A deep learning / neuroevolution hybrid for visual control

This paper presents a deep learning / neuroevolution hybrid approach called DLNE, which allows FPS bots to learn to aim & shoot based only on high-dimensional raw pixel input. The deep learning component is responsible for visual recognition and translating raw pixels to compact feature representations, while the evolving network takes those features as inputs to infer actions. The results suggest that combining deep learning and neuroevolution in a hybrid approach is a promising research direction that could make complex visual domains directly accessible to networks trained through evolution.

[1]  Jürgen Schmidhuber,et al.  Evolving deep unsupervised convolutional networks for vision-based reinforcement learning , 2014, GECCO.

[2]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[3]  Bobby D. Bryant,et al.  Neuro-visual control in the Quake II game engine , 2008, 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence).

[4]  Shane Legg,et al.  Human-level control through deep reinforcement learning , 2015, Nature.

[5]  Julian Togelius,et al.  Neuroevolution in Games: State of the Art and Open Challenges , 2014, IEEE Transactions on Computational Intelligence and AI in Games.