Reachability computation of low-order models for the safety verification of high-order road vehicle models

We present an approach to verify the planned maneuvers of an automated car. The main idea is to compute the occupancy of the automated car on the road using reachable sets, which makes it possible to check if one collides with other traffic participants, or leaves the drivable area. The specialty of the presented approach is that all possible uncertainties in the form of sensor noise, uncertain friction coefficient, and uncertain initial states, are considered. Maneuvers are periodically verified on-board to account for the variety of possible traffic situations, requiring an efficient algorithm. Thus, the underlying vehicle model has to be a compromise between accuracy and simplicity. The inexactness of the model is compensated by adding disturbance to the model such that it contains high-order model behavior. This is demonstrated by exploring the state space with rapidly-exploring random trees (RRTs) of a high-order model and check whether it leaves the reachable area of the low-order model used for verification.