Real‑Time Optimal Trajectory Planning for Autonomous Driving with Collision Avoidance Using Convex Optimization

作     者：Guoqiang Li Xudong Zhang Hongliang Guo Basilio Lenzo Ningyuan Guo 

作者机构：School of Mechanical EngineeringBeijing Institute of TechnologyBeijing 100081China Institute for Infocomm ResearchASTARSingaporeSingapore Department of Industrial EngineeringUniversity of PadovaPaduaItaly 

出 版 物：《Automotive Innovation》 (汽车创新工程（英文）)

年 卷 期：2023年第6卷第3期

页      面：481-491页

核心收录：

学科分类：0402[教育学-心理学(可授教育学、理学学位）] 0711[理学-系统科学] 0401[教育学-教育学] 08[工学] 080204[工学-车辆工程] 0837[工学-安全科学与工程] 0802[工学-机械工程] 0811[工学-控制科学与工程] 

基　　金：supported by Natural Science Foundation of Beijing(Grant No.3212013) Young Elite Scientists Sponsorship Program by CAST and Beijing JinQiao Project. 

主　　题：Trajectory planning Collision avoidance Model predictive control Autonomous driving 

摘      要：An online trajectory planning method for collision avoidance is proposed to improve vehicle driving safety and comfort simultaneously.The collision-free trajectory for autonomous driving is formulated as a nonlinear optimization problem.A novel approximate convex optimization approach is developed for the online optimal trajectory in both longitudinal and lateral directions.First,a dual variable is used to model the non-convex collision-free constraint for driving safety and is calculated by solving a dual problem of the relative distance between vehicles.Second,the trajectory is further optimized in a model predictive control framework considering the safety.It realizes continuous-time and dynamic feasible motion with collision avoidance.The geometry of object vehicles is described by polygons instead of circles or ellipses in traditional methods.In order to avoid aggressive maneuver in the longitudinal and lateral directions for driving comfort,rates of the acceleration and the steering angle are restricted.The final formulated optimization problem is convex,which can be solved by using quadratic programming solvers and is computationally efficient for online application.Simulation results show that this approach can obtain similar driving performance compared to a state-of-the-art nonlinear optimization method.Furthermore,various driving scenarios are tested to evaluate the robustness and the ability for handling complex driving tasks.