By 2025, driverless cars will be an integral part of daily transportation. Understanding the reliability of self-driving cars is a crucial step to ensuring that the impending ubiquity of self-driving cars causes as few fatalities as possible. Components like actuators, sensors, and computational elements that make up such systems have inherent vulnerabilities to faults due to manufacturing defects, aging, cyberattacks, and environmental factors. Repair and replacement of such components may reduce the risk of fault occurrences, but may be infeasible in terms of cost, safety, and availability. Alternately, certain faults and their false positives may trigger unnecessary repair or cause unnecessary reactions by the vehicle. Therefore, it is necessary to quickly and accurately identify faults in real time. This research will facilitate the development of in-the-field error mitigation techniques, resulting in more reliable autonomous cars. Furthermore, this research will support the technical development and engagement of an underrepresented cohort of graduate and undergraduate students at North Carolina A&T State University and North Carolina Central University through curriculum enhancements and participation in extracurricular activities such as the AutoDrive Challenge, a national self-driving car competition.
The proposed work will provide real-time diagnosis of transient, intermittent, and permanent faults that occur in a self-driving car. This analysis will substantially improve the performance and accuracy of fault classification/identification in complex systems. Multi-perspective error detection techniques, including discrete-event system analysis, data-driven analysis, and chip-level analysis, will be combined to diagnose faults in automotive systems. The discrete-event system analysis will detect and isolate a system's fault occurrences from external observation of general behaviors of the system and in the absence of full observation of occurred events. The data-driven analysis will use a novel fuzzy type-2 clustering-based method to detect whether a fault degraded performance. The chip-level analysis will detect when a computational component is malfunctioning based on equivalence checking of logic signals and state traces. The combination of these approaches will facilitate fault diagnosis of automotive systems in real-time and with greater accuracy and speed. The multi-perspective analysis will improve the understanding of how each perspective interacts with the other and has the potential to identify new fault types and patterns. The enhanced awareness created by integrating these three unique methods will facilitate automotive system fault diagnosis in real time with greater accuracy and speed than could be achieved by any of the methods individually.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
