The study presents a novel approach to decision making for autonomous vehicles using reachable sets. The proposed decision module identifies potential driving corridors by computing the reachable set for the vehicle’s longitudinal position. It then selects the best driving corridor based on a cost function that penalizes lane changes and deviations from a desired velocity profile. The module generates a reference trajectory within the chosen corridor, guiding low-level trajectory planners. The decision module, when combined with a motion-primitive-based and optimization-based planner, shows significant speed-ups in performance when tested on 2000 challenging traffic scenarios and the CARLA simulator.
Publication date: 21 Sep 2023
Project Page: https://arxiv.org/abs/2309.12289v1
Paper: https://arxiv.org/pdf/2309.12289
