Hey, have you ever wondered how AI agents learn to play games? I’ve been fascinated by this topic for years, and recently, I dedicated some time to dive deeper into Reinforcement Learning (RL). I spent a couple of hours each day in July implementing various RL algorithms, including DQN and Policy Gradient (REINFORCE), from scratch across multiple Atari games. It was a great learning experience, and I also utilized Stable Baselines for standardized benchmarking.
Now, I’m excited to share my progress and plans for the future. My goal is to expand the number of games and algorithms, creating a unified model that can play them all, similar to previous publications. I’m also planning to extend this to board games, enabling the creation of customized agents. Some of these agents will rely on well-known planning algorithms like Monte Carlo Tree Search, while others can clone the behavior of famous players.
The engineering challenge of designing a smart storage solution to index and serve all the games is also intriguing. I’ve made my repository public, so feel free to check it out and stay tuned for updates!
Reinforcement Learning has come a long way, and I’m excited to be a part of this journey. I hope to inspire others to explore this fascinating field and contribute to its growth.