Research On Key Technologies Of Lane Change Decision And Planning For Autonomous Vehicles

Autonomous vehicles are the future development direction of the automotive field and the edge of scientific and technological innovation.Automatic lane change is one of the most complex autonomous driving tasks,and its performance represents the intelligence level of autonomous vehicles.It consists of two core issues including automatic lane change decisions and path planning.At present,the lever lane change assist driving function only has the ability of change lanes path planning,which requires the driver to trigger and monitor safety,the application scenario is limited,the user experience is poor;high-level autonomous vehicles have many accidents in the automatic lane change test process,and have not yet been able to enter the practical stage.It is obvious that the automatic lane change function is still in the initial stage,which is one of the major problems in the current field of automatic driving.This paper examines the two core issues of automatic lane change,and the specific work is as follows:(1)A reinforcement learning lane change decision algorithm based on fuzzy reasoning is proposed,which uses a multi-objective reinforcement algorithm to construct a state space using direct parameters such as the position and speed of the front car.The proposed algorithm describes uncertainty in a probabilistic manner,continuously learns from the observed system states so that it can improve the robustness of reinforcement learning lane-changing algorithms for autonomous vehicles in an uncertain dynamic interaction environment.Since the automatic lane change process has a continuous state space,traditional multi-objective reinforcement learning algorithms face the challenge of dimensional disasters.To resolve this issue,fuzzy reasoning is leveraged in the time difference learning algorithm to dispose of successive conditions and motions,so that the self-learning process can be promoted.Verification results show that the proposed lane change decision algorithm which is based on fuzzy reasoning is comprehensively improved in terms of collision rate,average lane change maneuver time,and performance value,improves the selflearning ability of multi-objective decision-making algorithms according to actual environmental changes,expands the application scenarios of automatic lane change,and improves the intelligent level of automatic lane change function compared with the algorithm which does not use fuzzy reasoning.(2)A real-time planning and tracking strategy is proposed which is based on scenario.The lane change trajectory is planned in real-time according to the state of motion of the vehicle in front and the remaining lane change time at each moment so that the safety of the automatic lane change process is improved at lower computing power and economic costs.Specifically,the classical five-spline mathematical model is used to describes the lane change trajectory and optimizes the solution;coordinate conversion is used to extend the automatic lane change function to the application of curved dojo scenes.The lane change process scenario is divided into normal lane change and dangerous lane change.In view of the two most dangerous scenarios in the process of automatic lane change,which are the car in front of this lane suddenly invading the target lane and the car in front of the target lane suddenly slowing down,a dual redundant safety decision tree and multi-mode switching control strategy with distance and acceleration are designed to achieve optimal control in dangerous scenarios and improve the safety of the lane change process.The control strategy is simulated and tested,and the results show that the lateral acceleration and swing angle velocity of the vehicle in the two dangerous lane change scenarios have good control performance.(3)The automatic lane change simulation test system is constructed,which first analyzes and studies the expected functional safety of the automatic lane change system and the decision planning function of the autonomous vehicle,identifies the requirements of the automatic lane change simulation test system,and proposes a scenario-based virtual simulation and parallel test method.Then,an automatic lane change simulation test system consisting of softwares and hardwares such as a simulation bench composed of real vehicle actuators,a high-precision vehicle dynamic model,an automatic lane change scenario library and virtual simulation software,and a parallel in-the-loop test environment based on 5G is constructed,which provides a basic support platform and service for automatic lane change decision-making and path planning research,carries out simulation test research on decision-making algorithms and lane change path planning strategies,improves research efficiency,reduces the risk of real vehicle road testing,improves safety,and accelerates the smooth development of various research work.(4)Finally,the research results are applied to the testing and verification of autonomous vehicles,and a large number of vehicle testing and verification studies are carried out from three dimensions:parallel testing of closed test sites,open road tests of demonstration areas and i-VISTA autonomous vehicle challenges.The results show that the evaluation indicators such as normal operation of the automatic lane change system,lane change time,longitudinal acceleration,lateral acceleration,and safety all meet the requirements of vehicle design constraints,the automatic lane change decision-making algorithm and path planning strategy studied perform well on the whole vehicle,which lays the foundation for the automatic lane change function to be practical,and makes positive contributions to the development of the industry.