This is an implementation of Deep Reinforcement Learning as proposed from DeepMind (see: https://www.cs.toronto.edu/~vmnih/docs/dqn.pdf)
It is based on the ViZDoom (https://github.com/Marqt/ViZDoom) learning environment. The deep convolutional network is implemented in tensorflow (https://www.tensorflow.org/)
This proof of concept implementation, can serve as a starting point for further investigation. The implemented deep network is able to learn the basic scenario from ViZDoom within 1 hour on a GeForce GTX 970 + Core i7 4770.
For a detailed explanation see the Report.pdf.
For further learning more complex scenarios an addional layer should be added to the network.
For this to work you need ViZDoom and Tensorflow. See the installation_nodes.