Research On Autonomous Driving Simulation Technology Based On Driving Simulator

Autonomous driving technology is a future-oriented new transportation technology,which is concerned and actively promoted by all countries in the world.At present,the main research method of autonomous driving is to train and optimize vehicle model by collecting different road scene data using real vehicle test or simulation test,so as to solve the problems of poor vehicle safety and insufficient environmental adaptability.The real vehicle test has problems such as high cost,difficult safety,poor repeatability and so on,while simulation test can effectively avoid these problems.Although the simulation test can not completely replace the real vehicle test,it is still a necessary research method in the development of autonomous driving technology,such as Waymo’s cumulative simulatio n test mileage of more than 5 billion miles for training and optimizing vehicle models.A large number of simulation tests can circumvent some potential risks for real vehicle testing,which has become a widely used method in autonomous driving technology research.In the current various traffic simulation systems,the vehicle behavior models mostly simulate the characteristics of manual driving,and it is expected to reproduce the macro-micro laws of manual driving.There are some differences with the behavior of autonomous driving vehicles.Therefore,based on the UC-winRoad driving simulator,this paper builds a simulation platform with the main functions of autonomous driving simulating and model testing,and then studies the behavioral model of autonomous driving vehicles.The following three aspects of autonomous driving simulation technology research are mainly carried out:Firstly,building the autonomous driving simulation platform based on driving simulator.The UC-winRoad simulation software and SDK are used for secondary development to build autonomous driving simulation platform.The interactive interface is designed according to the research requirements,and the logic of autonomous driving control is determined.The vehicle information creation module and vehicle information acquisition module are developed to realize the initialization of simulated vehicle information and the perception of simulated traffic environment data.The vehicle motion control module and the vehicle behavior control module are further developed to realize the movement control of simulated vehicle,and the external model can flexibly control the simulated vehicle behavior through the interface.Secondly,researching on the autonomous driving model of adaptive driving simulator.According to the characteristics of driving simulator and the running demand of auto nomous driving vehicle,the autonomous driving car-following model and autonomous driving lane-changing model adapted to the driving simulator are constructed.For the car-following model,the behavior threshold is selected to divide the following driving state of vehicle,and the the autonomous car-following model based on multi-mode selection is constructed to realize the control of the vehicle’s following state.For the lane-changing model,the speed tolerance and space permissibility are introduced to constrain the generation of vehicle lane change.At the same time,the space required for vehicle lane-changing is determined.The lane-changing speed control model based on the two-lane front vehicle is constructed,and the trajectory optimization strategy based on the lane-changing space is formulated to realize the real-time feedback control of vehicle lane-changing.Finally,the analysis of model control effect based on autonomous driving simulation platform Through the fuzzy processing of holographic traffic informatio n,the running effect of the controlled vehicle in the autonomous driving simulation platfo rm under different info rmation detection errors and information delay conditions are tested.The simulation results show that the model constructed in this paper has certain adaptability to information error and information delay.However,the control effect of model is reduced when the information error and information delay are too large.