pogema

POGEMA (Partially-Observable Grid Environment for Multiple Agents) is a grid-based environment designed for studying the Partially Observable Multi-Agent Pathfinding (PO-MAPF) problem, which differs from traditional MAPF by incorporating decentralized decision-making and limited agent observations. In this setting, agents must make decisions based on partial information about the environment, planning and executing actions simultaneously. The goal is for the agents to reach their respective targets while avoiding collisions with other agents and obstacles. POGEMA supports various configurations, including customizable grid sizes and agent densities, and generates random maps and start/goal locations, making it highly flexible for different PO-MAPF scenarios.

The POGEMA environment is scalable, offering configurations that range from simple to complex grid setups. It includes predefined environments with varying levels of difficulty, such as Easy, Normal, Hard, and ExtraHard grids of different sizes (e.g., 8×8, 16×16, 32×32). Users can also integrate POGEMA with popular Multi-Agent Reinforcement Learning (MARL) frameworks like PettingZoo, PyMARL, and SampleFactory for enhanced experimentations.