All Tools
V
AgentsFreeOpen Source
VIZDOOM
Doom-based reinforcement learning environments from pixels
MIT
ABOUT
Pixel-based RL needs a fast 3D world with custom rewards, but modern game engines are slow to reset and hard to script. ViZDoom runs Doom scenarios with API-level control of observations, buttons, and maps so researchers can train visual agents without waiting on a heavyweight simulator.
INSTALL
pip install vizdoomINTEGRATION GUIDE
1. Train a pixel-based DQN agent on a built-in ViZDoom combat or navigation scenario
2. Wrap ViZDoom with Gymnasium and plug it into Stable Baselines3 or Tianshou
3. Design a custom scenario with scripted rewards for multi-agent deathmatch research
4. Benchmark sample-efficient visual RL algorithms on a cheap, deterministic 3D env
TAGS
pythonreinforcement-learninggymnasiumvizdoompixelsfaramasimulation