Classification and Description of AI Component Requirements in Autonomous Driving System

With the continuous advancement of Artificial Intelligence (AI) technologies, the demand for integrating AI components into system development has grown significantly, particularly in complex systems such as autonomous driving systems (ADS). However, current mainstream requirements engineering approaches still exhibit ambiguity in characterizing the functional properties of AI components, which considerably hinders development efficiency. In this paper, we propose a classification and description method for AI components in ADS, based on an extended problem frames approach. By introducing a structured framework and expanding the concept of symbolic phenomena, the proposed method enables a systematic representation of AI components' functional attributes, input/output semantics, and their roles within the system. According to their manifestation within the problem frames, AI components in ADS can be categorized into three categories: perception, directive, and interaction components. Preliminary case studies demonstrate that this method shows strong potential in supporting the development of clear and reusable requirement specifications for AI-driven systems.