Datasetpublic

cloudwalk-research/psi0-g1-sneaker-94ep-v1

Psi0 G1 Sneaker-in-Box — 94 episodes (v1) 94 teleoperated episodes of a Unitree G1 humanoid (with Inspire RH56DFTP hands) picking up a sneaker and placing it inside a box. Right-hand only, single-handed task. Used to fine-tune the v1 Psi0 checkpoint at the CloudWalk Robotics Lab (CW-RL), 2026-04. To our knowledge, the v1 fine-tune was the first publicly reported external reproduction of the Psi0 paper's full pipeline (collect → finetune → real-robot deploy). Quick… See the full description on the dataset page: https://huggingface.co/datasets/cloudwalk-research/psi0-g1-sneaker-94ep-v1.

sourceHugging Faceapache-2.0updated 4mo agoView on Hugging Face

0likes244downloads

Dataset Card

Psi0 G1 Sneaker-in-Box — 94 episodes (v1)

94 teleoperated episodes of a Unitree G1 humanoid (with Inspire RH56DFTP hands) picking up a sneaker and placing it inside a box. Right-hand only, single-handed task. Used to fine-tune the v1 Psi0 checkpoint at the CloudWalk Robotics Lab (CW-RL), 2026-04. To our knowledge, the v1 fine-tune was the first publicly reported external reproduction of the Psi0 paper's full pipeline (collect → finetune → real-robot deploy).

Quick facts


Episodes	94
Frames	26 192 @ 30 fps
Format	LeRobot v2.1
Camera	egocentric, 640×480, H.264 yuv420p
Robot	Unitree G1 + Inspire RH56DFTP (5-finger, 6 active DoF/side, padded to 14-D in dataset)
Hand-side composition	94 RIGHT-hand episodes; LEFT hand held open (`1.0` × 6) throughout
Action dim	36 (6 LEFT hand + 1 pad + 6 RIGHT hand + 1 pad + 7 LEFT arm + 7 RIGHT arm + 3 torso RPY + 1 torso height + 4 locomotion/target)
State dim	32 (action minus 4 locomotion/target dims)
Task instruction	`"Pick up the sneaker and place it inside the box."`
Episode length	min 178, median 273, mean 279, max 469 frames
Data collection	Pico 4 Ultra teleop with hand trackers (CloudWalk Robotics Lab)
License	Apache-2.0

Action and state layout

Standard Psi0 layout shared across all sibling datasets in this collection. Dataset-specific behaviour for v1:

Range	Meaning	v1 (this dataset)
`action[0:6]`	LEFT hand 6 DoF (Inspire)	constant `1.0` (hand held open) — std = 0
`action[6]`	LEFT hand pad (Inspire→14-D align)	constant `0.0`
`action[7:13]`	RIGHT hand 6 DoF	active, mean ≈ 0.574, std ≈ 0.482, all 6 DoF in lockstep (Pico 4 Ultra trigger broadcast)
`action[13]`	RIGHT hand pad	constant `0.0`
`action[14:21]`	LEFT arm 7 joints (rad)	active
`action[21:28]`	RIGHT arm 7 joints (rad)	active
`action[28:31]`	Torso RPY	constant `0`
`action[31]`	Torso height	near-constant ≈ 0.749
`action[32:36]`	Locomotion (vx, vy, vyaw, target_yaw)	constant `0` (stationary task)

For the underlying hand hardware (Inspire RH56DFTP + Pico 4 Ultra trigger broadcasting) and the deploy-side bridge required to publish 36-D Psi0 actions to the Inspire DDS topic, see the v1 fine-tune model card: `cloudwalk-research/psi0-g1-sneaker-94ep-v1-finetune`.

How to download

python

import os
from huggingface_hub import snapshot_download
snapshot_download(
    repo_id='cloudwalk-research/psi0-g1-sneaker-94ep-v1',
    repo_type='dataset',
    local_dir='data/real/psi0-g1-sneaker-94ep-v1',
)

After download, run scripts/data/patch_lerobot_meta.py to strip stale huggingface metadata keys from the parquet schemas (required by datasets==3.6.0; see Psi0 issue #3).

Fine-tune produced from this dataset

—Model repo: `cloudwalk-research/psi0-g1-sneaker-94ep-v1-finetune`
—Training summary: 40 000 steps, 8 × B200, 3 h 58 min, eval/loss 37.88 → 0.0675 (full metrics on the model card).
—Real-robot deployment results: documented on the v1 model card.

Related datasets

Repo	Episodes	Hand split	Purpose
`psi0-g1-sneaker-94ep-v1` (this)	94	0 L / 94 R	v1 fine-tune (right-hand only)
`psi0-g1-sneaker-205ep-v2-source`	205	111 L / 94 R	canonical v2 source; the 94 R episodes here are the same as in this v1 dataset (indices 0–93 in the source)
`psi0-g1-sneaker-199ep-v2`	199	108 L / 91 R	v2 fine-tune training set
`psi0-g1-sneaker-6ep-v2-eval`	6	3 L / 3 R	v2 held-out open-loop eval

Attribution

Developed by cloudwalk-research in the CloudWalk Robotics Lab (CW-RL). Part of the Psi0 collection. Built to fine-tune the Psi0 base model; stored in the LeRobot v2.1 format.

Citation

If you build on this dataset, please cite both upstream Psi0 and this release:

bibtex

@misc{cwrl_psi0_g1_sneaker_94ep_v1_2026,
  title        = {Psi0 G1 Sneaker-in-Box --- 94 episodes (v1, right-hand only)},
  author       = {{CloudWalk Robotics Lab}},
  year         = {2026},
  howpublished = {Hugging Face dataset},
  url          = {https://huggingface.co/datasets/cloudwalk-research/psi0-g1-sneaker-94ep-v1}
}

@article{psi0_2026,
  title         = {Psi0: A Generalist Vision-Language-Action Model for Humanoid Robots},
  author        = {Wei, Songlin and others},
  year          = {2026},
  eprint        = {2603.12263},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2603.12263}
}

@article{black2024pi0,
  title   = {{$\pi_0$}: A Vision-Language-Action Flow Model for General Robot Control},
  author  = {Black, Kevin and Brown, Noah and Driess, Danny and others},
  journal = {arXiv preprint arXiv:2410.24164},
  year    = {2024}
}

@software{lerobot2024,
  title  = {{LeRobot}: State-of-the-art Machine Learning for Real-World Robotics in {PyTorch}},
  author = {{Hugging Face}},
  url    = {https://github.com/huggingface/lerobot},
  year   = {2024}
}