Athens
This is an environment to do RL training and other forms where an LLM can play the game teeworlds https://teeworlds.com/, 4 bots shown in video
Core trick
Wraps the real Teeworlds 0.7.5 server with Python bot clients so an LLM can play multi-agent deathmatch via OpenEnv.
Tags
Description
This is an environment to do RL training and other forms where an LLM can play the game teeworlds https://teeworlds.com/, 4 bots shown in video
README excerpt
# TeeUnit: Multi-Agent Arena Environment A multi-agent deathmatch environment for LLM reinforcement learning, powered by the real [Teeworlds](https://www.teeworlds.com/) game engine. Built on the [OpenEnv](https://github.com/meta-pytorch/OpenEnv) framework. [](https://github.com/meta-pytorch/OpenEnv) [](https://github.com/teeworlds/teeworlds) [](LICENSE) --- ## Overview TeeUnit wraps the **real Teeworlds 0.7.5 game server** with Python bot clients, exposing it as an OpenEnv-compatible environment for **multi-agent reinforcement learning research**. ### Why Real Teeworlds? - **Authentic physics**: Proper momentum, hook mechanics, projectile behavior - **Battle-tested netcode**: Robust UDP protocol with delta compression - **Rich game state**: Full snapshot data (positions, velocities, weapons, projectiles) - **Extensible**: Access to all Teeworlds game modes and maps ### Key Features - **Turn-based mode**: Configurable ticks-per-step for LLM compatibility - **4 bot agents**: Connect as real game clients via Teeworlds protocol - **Full observability from snapshots**: Positions, health, weapons, projectiles - **Simple K/D rewards**: Clear reward signal for training - **OpenEnv compatible**: Works with TRL, torchforge, and more - **Docker deployable**: Ready for