ArnaudFickinger/gym-multigrid

Built on MiniGrid, this environment extends gridworld simulation with configurable multi-agent support, supporting both fully and partially observable state representations where each cell encodes object type, color, agent direction, and carried items. Core actions include movement, object manipulation (pick up/drop/toggle), and team-based gameplay through included environments like SoccerGame and CollectGame with customizable agent counts, goals, and reward structures. Integrates directly with OpenAI Gym as a standard environment, requiring only Python 3.5+, NumPy, and Matplotlib.

213 stars. No commits in the last 6 months.