Integrating Global Path Information With Advanced Risk Assessment: An Enhanced Potential Field Method for Intelligent Connected Vehicles Local Path Planning

Potential field (PF) approaches are widely used in local path planning due to their computational simplicity, but they still have many drawbacks when applied to self-driving cars, and the traditional PF suffers from isotropy, ignoring the real-time changes in vehicle behavior and local optimal traps. To address the above problems, global path information enhanced risk PF (RPF)-based local path is proposed. The RPF is based on the traditional PF, which is composed of the vehicle’s motion state information combined with a Gaussian function as a way to overcome the limitation of isotropy. In addition, we collect information about the surrounding environment through various types of traffic sensors, build directed graphs to describe the motion dependencies in the traffic environment, and design a vehicle-vehicle risk matrix based on the graph attention network model, which is able to adjust the RPF according to the real-time motion state of the vehicle. For the local optimal trap problem, the global path information is quickly obtained by combining with the heap logarithm A* (HLA*) algorithm, which helps the RPF to jump off faster from the local minimum problem. Finally, local paths are smoothed by combining Quintic Bézier curve taking into account vehicle kinematics and curvature constraints. The results of simulation experiments and real-vehicle tests show that our proposed method performs superiorly in local path planning compared with the traditional PF method.