Datawhale/datawhale_eai_pnp_language

Datawhale Every Embodied MuJoCo PnP Language Dataset

This is a small LeRobot-format teaching dataset for the Every Embodied MuJoCo pick-and-place tutorial. It is used by the Pi0 and SmolVLA examples in:

06-策略抓取或抓取VLA/大模型控制、VLA、VLM/04mujoco复现ACT、Pi0、SmolVLA

Dataset Summary

• Format: LeRobot dataset v2.1
• Environment: MuJoCo OMY pick-and-place
• Episodes: 20
• Frames: 2621
• FPS: 20
• Tasks:
  • Place the blue mug on the plate.
  • Place the red mug on the plate.
• Main features:
  • observation.image: 256 x 256 RGB image
  • observation.wrist_image: 256 x 256 RGB wrist image
  • observation.state: 6-D robot state
  • action: 7-D action
  • obj_init: 9-D object initialization vector

Usage

Download the dataset into the tutorial directory:

git lfs install
git clone https://huggingface.co/datasets/Datawhale/datawhale_eai_pnp_language demo_data_language

Then keep the training configs as:

dataset:
  repo_id: datawhale_eai_pnp_language
  root: ./demo_data_language

The same dataset can be used by both pi0_omy.yaml and smolvla_omy.yaml.

Collecting Your Own Data

You do not have to use this dataset. The tutorial also provides notebooks and scripts for collecting your own MuJoCo demonstrations:

• 1.collect_data.ipynb: basic keyboard teleoperation dataset collection
• 5.language_env.ipynb: language-conditioned red/blue mug dataset collection
• 6.visualize_data.ipynb: dataset visualization

In the language-conditioned workflow, 5.language_env.ipynb creates a local LeRobot dataset at ./demo_data_language. You can train directly from that local folder, or upload your own collected dataset to Hugging Face with huggingface_hub / git-lfs and point dataset.root to the downloaded folder.

Notes

This dataset is intended for tutorial and reproduction use. For stable policy performance, inspect collected trajectories visually and verify success with the provided evaluation notebooks/scripts, since small demonstration datasets can be sensitive to action timing, gripper release, and object initialization.