Decision Making through Occluded Intersections for Autonomous Driving

Driving in unsignalized intersections poses significant challenges for autonomous vehicles, as various uncertainties are involved. On one hand, behaviors of vehicles in the observable area need to be carefully predicted, on the other hand, the occluded area has to be addressed judiciously, as potential obstacles might cause serious danger. In this paper, decision making through occluded intersection is discussed and both static and dynamic obstacles are considered. A general approach is proposed to deal with different occlusion scenarios. The approach is based on partially observable Markov decision process (POMDP) and customized for autonomous driving. A variety of experiments are conducted to evaluate the proposed approach, and results show that the proposed approach enables autonomous vehicle to exhibit a human-like driving behavior when facing occlusions.