vpraise00/dynavla-libero-goal-t7-frictionloss-oracle-stall-val
This dataset was created using LeRobot. Dataset Structure meta/info.json: { "codebase_version": "v3.0", "robot_type": "panda", "total_episodes": 20, "total_frames": 2223, "total_tasks": 1, "chunks_size": 1000, "data_files_size_in_mb": 100, "video_files_size_in_mb": 200, "fps": 20, "splits": { "train": "0:20" }, "data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet", "video_path":… See the full description on the dataset page: https://huggingface.co/datasets/vpraise00/dynavla-libero-goal-t7-frictionloss-oracle-stall-val.
This dataset was created using LeRobot.
<a class="flex" href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=vpraise00/dynavla-libero-goal-t7-frictionloss-oracle-stall-val"> <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
- Homepage: [More Information Needed]
- Paper: [More Information Needed]
- License: apache-2.0
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "panda",
"total_episodes": 20,
"total_frames": 2223,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 20,
"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": {
"observation.images.image": {
"dtype": "video",
"shape": [
360,
360,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 360,
"video.width": 360,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 20,
"video.channels": 3,
"has_audio": false
}
},
"observation.images.image2": {
"dtype": "video",
"shape": [
360,
360,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 360,
"video.width": 360,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 20,
"video.channels": 3,
"has_audio": false
}
},
"observation.state": {
"dtype": "float32",
"shape": [
8
],
"names": [
"eef_x",
"eef_y",
"eef_z",
"axis_x",
"axis_y",
"axis_z",
"gripper_0",
"gripper_1"
]
},
"action": {
"dtype": "float32",
"shape": [
7
],
"names": [
"dx",
"dy",
"dz",
"drx",
"dry",
"drz",
"gripper"
]
},
"dynavla.policy.a_base": {
"dtype": "float32",
"shape": [
7
],
"names": null
},
"dynavla.policy.chunk_fresh": {
"dtype": "bool",
"shape": [
1
],
"names": null
},
"dynavla.dynamics.param": {
"dtype": "string",
"shape": [
1
],
"names": null
},
"dynavla.dynamics.value": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"dynavla.dynamics.readback": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"dynavla.dynamics.nominal": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"dynavla.dynamics.is_nominal": {
"dtype": "bool",
"shape": [
1
],
"names": null
},
"dynavla.task_id": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"dynavla.seed": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"dynavla.success": {
"dtype": "bool",
"shape": [
1
],
"names": null
},
"dynavla.reward": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"dynavla.steering.gain": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"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
}
}
}Citation
BibTeX:
[More Information Needed]DynaVLA
Purpose: Dynamics Adapter correction targets. Oracle stall-triggered ramp steering (data-generation only, uses privileged knob qpos/qvel): while the gripper is closed and the knob does not move despite a twist command, the gain on dz/drz ramps up (0.1/step, max 2.0) and ramps down (0.2/step) once it moves; the steering delta is rate-limited (0.15/step). action = a, `dynavla.policy.a_base` = raw VLA output, `dynavla.steering.gain` = gain at that step. Success episodes only*; failed seeds re-collected at the same init state with a fresh noise draw.
Split: validation split — values [0, 4, 8, 12] × seeds 40–44 (20 episodes)
liberogoal task 7 `turnonthestove, dynamics = joint frictionloss of flatstove1button` (nominal 0). Frozen SmolVLA (`HuggingFaceVLA/smolvlalibero, n_action_steps=10, 360x360, fixed benchmark init states; seed = init-state index). Standard features match the checkpoint I/O: observation.images.image/image2 (180° rotated, checkpoint convention), observation.state (8: eef_pos, axis-angle, gripper_qpos), action (7: dpos, daxisangle, gripper). Extra dynavla.* columns (not policy inputs): dynavla.policy.abase` (raw VLA output), `dynavla.policy.chunkfresh, dynavla.dynamics.{param,value,readback,nominal,isnominal}` (ground-truth dynamics label, never a policy input), `dynavla.taskid/seed/success/reward, dynavla.steering.gain. Episode-level contract in meta/dynavlaepisodes.jsonl`, dataset manifest in `meta/dynavlamanifest.json, validator report in meta/dynavla_validation.{json,txt}`. Source: https://github.com/vpraise00/DynaVLA
Episodes: 20, frames: 2223, fps 20.