ehalicki/lewam_eval_pusht_selfwam_mot_mse
This dataset was created using LeRobot. Dataset Description Real-robot evaluation rollouts recorded on a Rebot B601 7-DoF arm, from the LeWAM project. Every episode here is a policy rollout on the physical robot — not a teleoperated demonstration — scored by a human operator immediately after it ran. Task: non-prehensile PushT (b601_pusht): the arm pushes a T-shaped block into a target pose without grasping it. Shares only its name with the simulated PushT. Trained on… See the full description on the dataset page: https://huggingface.co/datasets/ehalicki/lewam_eval_pusht_selfwam_mot_mse.
This dataset was created using LeRobot.
<a class="flex" href="https://huggingface.co/spaces/lerobot/visualizedataset?path=ehalicki/lewamevalpushtselfwammotmse"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/> </a>
Dataset Description
Real-robot evaluation rollouts recorded on a Rebot B601 7-DoF arm, from the LeWAM project. Every episode here is a policy rollout on the physical robot — not a teleoperated demonstration — scored by a human operator immediately after it ran.
- Task: non-prehensile PushT (
b601_pusht): the arm pushes a T-shaped block into a target pose without grasping it. Shares only its name with the simulated PushT. Trained on `ehalicki/b601_pusht`. - Policy: LeWAM self-supervised WAM (
selfwam_mot_mse), a mixture-of-transformers ViT predictor trained jointly on the action chunk and the future scene/wrist/state latents, with the vision encoder fine-tuned end to end - Training seed: 27 (the other arms in this set were trained with seed 14)
- Success rate: 8/30 = 26.7%
- Episodes: 30 · frames: 33020 · fps: 30 · robot time: 18.3 min
- Wall-clock: 40.1 min (median 73s, max 192s per episode) — longer than robot time because inference on this hardware runs slower than real time
- wandb run: `39zi1axd` (
eval-selfwam_mot_mse_b601_pusht_s27) - Episode cap: 5400 steps — an episode that hits it is scored a failure.
Observations
Three synchronised camera streams (observation.images.zed_left, observation.images.zed_right from a ZED Mini, observation.images.wrist_cam from a wrist-mounted USB camera) plus observation.state. action is the 7-DoF joint command the policy emitted.
Note the policies were served torque-only state (observation.state sliced to *.torq); the observation.state recorded here is the robot's full state vector, so reproducing what the policy actually saw means applying the same slice.
Per-episode scoring
eval_session.json holds the operator's label for each episode:
- Homepage: https://github.com/erykhalicki/LeWAM
- Paper: [More Information Needed]
- License: apache-2.0
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
21
],
"names": [
"shoulder_pan.pos",
"shoulder_lift.pos",
"elbow_flex.pos",
"wrist_flex.pos",
"wrist_yaw.pos",
"wrist_roll.pos",
"gripper.pos",
"shoulder_pan.torq",
"shoulder_lift.torq",
"elbow_flex.torq",
"wrist_flex.torq",
"wrist_yaw.torq",
"wrist_roll.torq",
"gripper.torq",
"shoulder_pan.vel",
"shoulder_lift.vel",
"elbow_flex.vel",
"wrist_flex.vel",
"wrist_yaw.vel",
"wrist_roll.vel",
"gripper.vel"
]
},
"observation.images.zed_left": {
"dtype": "video",
"shape": [
376,
672,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"is_depth_map": false,
"video.height": 376,
"video.width": 672,
"video.codec": "h264",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": null,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {}
}
},
"observation.images.zed_right": {
"dtype": "video",
"shape": [
376,
672,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"is_depth_map": false,
"video.height": 376,
"video.width": 672,
"video.codec": "h264",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": null,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {}
}
},
"observation.images.wrist_cam": {
"dtype": "video",
"shape": [
360,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"is_depth_map": false,
"video.height": 360,
"video.width": 640,
"video.codec": "h264",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": null,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {}
}
},
"action": {
"dtype": "float32",
"shape": [
7
],
"names": [
"shoulder_pan.pos",
"shoulder_lift.pos",
"elbow_flex.pos",
"wrist_flex.pos",
"wrist_yaw.pos",
"wrist_roll.pos",
"gripper.pos"
]
},
"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
}
},
"total_episodes": 30,
"total_frames": 33020,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"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",
"robot_type": null,
"splits": {
"train": "0:30"
}
}Citation
BibTeX:
[More Information Needed]