README.md

metadata

task_categories:
  - robotics
tags:
  - LeRobot
  - robotics
  - robot-learning
  - imitation-learning
  - manipulation
  - pick-and-place
  - so101
  - domain-randomization
  - vision
configs:
  - config_name: default
    data_files: data/*/*.parquet

This dataset was created using LeRobot.

Dataset Summary

• Task: Move a red block into a white box (pick-and-place / block-in-bin).
• Scene setup: Tabletop scene with a fixed target receptacle (white box) and a red block.
• Variations across episodes:
  • Lighting (primary factor):
    • Warm vs. cool illumination (different color temperatures).
    • Different lighting angles/positions (“from different viewpoints”), producing diverse shadow directions and contrast patterns.
  • Red block position (secondary factor): The block start position varies within a small neighborhood around a nominal location (small translations), consistent with the previous dataset in this series.
• Learning goal: Train and evaluate policies that are robust to large appearance changes caused by lighting while still solving a straightforward manipulation task.

Why This Dataset

Lighting changes can significantly alter pixel appearance even when geometry is unchanged. This dataset targets:

• color constancy / illumination invariance challenges,
• robustness to strong shadows and specular highlights,
• stable perception of object boundaries and grasp points under shifting contrast.

It is especially useful as a stress test for vision-based policies that otherwise perform well in stable lab lighting.

Supported Tasks and Use Cases

This dataset is suitable for:

• Imitation learning for pick-and-place with heavy illumination randomization.
• Robust visual policy training (e.g., augmentation studies, representation learning).
• Generalization benchmarks: train on a subset of lighting conditions, test on unseen ones.
• Comparisons against:
  • fixed-lighting datasets,
  • mildly varied-lighting datasets,
  • position-only randomized datasets.

Task Description

Instruction: “Place the red block into the white box.”

A typical episode: approach the block → grasp → transport → release into the box → optional return.

Under varied lighting, successful policies must: - reliably detect the block despite changes in brightness and color cast, - handle shadowed/overexposed regions, - maintain consistent grasp approach when the block’s visual features shift.

Dataset Structure

meta/info.json:

{
    "codebase_version": "v3.0",
    "robot_type": "so_follower",
    "total_episodes": 50,
    "total_frames": 22165,
    "total_tasks": 1,
    "chunks_size": 1000,
    "data_files_size_in_mb": 100,
    "video_files_size_in_mb": 200,
    "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.gripper": {
            "dtype": "video",
            "shape": [
                1080,
                1920,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 1080,
                "video.width": 1920,
                "video.codec": "av1",
                "video.pix_fmt": "yuv420p",
                "video.is_depth_map": false,
                "video.fps": 30,
                "video.channels": 3,
                "has_audio": false
            }
        },
        "observation.images.top": {
            "dtype": "video",
            "shape": [
                1080,
                1920,
                3
            ],
            "names": [
                "height",
                "width",
                "channels"
            ],
            "info": {
                "video.height": 1080,
                "video.width": 1920,
                "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
        }
    }
}

📊 Interested in the dataset? Contact Unidata to discuss purchase terms: https://unidata.pro

Citation

BibTeX:

[More Information Needed]