Research On Motion Planning Of Autonomous Driving On Urban Road

With the development of science and technology,especially the significant progress of computer vision,autonomous driving technology has become a popular research field in recent years.Autonomous driving is a cross disciplinary and interdisciplinary technology.With the development of automotive intelligence,autonomous driving has gradually become a consideration factor for more and more consumers when purchasing vehicles.At the same time,autonomous driving technology can significantly increase urban traffic efficiency and significantly reduce accident rates.Due to the high complexity and random scenarios of urban roads,the autonomous driving technology of urban roads is currently the main technical difficulty.To achieve autonomous driving,multiple algorithm modules need to work together,and the planning and control module is one of the key modules of autonomous driving.The research content of this paper is the motion planning algorithm in the planning and control module.According to the two common conditions in automatic driving,the motion planning algorithm is divided into the trajectory planning algorithm under the structured road and the trajectory planning algorithm in the open space.The main work of this thesis is as follows:Firstly,the vehicle kinematics equation and vehicle dynamics equation are derived for the driverless vehicle according to the bicycle model hypothesis,vehicle geometric relationship and Newton mechanical relationship,and the coordinate transformation formula in the Frenet frame is derived,as well as the method of approximate vehicle model in the Frenet frame,Piesewise Jerk.In order to achieve obstacle avoidance function,several commonly used vehicle contour description methods and obstacle avoidance constraint conditions were introduced and derived,including rectangular description,circular description,and elliptical description.Then,for the autonomous driving trajectory planning algorithm on structured roads,a spatiotemporal decoupling algorithm framework was adopted.By introducing the Frenet frame,the spatiotemporal coupling problem originally in the Cartesian frame was transformed into a problem in the S-L-T coordinate system.At the same time,the S-L-T coordinate system can be decoupled in pairs,transforming it into a path planning problem in the S-L coordinate system and a speed planning problem in the S-T coordinate system.Then,the PienewserJerk method is introduced to transform the path planning problem into a Quadratic programming(QP)problem in the Frenet frame.Due to the drawbacks of this algorithm in dealing with complex environments,this thesis proposes an obstacle avoidance cost function in the Frenet frame,and adds this cost function as a soft constraint to the total cost.An adaptive iterative solution strategy is proposed,which can dynamically adjust the obstacle avoidance amplitude based on the distance of the obstacle entering the current lane,making the algorithm more robust,more adaptable to the complex working conditions of urban roads.Through simulation experiments,the improved algorithm proposed in this thesis has better comfort and real-time performance compared to the EM Planner based on the Frenet frame and the hybrid A* algorithm based on search strategy.Secondly,for scenarios such as u-turn and unprotected left turns,the trajectory planning algorithm under structured roads has certain flaws that make it difficult to handle.Therefore,this thesis studies trajectory planning algorithms in open spaces and adopts a front-end search and back-end optimization algorithm architecture.This thesis uses the hybrid A* algorithm to search for initial feasible solutions and serves as input for the back-end optimization algorithm.Secondly,this thesis makes certain adaptive adjustments to the hybrid A* based on the special scenarios,such as replacing the Reed Shepp curve with the DUBINS curve.In terms of backend optimization algorithms,this thesis adopts the CILQR algorithm based on optimal control theory to achieve smooth and precise obstacle avoidance of trajectories,and proposes to combine the control barrier function with the CILQR algorithm.Through simulation experiments,the CILQR-CBF algorithm proposed in this thesis can better handle two scenarios:u-turn and unprotected left turns.