the green and white object. The arms then start to manipulate the object, pulling and adjusting it. The object is moved around on the table, with the arms working together to unfold and spread it out. The arms continue to adjust the object, ensuring it is properly positioned on the table.",
  "audio_description": "The video does not contain any audio.",
  "subjects": [
    {
      "description": "A pair of robotic arms with black and white segments, equipped with grippers at the ends.",
      "appearance_details": "The arms have a sleek, modern design with visible joints and cables. The grippers are black and appear to be made of a durable material.",
      "relationship": "The arms are working together to manipulate the green and white object.",
      "location": "Center of the frame, positioned on either side of the table.",
      "relative_size": "Large within the frame, dominating the foreground.",
      "orientation": "Facing the table, with the grippers pointing towards the center.",
      "pose": "The arms are extended and actively moving, with the grippers gripping the object.",
      "action": "The arms are manipulating the green and white object, pulling and adjusting it on the table.",
      "state_changes": "The arms move continuously, adjusting their grip and position as they manipulate the object."
    },
    {
      "description": "A green and white object, possibly a piece of fabric or clothing.",
      "appearance_details": "The object has a green base with white patterns or text on it. It appears to be flexible and can be folded or unfolded.",
      "relationship": "The object is being manipulated by the robotic arms.",
      "location": "Center of the frame, on the table.",
      "relative_size": "Medium within the frame, occupying a significant portion of the table.",
      "orientation": "The object is being moved and adjusted by the arms, changing its orientation and shape.",
      "pose": "The object is being pulled, folded, and unfolded by the arms.",
      "action": "The object is being manipulated by the arms, changing its shape and position on the table.",
      "state_changes": "The object changes shape and position as it is manipulated by the arms."
    }
  ],
  "background_setting": "The scene takes place indoors, with a table covered in a red and white patterned tablecloth. Behind the table, there is a large poster featuring images of people and text. To the right, there is another smaller poster with text and images. The background is relatively plain, with a light-colored wall.",
  "lighting": {
    "conditions": "Bright and even lighting, illuminating the scene clearly.",
    "direction": "The lighting appears to be coming from above, casting minimal shadows.",
    "shadows": "Minimal shadows are visible, indicating a well-lit environment.",
    "illumination_effect": "The lighting highlights the details of the robotic arms and the object being manipulated."
  },
  "aesthetics": {
    "composition": "The composition is centered, with the robotic arms and the object being the main focus. The background elements provide context but do not distract from the main action.",
    "color_scheme": "The color scheme is vibrant, with the red and white tablecloth contrasting with the black and white robotic arms and the green and white object.",
    "mood_atmosphere": "The mood is focused and precise, reflecting the careful manipulation of the object by the robotic arms.",
    "patterns": "The tablecloth features a repeating pattern of red and white stripes with small decorative elements."
  },
  "cinematography": {
    "camera_motion": "The camera remains stationary throughout the video, providing a consistent view of the action.",
    "framing": "The framing is wide, capturing the entire scene including the robotic arms, the object, and the background elements.",
    "camera_angle": "The camera angle is slightly elevated, providing a clear view of the table and the objects on it.",
    "depth_of_field": "The depth of field is deep, with both the foreground and background elements in focus.",
    "focus": "The focus is sharp, capturing the details of the robotic arms and the object being manipulated.",
    "lens_focal_length": "The lens appears to be a standard focal length, providing a natural perspective of the scene."
  },
  "style_medium": "The video is a real-world recording, capturing the robotic arms manipulating the object in a physical environment.",
  "artistic_style": "The style is realistic and documentary-like, focusing on the precise movements of the robotic arms.",
  "context": "The video appears to be a demonstration of robotic manipulation, showcasing the capabilities of the robotic arms in handling and adjusting objects.",
  "actions": [
    {
      "time": "0:00-0:05",
      "description": "The robotic arms are holding the green and white object, preparing to manipulate it.",
      "actor": "Robotic arms",
      "object": "Green and white object",
      "contact_event": "The grippers are gripping the object.",
      "state_changes": "The arms begin to move, adjusting their grip on the object."
    },
    {
      "time": "0:05-0:15",
      "description": "The robotic arms start to pull and adjust the object, unfolding it on the table.",
      "actor": "Robotic arms",
      "object": "Green and white object",
      "contact_event": "The grippers continue to grip the object as it is moved.",
      "state_changes": "The object changes shape and position as it is manipulated by the arms."
    },
    {
      "time": "0:15-0:20",
      "description": "The robotic arms continue to adjust the object, ensuring it is properly positioned on the table.",
      "actor": "Robotic arms",
      "object": "Green and white object",
      "contact_event": "The grippers maintain their grip on the object.",
      "state_changes": "The object is further adjusted and spread out on the table."
