Research On Safety Evaluation Of Autonomous Vehicle For Freeway Scene

With the continuous promotion of autonomous driving technology,vehicle safety assessment has become the difficulty of advanced autonomous driving technology.At present,the single ADAS function of the existing L1-L2 level autonomous vehicle does not have the capability of continuous monitoring and evaluation of vehicle safety.In order to solve the problem that the vehicle safety status assessment method is limited by a single driver assistance system and the driving risk monitoring and evaluation is not comprehensive,this paper proposes a safety assessment method for autonomous vehicles oriented to the highway scene.Firstly,the components of highway scene are analyzed,and the safety evaluation framework of highway scene is proposed to study the impact of static scene,dynamic scene and vehicle safety risk on the driving safety of autonomous vehicles.In view of static scene evaluation,AHP is used to carry out weight design for the criterion layer and scheme layer of static scene,and calculate the static scene complexity of common scenes on highways,so as to provide selection basis for generating key scene data sets.Next,in the aspect of vehicle safety risk assessment,the evaluation index of vehicle longitudinal collision is introduced and the safe braking distance model is established to obtain vehicle driving state parameters,which makes preparation for the calculation of vehicle safety risk assessment and dynamic scene evaluation in the following paper.Then,the dynamic model of the vehicle on the curved slope road is established,and the safe driving speed of the vehicle is obtained considering the critical conditions of sideslip and rollover.Finally,the historical accident database is introduced,the traffic accident fault tree is established,and the accident probability information of different sections is obtained as the vehicle accident risk assessment coefficient.The vehicle risk degree is obtained by combining the safety speed and the self-drive speed as the vehicle risk assessment result,and the simulation verification is carried out in the design scenario.Finally,in the dynamic scene safety evaluation problem,the dynamic scene environment is divided into the scene where the main car goes straight and the scene where the main car changes lanes.The combination parameters of various dangerous operating conditions were preliminarily set in the simulation platform by using the PICT combination overlay tool to collect the required safety evaluation parameters and the state parameters of self and cycle vehicles.The pre-processed safety parameters are analyzed by Gaussian mixture cluster,and the scene classification is trained to obtain a scene safety class classifier.The designed classifier can be used to evaluate the safety class of vehicles in different scenarios.The Carsim and Simulink co-simulation platforms were used for verification and analysis,and compared with the clustering results,the risk identification accuracy of straight-line and lane change scenarios was 94% and 95.8%,respectively.The designed safety evaluation model is transplanted to Baidu Apollo platform as a single evaluate module to verify the applicability of the evaluation method.