A-MARL Framework Demonstrating Adversarial Learning Abilities

Alpha Tank is a multi-agent tank battle game built with Pygame and designed for Reinforcement Learning (RL) training. We want to create a full end-to-end RL pipeline (from environment to learning algorithms) as a deomstartion of showcasing how RL may learn from their opponents (depends on who, maybe another RL agent (i.e. PPO, SAC) or an intelligent bot (i.e. BFS bot, A* bot)) and use their charcteristics, along with the environement setup, to fight againts them and optimzie the reward.

Here is our Demo Wandb Report for our training. We have prsented our project at Galbraith Hall, UCSD, winter 2025. Checkout our latest manuscrip uploaded here. Our project was one of the top from CSE151b/251b in UCSD Winter 2025 among 30-ish other projects.

Demo of the agent fighting against bots.