    }
  ],
  "text_and_signage_elements": [
    {
      "text": "The poster in the background features text and images of people, but the specific text is not clearly legible.",
      "category": "physical_in_scene",
      "appearance": "The text is printed on the poster, with a mix of colors and fonts.",
      "spatial": "The poster is located behind the table, occupying a large portion of the background.",
      "temporal": "The poster remains stationary throughout the video.",
      "context": "The poster provides context to the setting, suggesting a public or commercial space."
    }
  ],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0:00-0:05",
      "description": "The robotic arms are holding the green and white object, preparing to manipulate it.",
      "key_changes": "The arms begin to move, adjusting their grip on the object.",
      "camera": "The camera remains stationary, providing a consistent view of the action.",
      "robot_motion": "The arms start to move, adjusting their grip on the object.",
      "object_motion": "The object is held by the arms, with minor adjustments in position.",
      "physical_logic": "The arms are preparing to manipulate the object, setting the stage for the subsequent actions."
    },
    {
      "segment_index": 1,
      "time_range": "0:05-0:15",
      "description": "The robotic arms start to pull and adjust the object, unfolding it on the table.",
      "key_changes": "The object changes shape and position as it is manipulated by the arms.",
      "camera": "The camera remains stationary, providing a consistent view of the action.",
      "robot_motion": "The arms actively manipulate the object, pulling and adjusting it.",
      "object_motion": "The object is pulled and unfolded by the arms.",
      "physical_logic": "The arms are working together to unfold and spread out the object on the table."
    },
    {
      "segment_index": 2,
      "time_range": "0:15-0:20",
      "description": "The robotic arms continue to adjust the object, ensuring it is properly positioned on the table.",
      "key_changes": "The object is further adjusted and spread out on the table.",
      "camera": "The camera remains stationary, providing a consistent view of the action.",
      "robot_motion": "The arms continue to manipulate the object, making final adjustments.",
      "object_motion": "The object is further adjusted and spread out on the table.",
      "physical_logic": "The arms are ensuring the object is properly positioned, completing the manipulation task."
    }
  ],
  "transitions": [
    "The video is a continuous shot, with no noticeable transitions between segments.",
    "The actions of the robotic arms flow smoothly, with no abrupt changes or interruptions.",
    "The object being manipulated changes shape and position gradually, reflecting the continuous nature of the task."
  ],
  "resolution": {
    "W": 640,
    "H": 400
  },
  "aspect_ratio": "16,9",
  "duration": "0:20",
  "fps": 30,
  "robotics": {
    "embodiment": "The robotic arms have a sleek, modern design with visible joints and cables. The grippers are black and appear to be made of a durable material.",
    "end_effector": "The grippers are black and appear to be made of a durable material, designed to grip and manipulate objects.",
    "manipulated_objects": "The green and white object, possibly a piece of fabric or clothing.",
    "task_goal": "The task is to manipulate the green and white object, unfolding and spreading it out on the table.",
    "initial_state": "The robotic arms are holding the green and white object, preparing to manipulate it.",
    "terminal_state": "The object is fully unfolded and spread out on the table, with the robotic arms making final adjustments.",
    "affordances": "The robotic arms are capable of precise manipulation, adjusting their grip and position to handle the object effectively.",
    "failure_modes": "Potential failure modes include the grippers slipping off the object or the object becoming tangled or misaligned during manipulation."
  },
  "physical_reasoning": {
    "contact_and_grasp": "The grippers maintain a firm grip on the object, allowing for precise manipulation.",
    "support_and_stability": "The robotic arms provide stable support, ensuring the object remains under control during manipulation.",
    "force_gravity_friction": "The arms apply appropriate force to pull and adjust the object, overcoming gravity and friction.",
    "causal_chain": "The arms initiate the manipulation, pulling and adjusting the object. The object responds to the applied forces, changing shape and position on the table.",
    "object_permanence": "The object remains a single, continuous entity throughout the manipulation, with no parts becoming detached or lost.",
    "next_action_prediction": "The robotic arms will likely continue to make minor adjustments to ensure the object is perfectly positioned on the table."
  },
  "uncertainties": [
    "The specific text on the poster in the background is not clearly legible, making it difficult to determine its exact content.",
    "The exact nature of the green and white object is not fully clear, though it appears to be a piece of fabric or clothing."
  ],
  "free_form_caption": "The video shows two robotic arms manipulating a green and white object on a table covered with a red and white patterned tablecloth. The arms work together to pull, adjust, and unfold the object, spreading it out on the table. The background features a large poster with images of people and text, as well as another smaller poster on the right side. The lighting is bright and even, illuminating the scene clearly. The video captures the precise movements of the robotic arms, highlighting their capabilities in handling and adjusting objects."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms116
cosmos_terms151
intersection39
jaccard0.171

