Reinforcement learning is considered to be a strong AI paradigm which can beused to teach machines through interaction with the environment and learningfrom their mistakes. Despite its perceived utility, it has not yet beensuccessfully applied in automotive applications. Motivated by the successfuldemonstrations of learning of Atari games and Go by Google DeepMind, we proposea framework for autonomous driving using deep reinforcement learning. This isof particular relevance as it is difficult to pose autonomous driving as asupervised learning problem due to strong interactions with the environmentincluding other vehicles, pedestrians and roadworks. As it is a relatively newarea of research for autonomous driving, we provide a short overview of deepreinforcement learning and then describe our proposed framework. Itincorporates Recurrent Neural Networks for information integration, enablingthe car to handle partially observable scenarios. It also integrates the recentwork on attention models to focus on relevant information, thereby reducing thecomputational complexity for deployment on embedded hardware. The framework wastested in an open source 3D car racing simulator called TORCS. Our simulationresults demonstrate learning of autonomous maneuvering in a scenario of complexroad curvatures and simple interaction of other vehicles.
展开▼
