Towards tactical behaviour planning under uncertainties for automated vehicles in urban scenarios

In recent years, automated driving has gained the public spotlight as more prototypes are tested amongst human drivers. Since driving is a social behavior, automated vehicles (AVs) have to cooperate with other road users in shared driving spaces. Therefore tactical behaviour planning with respect to the intentions of the surrounding road users is an essential functionality at the maneuver level. Moreover, the incorporation of uncertainty in road users intentions and sensor measurements is of high importance, especially in urban scenarios. This paper presents a generalized approach towards high level decision making for automated vehicles encountering by formulating the problem as a continuous Partially Observable Markov Decision Process (POMDP) by combining the advantage of two state of the art solvers - MCVI and SARSOP. Furthermore, the uncertainties in intention estimation model of the road users are incorporated using a discrete bayesian network. Validation has been done for longitudinal behavior planning in varied unsignalized intersection-based scenarios using the simulation software IPG CarMaker.