Multi-scenario Trajectory Generation And Optimization For Autonomous Driving

With the rapid development of the automobile industry and the government support,electrification and intelligence are two important development directions of the automobile industry at present.Meanwhile,research institutes and enterprises are vigorously conducting research.The technical level of autonomous driving is divided into four main parts:perception,map,PNC and operating system.The planning module in PNC part is vividly called the core brain module of autonomous driving,which integrates the information of upstream perception and high definition map.The prior information is used to decide the future driving behavior of the vehicle and calculate the future trajectory information of the ego vehicle.The quality of the trajectory generation will directly affect its performance on the real world.The saying"Safety does not mean cowardice"points out the two core indicators of trajectory planning.The first is safety.Accidents should not happen.The second is to be intelligent enough,not blindly conservative.If the self-driving vehicle is too conservative,it will affect the driving conditions of other vehicles on the road and slow down the traffic speed of the entire road.In this paper,several common scenarios are studied,and a complete set of decision-making,mission and local planning algorithms are proposed to generate trajectories.The thesis mainly includes the following parts:(1)Mission Planning and Decision-Making.First,the high definition map format required in the entire trajectory planning is introduced,which provides necessary road information for the subsequent mission planning and local planning.the~*algorithm is introduced,and used to get the mission path information from the starting point to the ending point in the mission planning.The global path will provide essential reference information for the subsequent local planning.The ego car needs to make different driving behaviors in different scenarios,for example,lane keeping,lane change,nudge,etc.This paper analyzes the differences between the above driving behaviors and gives the transition conditions for above behaviors.(2)Trajectory Generation and Optimization in Urban Roads.In order to eliminate the influence of road curvature on the trajectory planning algorithm in the case of urban roads with guide lines,the Frenet Frame is introduced and its advantages are pointed out,and then the relationship between the Frenet Frame and the Cartesian Frame is deduced in detail.In order to improve the passenger’s ride comfort,it is proved that the trajectory of the fifth-order polynomial can minimize jerk of the host vehicle.Then,the target position information under lane keeping and lane changing conditions is determined,combined with the initial position information of the vehicle and the method of fitting the trajectory with a quintic polynomial,a minimum jerk trajectory can be given.Then,for the nudge behavior,the initial trajectory is generated by offsetting several important trajectory points.After that,the quadratic optimization method is used to optimize the trajectory generated under various working conditions before.The optimization can make the trajectory of the ego vehicle meet more constraints,such as speed and acceleration limits.(3)Trajectory Generation and Optimization in Multi-Scenarios.A complex scene refers to a structured road and an unstructured road in the current environment of the ego vehicle,and the structured road may have lane keeping,overtaking,nudge,etc.,and a complex scene of one or more of the above scenarios.This part will introduce the trajectory planning algorithm under the public space.We use the(74)(9~*algorithm to generate an initial trajectory,and then use the quadratic optimization method to minimize the total acceleration and length of the trajectory of the vehicle during driving.Finally,the method of switching vehicles between structured roads and public space will be introduced.This part completes the closed loop of trajectory generation of the entire complex scene,which can realize any point-to-point autonomous driving of the ego vehicle.(4)Verification of Trajectory Planning Algorithm based in Simulation Environment.In order to verify the feasibility of the trajectory algorithm,this paper uses the Ubuntu environment,Autoware proposal platform,and C++to complete the writing of the core code,and verify the effectiveness of the algorithm on the basis of ROS as a communication tool.Make the scenario editor write common simulation scenarios into the yaml format file,which can greatly facilitate the comparison of the performance relationship of algorithm parameter adjustment.we also use rviz or python for display.Finally,several experimental scenarios are tested to verify the switching capability of multiple working conditions,and the functions such as lane changing and lane keeping are verified in the high-speed structured road scene.