Methodologies to better handle the evolving requirements of autonomous vehicle perception systems

Abstract

The development of autonomous vehicles (AV) has become an emerging field with various advanced technologies like Machine Learning (ML) and Artificial Intelligence (AI). However, the dynamic nature of the automotive industry, with constantly evolving requirements, presents significant challenges in developing these perception systems. This study investigates the changing functional and non-functional requirements of AV perception systems, focusing on the challenges and consequences these systems face in adapting to environmental changes, technological advancements, and regulatory demands. Through qualitative interviews with professionals from leading automotive companies, including system designers, solution architects, and embedded software engineers, the research explores how different development methodologies, particularly agile approaches, are crucial in addressing these evolving requirements. The findings reveal that the functional requirements in AV perception systems are evolving toward AI-enabled perception, real-time data processing, and advanced sensor fusion to enhance object detection, localization, and environmental understanding. Non-functional requirements such as safety, cybersecurity, and system reliability are becoming more complex due to increasing expectations and regulatory pressures. These evolving needs lead to significant challenges, including sensor uncertainties, higher development costs, and decision-making difficulties, which are being addressed through adaptive software development practices like Agile, SAFe, and hybrid approaches that support flexibility and rapid iteration. The insights gained from this research aim to improve the development processes for AV perception systems development and provide important suggestions and insights for engineers and researchers in the AV perception system. This work sets the stage for future work to explore hybrid development methodologies, real-time data processing optimization, and the potential of cutting-edge technologies, such as quantum computing, to overcome current limitations in AV perception and decision-making processes.