Hackathon_Team01_v3 / README.md
ampicard's picture
Update README.md
995bbf6 verified
metadata
license: apache-2.0
task_categories:
  - robotics
tags:
  - LeRobot
configs:
  - config_name: default
    data_files: data/*/*.parquet

This dataset was created using LeRobot.

Dataset Description

This dataset contains the final batch of teleoperated demonstrations collected during a two-day hackathon using the LeRobot library and SO-101 robot arms in a leader–follower configuration. Each episode shows the follower arm picking two colored cubes (one after the other) and placing each into the matching colored cross within a 2×2 grid. Two RGB cameras were used:

Top camera: mounted above the workspace for a clear 2D view of the arm, cubes, and grid.

Front/low camera: slightly above the ground, facing the arm and grid to provide better z-axis cues and arm self-perception.

Despite cardboard backgrounding, the room’s illumination varied over time and is deliberately preserved in the data, as it proved to be a limiting factor and may be valuable for robustness research.

This dataset is intended for vision-based imitation learning (e.g., behavior cloning, goal-conditioned policies), multi-view fusion, and perception-control studies on tabletop manipulation.

Use Cases

  • Imitation Learning: Behavior cloning from teleop demonstrations.

  • Multiview Perception: Fusing top + front perspectives for depth inference without explicit depth sensors.

  • Robustness to Lighting: Evaluating policy sensitivity to illumination drift.

  • State–Action Alignment: Leveraging synchronized proprioception and images.

Data Collection

Teleoperation Setup

  • Leader–Follower: Human teleoperates a leader arm; follower SO-101 replicates motion to generate demonstrations.

  • Workspace: Tabletop with a 2×2 grid. Each cell contains a colored cross; two colored cubes must be placed on matching crosses.

  • Cameras:

    • Top: overhead, full scene.

    • Front: low angle, emphasizes depth and arm self-pose.

  • Background control: Cardboard panels; lighting varies during the day and is preserved in data.

Episode Protocol

1- Move to pre-grasp; localize target cube(s) visually.

2- Grasp first cube; transport; place on correct colored cross.

3- Repeat for second cube.

4- Return to neutral.

Known limitations

  • Lighting drift: Significant variation during the day; expect distribution shift. Consider color constancy or data augmentation.

  • Camera motion: Cameras are fixed for the batch, but small nudges may occur; rely on metadata intrinsics/extrinsics if provided.

  • Occlusions: Self-occlusion of the gripper and cubes in certain positions, especially from left camera during close approach.

  • No depth: RGB only

Additional Information

  • Homepage: deel-ai

  • License: apache-2.0

Dataset Structure

meta/info.json:

{
    "codebase_version": "v3.0",
    "robot_type": "so101_follower",
    "total_episodes": 50,
    "total_frames": 31189,
    "total_tasks": 1,
    "chunks_size": 1000,
    "data_files_size_in_mb": 100,
    "video_files_size_in_mb": 500,
    "fps": 30,
    "splits": {
        "train": "0:50"
    },
    "data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
    "video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
    "features": {
        "action": {
            "dtype": "float32",
            "names": [
                "shoulder_pan.pos",
                "shoulder_lift.pos",
                "elbow_flex.pos",
                "wrist_flex.pos",
                "wrist_roll.pos",
                "gripper.pos"
            ],
            "shape": [
                6
            ]
        },
        "observation.state": {
            "dtype": "float32",
            "names": [
                "shoulder_pan.pos",
                "shoulder_lift.pos",
                "elbow_flex.pos",
                "wrist_flex.pos",
                "wrist_roll.pos",
                "gripper.pos"
            ],
            "shape": [
                6
            ]
        },
        "observation.images.left": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.is_depth_map": false,
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false
            }
        },
        "observation.images.front": {
            "dtype": "video",
            "shape": [
                480,
                640,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 480,
                "video.width": 640,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.is_depth_map": false,
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false
            }
        },
        "timestamp": {
            "dtype": "float32",
            "shape": [
                1
            ],
            "names": null
        },
        "frame_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        },
        "episode_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        },
        "index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        },
        "task_index": {
            "dtype": "int64",
            "shape": [
                1
            ],
            "names": null
        }
    }
}