Research On The Rule Editing And Code Generation For The High-Level Decision System Of Unmanned Vehicles

With the development of information technology, the self-driving car becomes a hot topic. The decision subsystem is the core module of the self-driving car, it analyzes the environmental information and the information of the self-driving car itself, then give an operation command to the self-driving car. The decision subsystem plays a role as the command center. Because the road condition is very complex, the decision subsystem needs to compute a large number of signals and its code size is very huge. The maintainability and scalability of the system is so bad, and it is easy to lead to ill-considered.In order to solve the difficulties faced by software developers, we learn some mature software development methodologies in the field of the software engineering. And we propose a framework based on the technology of code generation and model checking. This framework can be used to develop the software of the decision system automatically. Due to the use of the code generation, the developers can focus on the design rather than the implementation. This technology reduces the workload of manual coding, improves the consistency of the code, and improves the maintainability of the software.On the basis of this framework, this paper focuses on the rule model and the code generation for the decision subsystem. We design an intermediate language used to describe the rules and design a algorithm to translate the intermediate language to C language. In order to prove the validity of the code conversion, the paper prove the semantic equivalence of the transcoding process.On the basis of the above methods, we design and implement a tool named UNMANNED_RULE_EDIT for the decision subsystem of the self-driving cars. This tool can be used for the rule modeling and the code generation. And in highway environment, we test this tool by using it to do the rule modeling and code generation for the process of the overtaking. Currently, the tool has been successfully applied in a self-driving car which is developed by a domestic team.