A New Approach to AD/ADAS Test Scenario Generation Using Open-Source Intelligence and Large Language Models

Abstract

In the evolving field of Autonomous Driving (AD) and Advanced Driver Assistance Systems (ADAS), the growing necessity for realistic simulation data coupled with the dynamic nature of real-world driving situations challenges traditional test scenario approaches. This is heightened by the increased complexity of AD/ADAS which necessitates rigorous testing to ensure safety and reliability.

This paper explores a new practical approach to AD/ADAS test scenario generation using Open-Source Intelligence (OSINT), generative Artificial Intelligence (AI) and Scenario Description Languages (SDL). This new approach utilises a vast array of openly accessible data on the Internet and holds potential for future research and applications.

The study was performed following the framework of Design Science Research (DSR). In two DSR cycles, an artefact that generates dynamic scenarios by scraping internet sources and completing the scenarios with Large Language Models (LLMs) has been developed.

The developed artefact successfully generated scenarios in an automated manner using collected traffic incident data, demonstrating its practicability for AD/ADAS test scenario generation.

In conclusion, the combination of OSINT with generative AI for scenario generation holds the potential to be a viable approach for test scenario generation for AD/ADAS systems, potentially extending the existing scenario generation methods with a novel approach based on openly accessible online data. We have shown that the method can generate scenarios using keywords from online texts and adhere to a given scenario format, yet future research is encouraged to explore further and validate this approach to ensure its applicability in the automotive industry.