Git hub link:<p><a href="https://github.com/mwydmuch/ViZDoom" rel="nofollow">https://github.com/mwydmuch/ViZDoom</a><p>I preferred to install this with python3 on Ubuntu16.04, the make process from the repo is messy. Once you have cloned the sources and have vizdoom installed with pip3, you can dive right in to installing tensorflow with pip3 (and cuda if you have the hardware)<p>For now I believe tensorflow-1.0.1 and cuDNN-2 with cuda 6.5 support vizdoom for compute level cards 2.0 and higher, at least thats what a search suggests.<p><a href="https://en.wikipedia.org/wiki/CUDA" rel="nofollow">https://en.wikipedia.org/wiki/CUDA</a><p><a href="https://developer.nvidia.com/cuda-legacy-gpus" rel="nofollow">https://developer.nvidia.com/cuda-legacy-gpus</a><p><a href="https://developer.nvidia.com/cuda-toolkit-65" rel="nofollow">https://developer.nvidia.com/cuda-toolkit-65</a><p><a href="https://developer.nvidia.com/rdp/cudnn-archive" rel="nofollow">https://developer.nvidia.com/rdp/cudnn-archive</a><p>Does anyone know the lower limits of hardware, libraries and SDKs for tensor flow 1.0.1?