Research On Trajectory Planning Of Autonomous Driving Vehicles Under Urban Road

Autonomous driving is a complex system integrating information communication technology,artificial intelligence technology,vehicle chassis domain control and other technologies.It is an important development direction of the automobile industry.Among them,trajectory planning,as the key technology of autonomous driving,is an important link to reflect the level of intelligence.The quality of the planned trajectory directly determines the safety,efficiency,comfort and other performance of autonomous vehicle.Focuses on urban structured roads,this paper studies and designs a trajectory planning algorithm with lateral and longitudinal decoupling,which is suitable for roads and intersections,can deal with various working conditions such as lane changing,overtaking,yield,breaking and stop,and meets the requirements of realtime and inter-frame trajectory continuity.In this paper,trajectory planning is divided into behavior planning and motion planning.Firstly,the common geodetic,vehicle and Frenet coordinate systems are introduced,and the mutual conversion formula is given,and the upstream discrete guidance line is smoothed.Secondly,in order to handle complex urban traffic scenarios,the behavior planning algorithm based on hierarchical state machine is designed.At the top level,the intersection and in-road state are divided and switched according to the vehicle location and high-precision map information.At the middle level,each sub-state is divided based on rules.According to the in-road lane change scenario,a three-lane multi-vehicle lane change model is established under the Frenet coordinate system,and the possible collision forms are analyzed critically,so as to realize the preliminary judgment of the feasibility of lane change.Through the state switch and decision output of the state machine,it provides guidance for the underlying motion planning and limits its planning space,thereby reducing unnecessary sampling.In motion planning,in order to reduce the complexity of solution,the three-dimensional planning problem is decoupled into two two-dimensional planning problems.Through path planning and speed planning,the planning trajectory with time information is finally generated.In path planning,firstly,S-L graph is constructed based on smoothed local guideline,road information,behavior planning results,and static obstacles,and feasible planning space is generated.In the process of lateral and longitudinal sampling,multiple stages are divided in the longitudinal direction,and adjacent stages and cross-stage connections are supported,which can improve the diversity of sampling paths.In the lateral direction,step sampling is adopted,that is,coarse sampling is used to sparsely sample within the feasible range,the quality of each candidate path is evaluated,and the optimal path is solved by dynamic programming,and then the vicinity of the path solution is densely sampled through fine sampling,so that obtain better quality solutions.The evaluation of path quality takes into account factors such as curvature,safety,and lateral deviation from the target lane guideline.In order to further improve the path quality,by establishing the objective function and imposing constraints,it is optimized by quadratic programming,and the final path is generated.For speed planning,based on the construction of S-T graph,this paper realizes speed curve planning through two processes of search and optimization.Firstly,the discrete S-T graph is constructed based on the optimal path generated by path planning and the predicted trajectory of dynamic obstacles in the future.In order to narrow the search scope,invalid areas are eliminated according to the current initial state of the vehicle,acceleration and deceleration capacity,speed limit and other factors.During node-searching,the improved JPS algorithm is adopted,and the three-direction expansion mode is adopted to meet the nonreversing driving demand of vehicles.Its expansion cost comprehensively considers the driving safety,efficiency and comfort,and improves the quality of the rough solution of speed planning.Considering that the search target is not unique,its heuristic function is set as the weighted average of different target states.For the rough solution of the speed curve,after the preliminary treatment of reconnection,the quality of the solution is further optimized by the quadratic programming method,and finally a safe and comfortable speed curve that meets the vehicle kinematics constraints is obtained.At the same time,in order to ensure the continuity of the trajectory of two adjacent frames of the planning algorithm,this paper defines the index of trajectory similarity between frames,gives its calculation and evaluation methods,and takes it as a requirement for screening path solutions during path planning.By reasonably setting the matching distance threshold and similarity threshold,the path with a large difference from the previous frame trajectory can be deleted from the candidate path,thus ensuring that the inter-frame trajectory maintains good continuity under the same behavior state.Finally,a joint simulation platform based on SCANe R/Simulink/ROS is built,and the simulation and analysis of various working conditions are realized by combining the upstream and downstream modules such as navigation and positioning,sensing and tracking control of autonomous vehicle.The simulation results show that the algorithm proposed in this paper can meet the planning requirements of driving vehicles on urban structured roads,effectively deal with various working conditions such as in-road and intersection,and improve and verify the inter-frame trajectory continuity.