A offline 2D tank-duel arena and a study of training an agent to play it with imitation + self-play. The goal is to reproduce the strategic mechanics of the ricochet-tank genre as an RL testbed
javascript python reinforcement-learning dagger pytorch behavioral-cloning imitation-learning game-ai 2d-game gymnasium ppo domain-randomization self-play stable-baselines3
-
Updated
Jul 2, 2026 - JavaScript