cair/deep-rts

C++ core engine achieves 6M+ simulation steps/second (15,000% faster than StarCraft), with Python bindings for ML toolkit integration. Provides multi-agent scenarios with configurable reward functions for task-specific training (resource collection, combat objectives), supporting map variants from 10x10 to 31x31 with 2-6 player free-for-all configurations. Designed as a lightweight alternative to commercial RTS environments for reinforcement learning research.

247 stars. No commits in the last 6 months.