agibot_web_009

top

Webpage Caption

Captured from the robot's right-hand camera view (hand_right), the sequence begins with a wide-angle, fisheye-distorted perspective of a modern interior space. The robot's right arm is visible in the upper left, and the camera is mounted on the wrist, looking forward. At the start, the robot is positioned high, looking down at a white table surface where several items are placed, including a small white rectangular object and some plastic packaging. As the clip progresses, the robot moves its right arm forward and downward, causing the viewpoint to shift rapidly from the room's ceiling and walls toward the table. The right hand, a black multi-fingered gripper, enters the center of the frame as it descends toward the target white object. The left hand is not visible throughout the entire sequence. As the right hand reaches the table, it carefully aligns its fingertips with the sides of the small white rectangular object. The robot then executes a pinch grasp, closing its fingers around the object to secure it. By the end of the clip, the right hand has successfully made contact and gripped the object, which is now held firmly between the gripper's fingers, while the surrounding items on the white table remain undisturbed. The final frame shows a close-up of the gripped object from the hand_right perspective, with the gripper's structure framing the target.

Webpage structured JSON caption
{
  "summary": {
    "narrative": "The robot uses its right hand to approach and grasp a small, rectangular white object on a white table surface. The robot's right-hand camera provides a first-person view of the approach and the subsequent pinch grasp of the object.",
    "genre": "robot manipulation demonstration"
  },
  "reasoning": {
    "timeline": [
      {
        "time": "00:00:00–00:00:03",
        "actions": [
          "The robot moves its right arm and camera forward and downward toward a white table."
        ],
        "interactions": [
          "The camera view shifts from a wide room view to a close-up of the table surface."
        ],
        "causal": [
          "Robot body motion causes the viewpoint to change from a high angle to a low, table-level angle."
        ]
      },
      {
        "time": "00:00:03–00:00:06",
        "actions": [
          "The right hand descends toward a small white rectangular object."
        ],
        "interactions": [
          "The right hand's gripper aligns with the target object."
        ],
        "causal": [
          "The downward motion of the arm brings the gripper into the immediate vicinity of the object."
        ]
      },
      {
        "time": "00:00:06–00:00:10",
        "actions": [
          "The right hand closes its fingers to pinch the white rectangular object."
        ],
        "interactions": [
          "The fingertips make contact with the sides of the white object, securing it in a pinch grip."
        ],
        "causal": [
          "The closing of the gripper fingers applies pressure to the object, lifting it slightly or stabilizing it for removal."
        ]
      }
    ],
    "relations": {
      "spatial": [
        "The white object is centered on the white table surface.",
        "The right hand is positioned directly above and then around the object.",
        "The left hand is not visible in the frame."
      ],
      "temporal": [
        "The approach phase precedes the grasping phase."
      ],
      "comparisons": [
        "The camera view transitions from a wide-angle room perspective to a focused, close-up view of the manipulation area."
      ]
    },
    "interpretation": {
      "intent": [
        {
          "id": "right hand",
          "desc": "To accurately locate and grasp a small white rectangular object from a flat surface."
        }
      ],
      "theme": "precision grasping",
      "social": "null"
    }
  },
  "prediction": {
    "next": "The robot will likely lift the white object from the table and move it to another location.",
    "outcome": "The object will be successfully transported while maintained in the right hand's grip.",
    "past": "The robot was likely scanning the environment or receiving a command to pick up the specific object.",
    "what_if": "If the gripper had closed too early, it might have pushed the object across the table instead of grasping it."
  },
  "meta": {
    "style": "wrist-mounted robot camera with fisheye distortion"
  },
  "final_dense_caption": "Captured from the robot's right-hand camera view (hand_right), the sequence begins with a wide-angle, fisheye-distorted perspective of a modern interior space. The robot's right arm is visible in the upper left, and the camera is mounted on the wrist, looking forward. At the start, the robot is positioned high, looking down at a white table surface where several items are placed, including a small white rectangular object and some plastic packaging. As the clip progresses, the robot moves its right arm forward and downward, causing the viewpoint to shift rapidly from the room's ceiling and walls toward the table. The right hand, a black multi-fingered gripper, enters the center of the frame as it descends toward the target white object. The left hand is not visible throughout the entire sequence. As the right hand reaches the table, it carefully aligns its fingertips with the sides of the small white rectangular object. The robot then executes a pinch grasp, closing its fingers around the object to secure it. By the end of the clip, the right hand has successfully made contact and gripped the object, which is now held firmly between the gripper's fingers, while the surrounding items on the white table remain undisturbed. The final frame shows a close-up of the gripped object from the hand_right perspective, with the gripper's structure framing the target.",
  "filtering": {
    "talking_head": false,
    "watermark": false,
    "template_or_text_dominated": false,
    "computer_use": false,
    "multi_panel_or_composite": false,
    "visual_quality_issue": false,
    "edited_or_discontinuous": false
  }
}

