Human-like Trajectory Planning for Autonomous Vehicles Using Flexible Virtual Reference Points in Car Overtaking Motorcycle Scenarios

Abstract

Traditionally, autonomous vehicles (AVs) prioritize safety and efficiency when planning trajectories. However, the lack of human-like driving behaviors can limit user trust and acceptance. This can be attributed to difficulties in communication and interpreting the intentions of other road users, particularly during interactions like overtaking maneuvers. This study proposes the Flexible Virtual Reference Point (FVRP) concept, an enhancement to model-based trajectory planning that facilitates human-like behavior during overtaking maneuvers. FVRP decomposes complex overtaking maneuvers into distinct phases. Within each phase, the concept strategically positions virtual reference points by considering both driver comfort data and prevailing traffic constraints. These reference points are then connected to form a piecewise trajectory, ensuring the optimization of both safety and comfort during overtaking maneuvers involving a motorcycle (MC). The results demonstrate that FVRP successfully generates trajectories that achieve both safety and driver comfort across all experimental settings. Furthermore, the trajectories generated by FVRP exhibit characteristics that resemble human drivers, while maintaining safe and comfortable distances compared to those generated by driver behavior models and optimal control models. The success of FVRP concept in car-to-MC overtaking maneuvers suggests its potential for adaptation to other driving maneuvers where comfort and safety need to be balanced.