About

Ultimate Volleyball was inspired by @hardmaru's Slime Volleyball gym. I built the 3D environment on Unity's game engine, and trained agents to play competitive volleyball using self-play reinforcement learning in ML-Agents.

The volleyball players (agents) are given a reward for either hitting the ball over the net, or scoring a point. Over the course of hundreds of thousands - millions of runs, the players eventually learn to return the volleyball as well as employ key tactics such as hitting it towards the edges of the court where it is harder for the opponent to return.

The project is fully open-source and available on GitHub (80+ stars).

I also wrote a 5-part deep reinforcement learning tutorial which teaches you how to train your own AI to play volleyball.