JSON-derived LLM Caption

ok
Qwen/Qwen3.5-27B · 2026-06-11T20:03:01+0000

Draft 3 (Polishing):* A first-person perspective from a robotic arm captures a tracking shot moving forward through a modern, dimly lit interior space featuring sleek surfaces and a futuristic aesthetic. The robotic embodiment consists of a black and white gripper mechanism with a black frame, metallic components, and a white cylindrical section attached to a larger structure. The camera maintains an eye-level angle with a wide-angle lens and deep depth of field, keeping the foreground and midground in sharp focus. Initially, the arm moves forward from a counter area into a narrow corridor with patterned walls under top-lit and ambient conditions that cast soft shadows. The robot approaches a white table where a small, rectangular light brown wooden box lies flat, containing a clear plastic bag with small items inside. Upon reaching the table, the gripper extends and makes contact with the wooden box, transitioning from a resting position to actively engaging with the object. The mechanism grasps the box, providing support and stability while applying force to lift and move the box against gravity. The box shifts from a stationary state to being manipulated by the gripper, remaining in the robot's grasp throughout the action. The continuous shot concludes with the robot having moved the wooden box from its initial position on the table, demonstrating physical logic where the gripper supports the object during the lift. Localized bright spots highlight metallic and glossy surfaces within the environment as the manipulation task progresses without cuts or transitions.

Rewrite raw output
{
  "rewritten_free_form_caption": "Draft 3 (Polishing):* A first-person perspective from a robotic arm captures a tracking shot moving forward through a modern, dimly lit interior space featuring sleek surfaces and a futuristic aesthetic. The robotic embodiment consists of a black and white gripper mechanism with a black frame, metallic components, and a white cylindrical section attached to a larger structure. The camera maintains an eye-level angle with a wide-angle lens and deep depth of field, keeping the foreground and midground in sharp focus. Initially, the arm moves forward from a counter area into a narrow corridor with patterned walls under top-lit and ambient conditions that cast soft shadows. The robot approaches a white table where a small, rectangular light brown wooden box lies flat, containing a clear plastic bag with small items inside. Upon reaching the table, the gripper extends and makes contact with the wooden box, transitioning from a resting position to actively engaging with the object. The mechanism grasps the box, providing support and stability while applying force to lift and move the box against gravity. The box shifts from a stationary state to being manipulated by the gripper, remaining in the robot's grasp throughout the action. The continuous shot concludes with the robot having moved the wooden box from its initial position on the table, demonstrating physical logic where the gripper supports the object during the lift. Localized bright spots highlight metallic and glossy surfaces within the environment as the manipulation task progresses without cuts or transitions.",
  "notes": [
    "plain-text response"
  ]
}

