Research On Autonomous Driving Decision And Planning Method For Meeting Scenarios

In recent years,with the rapid development of artificial intelligence technology,the fiery capital market for autonomous driving,and the continuous reduction in sensor costs,technological innovation and application in the field of autonomous driving have ushered in unprecedented development.As common and easy-to-solve problems continue to be solved,uncommon and difficult-to-solve problems gradually become the new challenges facing the development of autonomous driving.As one of the core technologies of autonomous driving,the difficulty of decision and planning technology lies in non-convexity and interactivity.The car meeting problem in meeting scenarios embodies these two difficulties.It is necessary to both consider the win-win result with meeting obstacles,and recognize the true intention of the obstacle and cooperate.Considering factors such as safety,traffic regulations,comfort,efficiency,and followability,this paper presents a set of decision and planning algorithms based on high-precision maps and oriented to structured roads.The algorithm includes several modules such as lane-level decision,path decision,path planning,speed decision,and speed planning.It takes into account upstream factors such as perception,maps,and positioning,and downstream factors such as control modules.The final output is a smooth trajectory including spatial information and velocity information.The initial state of the result is the current state of the ego vehicle.This scheme can not only realize the detour of static obstacles,but also realize the avoidance of dynamic obstacles such as crossing and the same direction.For high-interaction problems in meeting scenarios,the algorithm also designs a special interaction planner.The thesis mainly includes the following parts:(1)Lane level decision.This module is implemented based on a finite state machine,and realizes macro decision by limiting the feasible domain of path planning.This module comprehensively considers safety and efficiency,so divides lane-changing scenarios into necessary lane-changing and non-essential lane-changing,realizing lanelevel decision that is not aggressive in general scenarios and not blindly conservative in critical scenarios.(2)Decision and planning of path and speed.The path decision module is implemented based on the graph search method,which compresses the spatial feasible region into a convex feasible region;the path planning module is realized based on the value optimization method,and finds an optimal path in this convex feasible region.In this paper,a mapping method is designed in the path decision module,which can quickly search for the optimal path at one time.In the path planning module,a mathematical expression of soft and hard constraints is designed,and the quadratic programming method is used to solve the problem.The speed decision module is implemented based on the enumeration evaluation method,and the speed planning module is also implemented based on the value optimization method.The method of the path planning module is partially reused to improve simplicity.The combination of path information and speed information is the final decision and planning result.(3)Decision and planning algorithm for interactive scenarios.The interaction planner considers the decision and planning of both the ego vehicle and the meeting obstacle to achieve a win-win situation,and can adjust the strategy according to the real intention of the obstacle.This module specifically includes trajectories generation,joint evaluation,online identification of the intention,and final decision selection.This solution will first try to actively change the behavior of the meeting obstacle in the interactive scene,and recognize the real intention of the meeting obstacle and take cooperative actions when the meeting obstacle does not respond to the behavior of the ego vehicle.(4)Simulation experiment verification of decision and planning algorithm.In order to verify the effectiveness of the algorithm,a Simulink-Prescan co-simulation platform is built in this paper,and the vehicle dynamics model of Carsim is used to make the simulation more realistic.The experimental results show that the method in this paper has good performance under the scene with both static and dynamic obstacles.