Self-Adaptive Test Engineering in Embedded Intelligent Systems: A Machine Learning Approach

Abstract

The rapid development of autonomous intelligent systems, such as self-driving vehicles, smart healthcare devices, and IoT technologies, has introduced significant challenges in the domain of test engineering. Traditional testing approaches, such as manual testing and hardware-in-the-loop (HIL), often fail to accommodate the evolving and dynamic nature of these systems. In response to these limitations, this paper introduces a machine learning-based self-adaptive test engineering framework, aimed at improving test efficiency, coverage, and fault detection for embedded systems. The framework leverages reinforcement learning (RL) to dynamically generate and optimize test cases in real-time, adjusting based on system performance feedback. This approach allows for the detection of both known and unknown failure modes, ensuring a more comprehensive and adaptive testing process. Through the use of an autonomous vehicle simulation, the framework’s effectiveness is demonstrated, showing significant improvements in fault detection, test coverage, and overall resource efficiency compared to traditional methods. Moreover, the framework's real-time adaptability and scalability to larger, more complex systems are validated. While the current framework performs well for smaller systems, challenges related to generalization across different embedded systems and scalability for larger applications remain. Future work will explore the integration of zero-shot learning and transfer learning to address these challenges and enhance the framework’s ability to generalize across diverse autonomous systems. This paper presents a promising advancement toward a more adaptive, efficient, and scalable testing methodology for autonomous and embedded systems.