Interactive Planning for the Intersections with Uncertainty of Observed Vehicle’s Intentions and Occlusion Areas

Unsignalized intersections are one of the common urban driving scenarios, which are highly uncertain and challenging for planning. The main uncertainties at unsignalized intersections are intentions of observed vehicles and situations of occlusion areas. This paper proposes an algorithm to jointly infer the observed and occluded uncertainties, and apply it to the model of Partially Observable Markov Decision Process. This method can make full use of historical information observed by ego vehicle. Simulations demonstrate that the proposed interactive planner can generate more efficient trajectories.