This program trains RL agents to play games in µniverse. It uses Trust Region Policy Optimization to optimize neural networks policies.

This is intended to be used as a baseline. It gives an idea of how well traditional deep-RL does on µniverse games.

This package uses anyplugin. Thus, you can add -tags cuda to any go run, go build, or go get command to enable CUDA. You will want to see the cuda build instructions. You will also want to download muniverse-agent with its CUDA dependencies:

go get -u -tags cuda github.com/unixpickle/muniverse-agent