Modelling Autonomous Driving Systems:A Data-Driven And Model-Driven Collaboration Approach

With the rapid development of artificial intelligence,more and more Cyber-Physical Systems are using machine learning（ML）to accomplish specific tasks,and such systems are also called intelligent systems,such as Autonomous Driving System（ADS）.Considering the characteristics of intelligent systems,on the one hand,the development of the system has a high demand for the quality of high-quality data,and on the other hand,it is necessary to model the properties of artificial intelligence and neural network models in the system model.Traditional systems engineering is no longer sufficient the need for the modeling of intelligent systems,so modeling of intelligent systems has become a research hotspot in academia and industry,but since the research is still in its infancy,there are not yet complete theories and techniques to support the modeling of intelligent systems.So,this paper proposes key technologies for data-driven and model-driven collaborative modeling of intelligent systems.Firstly,based on the Meta-Object Facility（MOF）Meta Modeling standard,a spatio-temporal trajectory data（STTD）Meta Modeling technology system is proposed.Then,the intelligent system modeling language Sys ML4 AI is designed and implemented by making extensions to the Sys ML.Finally,the practicality and easy scalability of this paper’s approach are demonstrated by studying the modeling of Scenario Recognition Intelligent Component,which is based on spatiotemporal trajectory data-driven and model-driven collaborative.The main work points of this paper are as follows.1.In order to obtain high-quality domain data,this paper proposes a MOF-based STTD Meta Modeling technology,to guide data modeling and tool development.The spatio-temporal trajectory Meta Data structure is used to define the components of the data,the twelve-dimensional spatio-temporal trajectory data MetaModel is defined to represent the domain knowledge of the data,and the data quality management is designed.Finally,spatio-temporal trajectory data pre-processing tools are introduced to support the implementation of the above techniques.2.In order to support the modeling of intelligent systems,this paper proposes the Sys ML4 AI modeling language based on extending Sys ML.Firstly,we implement the extension of the Block Definition Diagram（BDD）from two perspectives of Steretype and Meta Model,add new Steretype to model the types of intelligent component,and add AIComponent in the Meta Model to model the artificial intelligence information in intelligent component.Then,add new port types in the Meta Model of the Internal Block Diagram（IBD）to support the modeling of complex internal structure.Finally,we add a new structure diagram in the Sys ML4 AI language: Artificial Neural Networks Diagram（ANN）,and a detailed Meta Model of ANN is given to support modeling the structure of neural networks.We also design the microservices-based intelligent systems architecture,and intelligent component combination method.Also,the Sys ML4 AI modeling tool is developed.3.In order to achieve data-driven and model-driven collaborative modeling of ADS,the spatio-temporal trajectory data-driven and model-driven collaborative closedloop model of ADS are designed,and how to model the intelligent component of the intelligent system is studied with the scenario recognition case.Firstly,we use Carla simulator to generate spatio-temporal trajectory data Set and get high quality spatio-temporal trajectory data by data pre-processing technique.Then,we design and implement the autonomous driving scenario recognition algorithm and neural network model construction.Finally,we use Sys ML4 AI language to implement the BDD modeling,IBD modeling and ANN modeling of scenario recognition.For the modeling of neural networks,it is based on the results of data-driven ML model training,which leads to collaborative modeling.