Research And Implementation Of Drum-based Intelligent-Vehicle-in-the-Loop Accelerated Evaluation

While new technologies empower intelligent vehicles,they also bring new challenges to traditional automotive test tools and test methods.At present,the test tools used for intelligent vehicle testing encounter the problems that single performance cannot satisfy complex test requirements and there is only a small proportion of extreme scenarios in test acceleration.In order to improve the testing capability of the traditional test program and ensure that the intelligent vehicles can be rigorously and fully tested before the real road test,this paper proposes an intelligent-vehicle-in-the-loop acceleration test program based on the drum platform.The proposed test tool can realize the multi-source information simulation of the perception layer of the tested vehicle,the dynamic realization of the road model,and the following of the vehicle steering behavior.Based on the NGSIM data set,a combined carfollowing model is built to test the scenarios generated by the cross-entropy important sampling method acceleration.The research contents mainly include:An intelligent-vehicle-in-the-loop test framework composed of road simulation subsystem and signal simulation subsystem is designed.The actual perception data of the signal simulation subsystem is mainly generated by the millimeter wave radar black box and the video black box,and the virtual simulation data is mainly obtained from the sensor module of the scene software.The road simulation subsystem can control the servo motor inside the drum platform structure through PID algorithm and realize the simulation of road resistance,roll angle and pitch angle and steering following in the test scenarios.The fuzzy sliding mode control algorithm of the steering-following part realizes the following and status monitoring of the steering behavior according to the measured vehicle tire angle and steering wheel angle.The experimental results show that the steering-following system can follow the intelligent vehicles’ steering behavior accurately and steadily.The IDM-SVR car-following model is constructed based on theoretical driving and data driving models.The data used to build the model are all car-following data selected from the NGSIM data set.The parameters of the IDM model are calibrated by simulated annealing algorithm,and the parameters of the SVR model are optimized in conjunction with the particle swarm optimization algorithm.The weighting coefficient of the model is optimized to generate the IDM-SVR car-following model that can reflect the theoretical attributes and the data attributes.The speed and acceleration fluctuations of the combined model have been significantly improved.Based on the MCMC random sampling test scenario,the cross-entropy important sampling method is used to accelerate the generation of extreme scenarios in the test,which solves the problem of low generation rate of extreme scenarios.During the experiments,the upper computer software of the test system can realize the parameter setting of the test tool,and the millimeter-wave radar dark box is used to simulate the test scenario signal.The combined model is used to evaluate the generated test scenarios.The results show that the accident rate of the improved accelerated sampling scheme is significantly faster than that of the actual road,which can effectively promote the efficiency of the accelerated testing.