A Robust Trajectory Planning Method Based on Historical Information for Autonomous Vehicles

Trajectory planning is key component for autonomous vehicles. However, most existing trajectory planning methods which utilized a set of fixed weights to evaluate an optimal trajectory from a set of trajectory candidates may change abruptly in complex dynamic scenarios. In this study, a robust trajectory planning was proposed based on historical information. The developed trajectory planning is mainly consisting of candidate trajectories generation module, the collision detection module, the stability detection module and the speed planning module. Firstly, the candidate trajectories are generated according to the vehicle kinematics model. And then the collision detection module is used to remove the dangerous trajectories. Then, the optimal trajectory is selected by the multi-attribute indicators evaluation, and the selected trajectory is checked by the stability detection module based on the lateral deviation of the optimal trajectory from the historical trajectory. Based on its result, the current optimal trajectory or the historical trajectory is chosen as the output trajectory. Finally, based on the output trajectory, the speed planning module is executed to generate a smooth trajectory. Simulation and field experiments were conducted to evaluate the effectiveness of the proposed method.