Automotive Software Development with Generative Pre-trained Transformer (DevGPT)

Project Description

With the increasing scope and complexity of software systems, the importance of programming tools is growing. Software development is being automated at all levels. This includes both the architecture and the generation of source code. Development should be as consistent as possible in line with the ideal of "correct-by-construction".

Artificial intelligence (AI) will further increase the degree of automation in the future. Model-based methods are often used to describe a wide variety of artifacts, e.g. requirements, architecture, implementation and testing. A textual representation is also possible in addition to a graphical representation (e.g. UML, SysML, AADL, AUTOSAR, UMLsec). This forms the basis for further analysis procedures.

Research Problem

The use of a large statistical language model based on generative pre-trained transformers (GPT) is particularly promising. However, such AI tools cannot currently be verified and the IT security of the generated code is not guaranteed. However, this is essential for the development of safety-critical applications such as automated or autonomous driving.

Development Goal

The open-source AI tool for automated software development "DevGPT" is being created. This forms the basis for the joint research focus "Secure software DevGPT 1 through artificial intelligence in software development" and enables research from three perspectives:

1. Generation of structured textual artifacts (e.g. architecture descriptions, security annotations of models and source code) for software development,
2. Verification of generated software architectures,
3. Verification of the implementation of security requirements, security review.

Project Partners

As part of this research project, existing research structures between the specialist areas

• „Spoken and Natural Language Understanding“ (Prof. Georges),
• „Modelbased Systems Engineering und Software Engineering“ (Prof. Kugele) and
• „Applied IT Security“ (Prof. Hof)

will be further expanded.