Motion planning for autonomous vehicles remains challenging, especially in environments with multiple vehicles and high speeds. Autonomous racing offers an opportunity to develop algorithms that can deal with such situations and adds the requirement of following race rules. We propose a hybrid local planning approach capable of generating rule-compliant trajectories at the dynamic limits for multi-vehicle oval racing. The planning method is based on a spatiotemporal graph, which is searched in a two-step process to exploit the dynamic limits on the one hand and achieve a long planning horizon on the other. We introduce a soft-checking procedure that can handle cases where no collision-free, feasible, or rule-compliant solutions are found to restore an admissible state as quickly as possible. We also present a state machine explicitly designed for fully autonomous operation on a racetrack, acting on a higher level of the planning algorithm. It contains the interface to a race control entity and translates the current race rules and conditions into interpretable requests for the local planning algorithm. We present the results of experiments with a full-scale prototype, including overtaking maneuvers at speeds of up to 74 m/s.     «

Motion planning for autonomous vehicles remains challenging, especially in environments with multiple vehicles and high speeds. Autonomous racing offers an opportunity to develop algorithms that can deal with such situations and adds the requirement of following race rules. We propose a hybrid local planning approach capable of generating rule-compliant trajectories at the dynamic limits for multi-vehicle oval racing. The planning method is based on a spatiotemporal graph, which is searched in a...     »