Reinforcement Learning for Constraint Satisfaction Game Agents

This work explores reinforcement learning for constraint satisfaction game environments, progressing from tabular Q-learning and DQN experiments to an adversarial PPO-based Wumpus agent. The system transforms the traditional static Wumpus into an adaptive RL-driven adversary capable of learning pursuit behaviour through scent-memory tracking and reward shaping. The project includes a full-stack deployment using FastAPI, React, Firebase, and Gymnasium-compatible training environments.