SwarmRL

I developed the SwarmRL Environment for the Swarm Rescue simulator, creating a reinforcement learning setup that supports both single-agent and multi-agent scenarios. The environment adheres to the Gymnasium interface for single-agent use and the PettingZoo interface for multi-agent settings. I also implemented a custom Gymnasium version for multi-agent scenarios, tailored for non-standard observation and action spaces. Additionally, I wrote training pipelines to train models for both single-agent and multi-agent settings using algorithms like SAC, PPO, and more.