Master's thesis; Using AI to create test scenarios for automated mobility from accident data

Cooperative, connected and automated mobility (CCAM) is expected to improve traffic efficiency, accessibility and road safety. However, a challenge is the safety validation of vehicles equipped with automated driving or advanced driver assist features, due to the complexity and uncertainty of the driving environment. This thesis is about leveraging LLMs to interpret accident databases and create executable test scenarios for CCAM systems.

Background
Several European research projects are working on different aspects of safety validation for CCAM systems. One of the projects is Synergies (https://www.ri.se/en/what-we-do/projects/synergies-scenarios-for-development-of-automated-vehicles), which aims to create a European dataspace with datasets, tools and best practices for creating test scenarios for CCAM systems. Test scenarios will be in the form of machine-readable descriptions that can be replayed in different test environments such as simulators or different types of test facilities.

Description and Key Responsibilities
To create representative scenario datasets, data is collected from various sources such as recordings from test vehicles, drones or infrastructure. Another interesting source is existing traffic accident databases, which contain descriptions of situations that have previously led to accidents. However, these descriptions are typically in the form of textual descriptions in natural language or sketches. A challenge is thus converting this information to a machine-readable scenario format. Recent AI advances in the form of Large Language Models (LLM) and Multimodal LLMs (MLLM) may be a useful tool to help automate such conversion.

As part of this master thesis, we want to investigate:

• The feasibility and suitable tools for leveraging LLM/MLLM to interpret accident databases and convert scenarios to a machine-readable format (e.g. OpenScenario/OpenDrive).
• Developing a prototype tool infrastructure that can import accident data and leverage different LLM/MLLMs for interpretation and conversion.
• Testing the prototype on a set of accident records and evaluating its performance. Investigate options for improvements.

Qualifications
We are looking for one or two Master Thesis student. As a master's thesis candidate in this project, you will work with researchers from the Dependable Transport Systems and Human-Centered AI departments at RISE. We will provide you with the infrastructure and support to perform your thesis work.

Terms
This thesis is located in Borås and Gothenburg, physical presence is expected to some degree. Candidates are expected to be enrolled in a master's program in a field related to computer science and engineering. Start is beginning of 2025. The compensation will be 30 000 SEK upon completion of a high-quality thesis.

Welcome with your application!
If this sounds interesting and you want to know more, please contact Fredrik Warg, (fredrik.warg@ri.se) or Åsa Olsson, (asa.olsson@ri.se), Dependable Transport Systems Unit, RISE. Last day of application is November 6, 2024. Selection and interviews take place on an ongoing basis during and after the application period.