Raw Cosmos3-Super Caption

ok
2026-06-10T23:05:59+0000 · display: free_form_caption

A first-person perspective from a robotic arm moving through a modern, dimly lit interior space. The robot navigates past a counter and enters a narrow corridor with patterned walls. It approaches a white table where a small wooden box and a clear plastic bag are placed. The robot's gripper is visible in the foreground, manipulating the wooden box. The environment features sleek surfaces, ambient lighting, and a futuristic aesthetic.

Raw free_form_caption
A first-person perspective from a robotic arm moving through a modern, dimly lit interior space. The robot navigates past a counter and enters a narrow corridor with patterned walls. It approaches a white table where a small wooden box and a clear plastic bag are placed. The robot's gripper is visible in the foreground, manipulating the wooden box. The environment features sleek surfaces, ambient lighting, and a futuristic aesthetic.
Cosmos structured JSON
{
  "scene_imagination": "A first-person perspective from a robotic arm moving through a modern, dimly lit interior space. The robot navigates past a counter and enters a narrow corridor with patterned walls. It approaches a white table where a small wooden box and a clear plastic bag are placed. The robot's gripper is visible in the foreground, manipulating the wooden box. The environment features sleek surfaces, ambient lighting, and a futuristic aesthetic.",
  "temporal_caption": "The robot arm moves forward from a counter area into a corridor, approaches a white table, and then focuses on manipulating a small wooden box placed on the table.",
  "audio_description": "Ambient room tone with subtle mechanical sounds from the robot's movement and gripper operation.",
  "subjects": [
    {
      "description": "A robotic arm with a black and white gripper mechanism, viewed from a first-person perspective.",
      "appearance_details": "The gripper has a black frame with metallic components and a white cylindrical section. It is attached to a larger robotic structure.",
      "relationship": "The primary subject performing actions in the video.",
      "location": "Foreground, moving forward.",
      "relative_size": "Large within frame.",
      "orientation": "Facing forward.",
      "pose": "Extended forward, manipulating objects.",
      "action": "Moving forward, approaching a table, and interacting with a wooden box.",
      "state_changes": "The gripper moves from a resting position to actively engaging with the wooden box."
    },
    {
      "description": "A small, rectangular wooden box with a clear plastic bag inside.",
      "appearance_details": "The box is light brown wood with a clear plastic bag containing small items inside.",
      "relationship": "The object being manipulated by the robot.",
      "location": "On a white table in the foreground.",
      "relative_size": "Small within frame.",
      "orientation": "Lying flat on the table.",
      "pose": "Stationary until interacted with.",
      "action": "Being moved by the robot's gripper.",
      "state_changes": "The box is lifted and moved by the gripper."
    }
  ],
  "background_setting": "A modern, dimly lit interior space with a counter area, a corridor with patterned walls, and a white table. The environment features sleek surfaces, ambient lighting, and a futuristic aesthetic.",
  "lighting": {
    "conditions": "Dim indoor lighting with localized bright spots.",
    "direction": "Top-lit and ambient.",
    "shadows": "Soft shadows cast by objects and the robot.",
    "illumination_effect": "The lighting creates a moody, futuristic atmosphere with highlights on metallic and glossy surfaces."
  },
  "aesthetics": {
    "composition": "First-person perspective with the robot's gripper in the foreground and the environment receding into the background.",
    "color_scheme": "Dark tones with metallic accents and warm highlights.",
    "mood_atmosphere": "Futuristic, focused, and slightly mysterious.",
    "patterns": "Geometric patterns on the walls and floor."
  },
  "cinematography": {
    "camera_motion": "Tracking shot moving forward.",
    "framing": "First-person perspective.",
    "camera_angle": "Eye-level.",
    "depth_of_field": "Deep.",
    "focus": "Sharp focus on the foreground and midground.",
    "lens_focal_length": "Wide-angle."
  },
  "style_medium": "Robotics video",
  "artistic_style": "Futuristic, realistic",
