PPO-PyTorch

This repository provides a Minimal PyTorch implementation of Proximal Policy Optimization (PPO) with clipped objective for OpenAI gym environments. It is primarily intended for beginners in Reinforcement Learning for understanding the PPO algorithm. It can still be used for complex environments but may require some hyperparameter-tuning or changes in the code.

Modified from https://github.com/tangyudi/Ai-Learn

Usage

• To train a new network : run PPO_continuous.py
• To train a new network : run PPO.py
• To train a test network : run test_continuous.py
• To train a test network : run test.py

Dependencies

Trained and tested on:

gym==0.19.0 
pyglet==1.5.27  
box2d box2d-kengz 
gym[box2d]
torch==2.0.1+cu117

If you still have problems, you can check requirement.txt.

References

• VMPO paper
• OpenAI Spinning up