Research On Modeling,Simulation And Verification Of Autonomous Driving Scenarios

The level of automation of autonomous vehicles continues to improve as autonomous driving technology advances,and more and more driving tasks are moved from the hands of human drivers to the Autonomous Driving System(ADS).The requirements for autonomous vehicle safety testing can no longer be addressed by testing procedures for traditional vehicles.The virtual test technique based on driving scenarios has considerable advantages in terms of efficiency and cost,and it will be the primary method for testing and verifying autonomous vehicles in the future.One of the critical challenges to be solved is how to flexibly express a specific driving scenario and execute the abstract driving scenario described in the virtual environment of an autonomous driving simulator or other verification tools.To address the aforementioned issues,this paper proposes SML4ADS(Scenario Modeling Language for Autonomous Driving System),a simple and practical DomainSpecific Modeling Language for Autonomous Driving Scenarios.On the basis of the modeling language,a graphical modeling tool supporting SML4 ADS is developed,providing modelers with convenience.In addition,a simulation method capable of executing the abstract SML4 ADS scenario model in the CARLA simulator and a verification method of the SML4 ADS scenario model based on UPPAAL SMC are implemented.The main work of this paper is as follows:1.This paper proposes a simple and practical domain-specific modeling language SML4 ADS for autonomous driving scenarios in order to flexibly represent diverse driving scenarios encountered by autonomous vehicles.First,based on the definition of the autonomous driving scenario term and domain knowledge of the autonomous driving,this paper analyzes the major content and modeling parts of the autonomous driving scenario.The abstract syntax,concrete syntax,and semantics of SML4 ADS are then designed using these crucial information and the modeldriven software development concept.Furthermore,in order to avoid text syntax duplication and complex structural difficulties,the SML4 ADS visual modeling tool is built,which offers modelers with ease and boosts modeling productivity.2.This paper designs and implements an approach of transforming the abstract scenario model of SML4 ADS into a concrete executable scenario in the autonomous driving simulator in order to actually carry out the virtual test based on the scenario.To demonstrate the generality of the SML4 ADS model simulation method,this paper first abstracts the execution process of the driving scenario in the simulator and transforms it into a specialized simulation process framework.Then,using the object-oriented thought characteristics of abstraction,inheritance,and polymorphism,and given that the autonomous driving simulator has a rich API interface,a general simulation technique of the SML4 ADS model is proposed.Finally,this paper provides an actual implementation of the SML4 ADS model simulation approach for the CARLA simulator.3.This paper designs the model transformation from the SML4 ADS model to the UPPAAL SMC model according to the semantics of SML4 ADS and implements the SML4 ADS scenario model verification method based on the UPPAAL SMC in order to verify and analyze the vehicle behavior in the SML4 ADS scenario model using the formal method.The model transformation is divided into two parts:transforming the SML4 ADS model data information into the data structure that can be stored in UPPAAL SMC and transforming the portion of the SML4 ADS model that describes the dynamic behavior of the vehicle into the network of automata in UPPAAL SMC.Finally,a direct connection is established between SML4 ADS and the UPPAAL SMC verification tool.