Agentic AI for Software-Defined Vehicles: A Generative Testing Framework for Autonomous Feature Assurance

Abstract

SDVs get advanced as a result of applying artificial intelligence (AI), yet they become increasingly difficult to test. The conventional techniques such as the scripted examinations or random scenarios are not sufficient to test all of the likelihood of driving that may contain peculiar and hazardous elements. The present research paper claims to introduce a novel conceptualization using agentic AI and generative AI (with no contact to real-life phenomena) to experiment in SDV attributes more effectively. The agentic system of AI will be autonomous that is, they plan, learn and innovate overtime. The Generative AI assists in supporting millions of test cases, unusual road dynamics, driver behavior and cyber-attacks. It experimented with its framework using modified systems of driver-assistance (ADAS) and the experiment was very successful. In comparison, agentic AI contains three times more opportunities to go off the rails relative to scripted tests and exceeding a regulation failure relative to what random tests. It also enabled the vehicles to be competent to pass safety standards like FMVSS and the ISO 26262 standard as risks may be noticed promptly. The model minimized the use of the network without the report of mines to the testing speed and coverage which was enabled with a combination of edge resources and cloud-based resources. This was what was called continuous learning as the system continued to improve as more vehicles provided the data. The result has already been proven that AI-based testing is scalable, viable and applicable to use in complex conditions. According to the study, agentic AI could greatly contribute to smarter and safer SDV testing that would give engineers the opportunity to identify and eliminate issues extremely quickly. Such an approach will go toward the adoption of faster and have more confidence in self-driving cars.