Minimial implementation of PPO for learning a robotics policy in Jax.

step.mp4

NanoPPO

This is a minimalist implementation of training and inferencing a PPO policy for making humanoid robots walk.

How do I run this?

0. Install dependencies: pip install --pre -r requirements.txt
1. Start training a model: python -m train
2. Run inference on the trained model: python -m play video will be saved to video.mp4

Contributing

Before committing changes, run the following formatting commands:

make format
make static-checks