https://github.com/user-attachments/assets/af664888-200a-4969-a248-1ef59d8b259c

BDX-R in mjlab

This repository showcase the implementation of the BDX-R robot into mjlab.

[!WARNING]
As mjlab is still in early development, this repository may be impacted by breaking changes. If an issue were to arise when running one of the scripts, feel free to open an issue or contribute to the project. Thanks you for your understanding!

🤖 What's mjlab?

mjlab is a project to have the Isaac Lab API using MjWarp as the backend. If you’re wondering about the motivation behind it or how it differs from Newton, you can learn more about it here.

🚀 Quickstart

Clone the repository.

bashCopy
git clone https://github.com/BDX-R/BDX-R-MjLab.git && cd BDX-R-MjLab

List available environments.

bashCopy
uv run bdx_r_list_envs

Use the dummy agents.

bashCopy
uv run bdx_r_play Mjlab-Velocity-Flat-BDX-R-Legs --agent zero # send zero actions to the robot
uv run bdx_r_play Mjlab-Velocity-Flat-BDX-R-Legs --agent random # send random actions to the robot

Velocity Tracking

Train the policy.

bashCopy
uv run bdx_r_train Mjlab-Velocity-Flat-BDX-R-Legs --env.scene.num-envs 4096

Evaluate the policy.

bashCopy
uv run bdx_r_play Mjlab-Velocity-Flat-BDX-R-Legs-Play --wandb-run-path your-org//run-id

🎯 Roadmap

• Use IMU for observations
• Use simplified meshes for collisions
• Deploy the policy on real robot
• Use rough terrains
• Add a head to the robot

Check here for the general roadmap of the project!

🛠️ Contributing

We look forward for contributions. Before submitting a PR, please run the following command for format:

bashCopy
make format

🙏 Acknowledgements

We're grateful to the people behind mjlab, MuJoCo Warp and Isaac Lab.