erikluo/vizdoom

Doom-based AI Research Platform for Reinforcement Learning from Raw Visual Information. :godmode:

ViZDoom

http://vizdoom.cs.put.edu.pl

ViZDoom allows developing AI bots that play Doom using only the visual information (the screen buffer). It is primarily intended for research in machine visual learning, and deep reinforcement learning, in particular.

ViZDoom is based on ZDoom to provide the game mechanics.

ViZDoom is the platform for Visual Doom Competition @ CIG 2017.

Features

• Multi-platform,
• API for C++, Lua, Java and Python,
• Easy-to-create custom scenarios (examples available),
• Async and sync single-player and multi-player modes,
• Fast (up to 7000 fps in sync mode, single threaded),
• Customizable resolution and rendering parameters,
• Access to the depth buffer (3D vision)
• Automatic labeling game objects visible in the frame
• Off-screen rendering,
• Episodes recording,
• Time scaling in async mode,
• Lightweight (few MBs).

ViZDoom API is reinforcement learning friendly (suitable also for learning from demonstration, apprenticeship learning or apprenticeship via inverse reinforcement learning, etc.).

Cite as

Michał Kempka, Marek Wydmuch, Grzegorz Runc, Jakub Toczek & Wojciech Jaśkowski, ViZDoom: A Doom-based AI Research Platform for Visual Reinforcement Learning, IEEE Conference on Computational Intelligence and Games, pp. 341-348, Santorini, Greece, 2016 (arXiv:1605.02097)

Bibtex:

@inproceedings{Kempka2016ViZDoom,
  author    = {Micha{\l} Kempka and Marek Wydmuch and Grzegorz Runc and Jakub Toczek and Wojciech Ja\'skowski},
  title     = {{ViZDoom}: A {D}oom-based {AI} Research Platform for Visual Reinforcement Learning},
  booktitle = {IEEE Conference on Computational Intelligence and Games},  
  year      = {2016},
  url       = {http://arxiv.org/abs/1605.02097},
  address   = {Santorini, Greece},
  Month     = {Sep},
  Pages     = {341--348},
  Publisher = {IEEE},
  Note      = {The best paper award}
}

Installation/Building instructions

• PyPI (pip/conda)

• LuaRocks

• Linux

• MacOS/OSX

• Windows

Windows build

For Windows we are providing compiled runtime binaries and development libraries:

1.1.0 (2016-12-23):

• Python 2.7 (64-bit)
• Python 3.4 (64-bit)
• Python 3.5 (64-bit)
• Lua 5.1 & LuaJIT (64-bit)
• Java (64-bit)

Examples

Before running the provided examples, make sure that freedoom2.wad is placed in the same directory as the ViZDoom executable (on Linux and MacOS/OSX it should be done automatically by the building process):

• Python (contain learning examples implemented in Theano and TensorFlow)
• C++
• Lua (contain learning example implemented in Torch)
• Java

Python examples are currently the richest, so we recommend to look at them, even if you plan to use other language. API is almost identical for all languages.

See also the tutorial.

Documentation

Detailed description of all types and methods:

• DoomGame
• Types
• Configuration files
• Exceptions
• Utilities

Changelog for 1.1.X version.

Contributions

This project is maintained and developed in our free time. All bug fixes, new examples and scenarios are welcome! We are also open to features ideas and design suggestions.

License

Code original to ViZDoom is under MIT license. ZDoom uses code from several sources with varying licensing schemes.