Virtual Traffic Scene Of Intelligent Behavior Validation Platform For Driverless Vehicle

Driverless vehicle integrates environment perception, decision-making programming and control, which relates to many subjects such as kinematics and dynamics, automatic control, sensor technology, pattern recognition, artificial intelligence, and incarnates the latest research results of computer and robot technologies. Development of driverless vehicle requires lots of test conditions to support, however, general vehicle test can not satisfy requirements of driverless vehicle validation. Firstly, safety problems of driverless vehicle are not validated effectively. Secondly, validation of driverless vehicle needs very slashing road conditions, so it costs much.Driverless vehicle drives automatically, so its security and reliability are very important, whose test had better be carried out under controllable, repeatable, effective and safe conditions. Therefore, developing safe and effective experiment methods of driverless vehicle and constructing economical, high robustness experiment environments are very important. Platform for intelligent behavior validation of driverless vehicle is a hardware-in-the-loop experiment platform that we construct, aiming to validate intelligent behavior of driverless vehicle with economical, safe and effective way, and virtual traffic scene is an important component of it. This dissertation surrounds with intelligent behavior validation of driverless vehicle, and studies construction of virtual traffic scene and behavior evaluation methods. Contents mainly contain following aspects:Firstly, construction methods of virtual traffic scene was researched, and general design project of virtual traffic scene of intelligent behavior validation platform for driverless vehicle was proposed. At present virtual traffic scene is mainly used in driving simulator, which includes training and developing types. Regardless of which type, they are all"human-vehicle-environment"closed loop systems. However, virtual traffic scene in intelligent behavior validation platform for driverless vehicle is a"vehicle-environment"closed loop system, and both have essential differences, so we need to study construction methods of virtual traffic scene suitable for intelligent behavior validation of driverless vehicle. With development of virtual environment, artificial intelligence, and artificial life, we can construct virtual traffic scene being provided with definite intelligence. On the one hand, validated vehicle can apperceive virtual environment for autonomous driving, on the other hand, test models in virtual traffic scene also can apperceive intelligent behavior of validated vehicle, which can provide original behavior data for evaluation, forming"vehicle-environment"closed loop intelligent system, and then provides a novel method for solving problems of intelligent behavior validation.Secondly, designing principles and implementation methods of static test models in virtual scene were studied. It mainly researches construction methods of road agent, rail agent, traffic light agent, and sign agent, and designs different road models such as"U turn","T type"road, and"S type"navigation path. To improve robustness of experiment validation, we design and implement sunny, rainy, fogy, and snowy natural environment models. These static test models provide test environment and behavior data for intelligent behavior validation of driverless vehicle.Thirdly, construction methods of dynamic test models for intelligent behavior validation of driverless vehicle were studied. Intelligent behavior of driverless vehicle mainly reflects in confronting with paroxysmal and uncertain events. Researching construction of dynamic test models is to design different types of paroxysmal and uncertain events, which creates test conditions for validation. In this chapter, follow vehicle agent model, its behavior modeling and simulation were studied, and some paroxysmal scenario models were implemented. At last construction method of behavior management agent was proposed.Finally, evaluation method of intelligent behavior was studied. Based on behavior data that validated vehicle has occurred with designed static test models and dynamic test models, a set of validation and evaluation rules were designed, and a evaluation method for intelligent behavior of driverless vehicle was presented.