RL Suite: Game AI Agents

A comprehensive exploration of Reinforcement Learning algorithms applied to complex environments. For Checkers, I engineered a custom Gymnasium environment from scratch to handle multi-step capturing logic and trained a DQN agent with a massive action space (4096 outputs), benchmarking performance via self-play ELO ratings.

For Atari Breakout, I implemented a Deep Q-Network with CNNs, utilizing frame stacking to allow the agent to perceive ball velocity and trajectory from raw pixels. Additionally, I built a pure algorithmic solver for a Block Game using Monte Carlo Tree Search (MCTS), enabling the AI to look ahead and strategize effectively without model training.