PPO impementation using pytorch

The code was adapted from https://github.com/vwxyzjn/ppo-implementation-details.

Please refer to https://iclr-blog-track.github.io/2022/03/25/ppo-implementation-details/ for more reading.

Get started

1. build the docker images

docker build . -t ppo-dev 

2. create a container and expose the port (for tensorboard)

# specify OUPUT_DIR 
docker run -v $OUTPUT_DIR:/app/ -p 6006:6006 --rm -it ppo-dev /bin/bash

3. run the training script (with your own parameters)

python train.py num_iterations=50000 rollout_length=100

The model file, training logs and evaluation video can be found in $OUTPUT_DIR.