This is a multi-agent reinforcement learning simulation I’ve been building as a personal project. It’s a grid-based combat environment with per-agent PPO training runtime. Some of the things I’ve been experimenting with: – Per-agent PPO (isolated optimizers per agent) – Runtime checkpointing and resume chains – Headless-mode live CSV telemetry logging – Config-driven experiment control The repo includes the simulation engine, PPO runtime, and telemetry tooling.