This is my first working(ish) attempt at making and deploygin a balancing policy for the 23DOF G1.

It borrows heavily from the different unitree repos and mujoco playground examples. If you make any improvements or create a separate balance policy, please let me know - I want to learn!

Qualifiers

• It's brittle and jank but it kind of works (i think?), so I thought I'd put it as a standalone repo on github in case this helps out someone else.
• I didn't cleanup/comment/organize the code at all. It's pretty small but might be difficult to follow, - you are absolutely welcome to ask me questions via email/twitter.
• This code is copy/pasted from the main repo I was derping around in and I didn't test it thoroughly - if you run into issues, again, you're totally welcome to ping me.

Instructions

Setup/pip install the following repos (https://github.com/unitreerobotics/unitree_sdk2_python) (https://github.com/google-deepmind/mujoco_playground) (https://github.com/google/brax)

(note: doing this from memory so there may be more...)

From inside the script folder, run python main.py to train a policy To deploy that policy run python deploy_real.py -- be sure to replace CHECKPOINT_PATH with your own and have an environment variable for NETWORK_CARD_NAME (per their docs)