Research On Test Scenario Generation And Related Testing Methods For Connected And Automated Vehicle

As connected and automated vehicle technology continues to evolve,the ability of vehicles to communicate online and drive autonomously has been significantly improved.However,rigorous testing is necessary to ensure the safety and stability of the functions before the vehicles are mass-produced.Traditional realvehicle testing suffers from low efficiency and high costs,while simulation testing faces challenges such as low coverage of test scenario libraries,low efficiency in test scenario construction and imperfect test evaluation systems.To address these challenges,the digital twin testing approach driven by automatic generation of test scenarios for smart connected vehicles shows strong advantages.This approach uses test intent to drive the automatic generation of scenarios,and can generate a large number of test scenarios with practical significance.Based on the generated test scenarios,this paper proposes a digital twin testing method based on multi-device compatibility.Based on a scenario-driven testing approach,this paper proposes and validates a collaborative vehicle convergence model.The main research contents and innovations of this paper are as follows:(1)In response to the difficulty of simulation testing due to the high cost of test map acquisition,this paper proposes a static road library generation method based on hierarchical clustering.Using open source structured maps,a largescale map of the area of interest is obtained.A simple road library is extracted based on road feature information,and the acquired roads are classified.Also by intersection clustering method,simple intersections as well as complex road network structures can be obtained.Experiments prove that it can generate corresponding test roads,and the generated test roads will be utilized in Chapter3 and Chapter 4.(2)To overcome the shortcomings of slow manual construction of test scenarios and low coverage of traditional automatic methods,this paper proposes a functional test scenario library generation method based on a partially observable Markov decision process,which takes advantage of the Markov decision process to model test intentions and combines the test scenario generation with the solution process to generate the corresponding test scenario libraries.Through experiments,it is proved that the relevant test scenario library can be generated by using this method,and there is an 83.9% improvement in scenario coverage.(3)In order to solve the problems of low efficiency of real-vehicle testing,poor correlation of simulation testing,and poor compatibility of traditional inthe-loop testing,based on the test scenario library automatically generated by the automatic method of test scenarios,this paper proposes a scenario-driven digital twin testing method.At the same time,this paper summarises the problems of the current test and the corresponding solutions.The experiments combined with forward collision warning test and intersection collision warning test prove that the test method has good stability and compatibility,and has been delivered to a number of enterprises for use.(4)To address the problem of low efficiency of conventional sensors at ramp junctions,this paper proposes a collaborative multi-agent reinforcement learningbased vehicle rendezvous approach combined with vehicle networked communication.The vehicle merging problem is constructed as a Markov decision process,with the vehicle intent and the initial state as inputs,and the model gives the suggested vehicle driving behavior.After model-in-the-loop testing as well as hardware-in-the-loop testing,it is demonstrated that the method is able to achieve safe vehicle convergence.This method enables a higher level of safety in the same amount of time as compared to conventional methods.