A platform for developing AI bots that play Doom using visual information, designed for reinforcement learning research.
ViZDoom is an open-source platform for developing AI bots that play the 1993 game Doom using only visual information from the screen. It is built on the ZDoom engine and provides a customizable environment for research in machine visual learning and deep reinforcement learning. The platform allows researchers to train agents in complex, interactive scenarios to advance AI capabilities in perception and decision-making.
Researchers and developers in machine learning, particularly those focused on reinforcement learning, visual learning, and game AI who need a rich, performant environment for experimentation.
Developers choose ViZDoom for its high performance, ease of creating custom scenarios, and seamless integration with popular reinforcement learning frameworks like Gymnasium. Its unique selling point is providing a visually complex, game-based environment that is both lightweight and highly customizable for AI research.
Reinforcement Learning environments based on the 1993 game Doom :godmode:
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Achieves up to 7000 frames per second in synchronous mode on a modern CPU, enabling rapid iteration for reinforcement learning experiments.
Supports easy creation of custom scenarios using visual editors and ACS scripting, allowing tailored environments for specific research questions.
Includes Gymnasium and Gym wrappers, making it straightforward to integrate with popular reinforcement learning frameworks like PyTorch and TensorFlow.
Provides access to depth buffers, audio, actor lists, and map geometry, offering multiple modalities for AI perception beyond visual input.
Does not distribute original Doom graphics; users must source their own WAD files, which can be a barrier and limits the out-of-the-box visual experience.
The Windows version is explicitly noted as less tested and not recommended for serious experiments, pushing users towards Linux or macOS for reliability.
Installing from source requires CMake, Boost, SDL2, and other libraries, which can complicate setup, especially on systems without pre-built wheels.