Balancing Computation Speed and Quality: A Decentralized Motion Planning Method for Cooperative Lane Changes of Connected and Automated Vehicles

This paper focuses on the multi-vehicle motion planning (MVMP) problem for cooperative lane changes of connected and automated vehicles (CAVs). The predominant decentralized MVMP methods can hardly explore and utilize the cooperation capability of a multi-vehicle team, thus they usually lead to low-quality solutions. This paper proposes a two-stage MVMP framework to find high-quality online solutions. Concretely, at stage 1, the CAV platoon transfers from its original formation to a sufficiently sparse formation; at stage 2, all the CAVs simultaneously change lanes with collision avoidance implicitly ensured. The CAVs only involve longitudinal rather than lateral motions at stage 1, thus the collision-avoidance constraints can be easily handled. Since stage 2 begins with a sparse formation, the implicitly ensured collision avoidance can be completely omitted then. Through this, the proposed method avoids directly handling the challenging collision avoidance conditions, thereby being able to compute fast. As the vehicles run cooperatively and simultaneously at either stage, the obtained solutions are near-optimal. The completeness, effectiveness, and quality of the proposed two-stage MVMP method are validated through theoretical analysis and comparative simulations.