opendilab/GoBigger

Provides a scalable gym-like environment with configurable multi-agent teams competing cooperatively within partially observable state spaces, featuring complex ball mechanics (food, thorns, spores, clones) and action primitives (move, split, eject). Built on efficient game engine architecture with rich observation data including global match state and per-player vision-bounded overlaps. Supports Python 3.6+ via PyPI/Conda with straightforward environment creation and step-based interaction compatible with standard RL frameworks.

481 stars. No commits in the last 6 months.