Farama-Foundation/ViZDoom

What it solves

ViZDoom is a research platform designed to help developers create AI bots that can play the game Doom using only visual information from the screen buffer. It provides a controlled environment for studying machine visual learning and deep reinforcement learning.

How it works

Built on the ZDoom engine, ViZDoom provides APIs for Python and C++ to interact with the game. It allows AI agents to perceive the game world through the screen buffer, depth buffer (3D vision), audio buffer, and lists of game objects and map geometry. It supports both synchronous and asynchronous modes for single-player and multiplayer experiences, and integrates with Gymnasium/Gym wrappers for easier reinforcement learning implementation.

Who it’s for

It is primarily intended for researchers in machine visual learning and deep reinforcement learning who need a high-performance, lightweight environment for training and testing AI agents.

Highlights

• High performance: capable of up to 7,000 frames per second in sync mode on a modern CPU.
• Comprehensive sensory input: access to visual, depth, and audio buffers, as well as in-game text and object lists.
• Flexible environment: supports custom scenario creation via visual editors and a powerful scripting language.
• Broad compatibility: multi-platform support for Linux, macOS, and Windows with Python and C++ APIs.