  "context": "A demonstration of a robotic arm performing a manipulation task in a simulated or real-world environment.",
  "actions": [
    {
      "time": "0-5s",
      "description": "The robot arm moves forward from a counter area into a corridor.",
      "actor": "Robot arm",
      "object": "N/A",
      "contact_event": "N/A",
      "state_change": "The robot moves from the counter to the corridor."
    },
    {
      "time": "5-10s",
      "description": "The robot arm approaches a white table.",
      "actor": "Robot arm",
      "object": "White table",
      "contact_event": "N/A",
      "state_change": "The robot reaches the table."
    },
    {
      "time": "10-15s",
      "description": "The robot arm focuses on a small wooden box on the table.",
      "actor": "Robot arm",
      "object": "Wooden box",
      "contact_event": "The gripper makes contact with the box.",
      "state_change": "The robot begins to manipulate the box."
    },
    {
      "time": "15-20s",
      "description": "The robot arm lifts and moves the wooden box.",
      "actor": "Robot arm",
      "object": "Wooden box",
      "contact_event": "The gripper grasps and lifts the box.",
      "state_change": "The box is moved from its initial position."
    }
  ],
  "text_and_signage_elements": [],
  "segments": [
    {
      "segment_index": 0,
      "time_range": "0-5s",
      "description": "The robot arm moves forward from a counter area into a corridor.",
      "key_changes": "The robot moves from the counter to the corridor.",
      "camera": "Tracking shot moving forward.",
      "robot_motion": "Moving forward.",
      "object_motion": "N/A",
      "physical_logic": "The robot is navigating through the environment."
    },
    {
      "segment_index": 1,
      "time_range": "5-10s",
      "description": "The robot arm approaches a white table.",
      "key_changes": "The robot reaches the table.",
      "camera": "Tracking shot moving forward.",
      "robot_motion": "Moving forward.",
      "object_motion": "N/A",
      "physical_logic": "The robot is approaching the target object."
    },
    {
      "segment_index": 2,
      "time_range": "10-20s",
      "description": "The robot arm focuses on and manipulates a small wooden box on the table.",
      "key_changes": "The robot begins to manipulate the box.",
      "camera": "First-person perspective.",
      "robot_motion": "Moving forward and manipulating the box.",
      "object_motion": "The box is lifted and moved.",
      "physical_logic": "The robot is performing a manipulation task."
    }
  ],
  "transitions": [
    "Continuous shot with no cuts or transitions."
  ],
  "resolution": {
    "W": 640,
    "H": 528
  },
  "aspect_ratio": "16,9",
  "duration": "20.00s",
  "fps": 30.0,
  "robotics": {
    "embodiment": "Robotic arm with a black and white gripper mechanism.",
    "end_effector": "Black and white gripper mechanism.",
    "manipulated_objects": [
      "Small wooden box",
      "Clear plastic bag"
    ],
    "task_goal": "Manipulate the wooden box on the table.",
    "initial_state": "The robot is positioned near a counter area.",
    "terminal_state": "The robot has moved the wooden box on the table.",
    "affordances": "The robot is capable of moving forward and manipulating objects.",
    "failure_modes": "N/A"
  },
  "physical_reasoning": {
    "contact_and_grasp": "The robot's gripper makes contact with the wooden box and grasps it.",
    "support_and_stability": "The robot's gripper provides support and stability for the box.",
    "force_gravity_friction": "The robot applies force to lift and move the box against gravity.",
    "causal_chain": "The robot moves forward, approaches the table, and manipulates the box.",
    "object_permanence": "The box remains in the robot's grasp throughout the manipulation.",
    "next_action_prediction": "The robot will continue to manipulate the box or place it in a new location."
  },
  "uncertainties": [],
  "free_form_caption": "A first-person perspective from a robotic arm moving through a modern, dimly lit interior space. The robot navigates past a counter and enters a narrow corridor with patterned walls. It approaches a white table where a small wooden box and a clear plastic bag are placed. The robot's gripper is visible in the foreground, manipulating the wooden box. The environment features sleek surfaces, ambient lighting, and a futuristic aesthetic."
}

Lightweight Text Overlap

This is only a lexical sanity check, not a semantic score.

web_terms104
cosmos_terms144
intersection33
jaccard0.153