Software and System Reliability Engineering for Autonomous Systems Incorporating Machine Learning

Reliable Unmanned Autonomous Systems: Conceptual Framework for Warning Identification during Remote Operations

“…As the enablers of autonomy, the reliability engineering approach to ML algorithms is similar to traditional software reliability assessment. Abstractly, ML performs perception tasks and informed decision-making; thus, most systems that incorporate ML will naturally include standard software components [3]. Reliability growth modeling that characterizes how the reliability of a system increases during testing [3] is one of the standard approaches to software reliability assessment.…”

“…Abstractly, ML performs perception tasks and informed decision-making; thus, most systems that incorporate ML will naturally include standard software components [3]. Reliability growth modeling that characterizes how the reliability of a system increases during testing [3] is one of the standard approaches to software reliability assessment. In ML, the reliability growth measures the accuracy as a fraction of correct predictions divided by a total number of predictions [3].…”

“…Autonomy is "an unmanned system's own ability of integrated sensing, perceiving, analyzing, communicating, planning, decisionmaking, and acting/executing to achieve its goals" [2]. Recent research on autonomous systems identifies common challenges within risk and trust of ML algorithms that enable autonomy [3], [4]. Formal definitions of technical tests and evaluation of UAS [5] highlight challenges of lacking the quantitative definitions of emergent behavior, human trust, reliability and resilience [3].…”

“…Recent research on autonomous systems identifies common challenges within risk and trust of ML algorithms that enable autonomy [3], [4]. Formal definitions of technical tests and evaluation of UAS [5] highlight challenges of lacking the quantitative definitions of emergent behavior, human trust, reliability and resilience [3]. However, to ensure the UAS's ability to act and make decisions to achieve the mission's goal, it is critical to explore the concept of calibrated trust [6].…”

“…ML allows computing systems to learn how to do tasks from significant amounts of data, rather than being programmed (human instructed) [1]. Therefore, there is a rising need to understand how ML capabilities can be integrated into existing systems engineering, and design processes [3]. The performance of ML algorithms can measure the majority of UAS's capabilities, inevitably measuring the system's reliability.…”