This dataset was created using LeRobot.
Dataset Summary
- Task: Place the red block into the white box (pick-and-place / block-in-bin).
- Key variation (hard negatives):
- The canonical “red start position” is occupied by either the green or beige block.
- The red block is relocated to a nearby position (close to the original spot).
- Additional variation (if applicable):
- Red block orientation varies across episodes.
- Decoy blocks (green/beige) may also vary in position/orientation.
Why This Dataset
In simpler setups, policies can succeed by memorizing a fixed reach to the usual red-block location. This dataset is designed to prevent that by introducing:
- decoy substitution at the most “obvious” location,
- a requirement for object identity-based selection (choose red, not “the block at the usual spot”),
- better testing of visual grounding and robustness to distractors.
Supported Tasks and Use Cases
This dataset is useful for:
- Imitation learning where the policy must truly identify the target object.
- Robust perception + control under distractors and small target translations.
- Failure analysis: measuring how often a policy grasps the decoy at the canonical position.
- Curriculum / benchmarking: compare performance vs.
- red-only fixed-position datasets,
- multi-block distractor datasets without decoy substitution.
Task Description
Instruction: “Place the red block into the white box.”
A typical episode: approach → grasp → transport → release into box → optional return.
The defining challenge is that the canonical start location is not reliable: it may contain a non-target block, while the red block is nearby.
Dataset Structure
{
"codebase_version": "v3.0",
"robot_type": "so_follower",
"total_episodes": 20,
"total_frames": 7148,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:20"
},
"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]
- Downloads last month
- 10