Dynamic Decision-making in Continuous Partially Observable Domains: A Novel Method and its Application for Autonomous Driving

Decision-making is a crucial challenge on the way to fully autonomous systems. In real world tasks, assessing the consequences of decisions is aggravated by two factors: uncertainty and the continuous nature of the environment. In this work, we develop a general method for solving continuous partially observable Markov decision processes (POMDPs) that combines learning and planning. We apply it to autonomous driving in urban scenarios with hidden objects and cooperative driver interactions.