M

MARL

Multi-agent reinforcement learning testbed: a top-down 2D defense game where a team of agents must protect a flag from waves of enemies that get faster and more numerous over time. Fully batched PyTorch simulation on CUDA (thousands of arenas in parallel), PPO / MAPPO with MLP, GRU, LSTM and attention policies, benchmarks. The goal: cooperation that emerges from a shared reward.