Predictive and Prescriptive Analyses of Autonomy Integration into the System of Systems

Stakeholders of existing system of systems (SoSs) are interested in integrating autonomous engineered systems to improve capabilities while minimizing the need for human workload. Such integration leads to a system of autonomous systems (SoAS) that introduces new systems engineering challenges in terms of integration, testing, and evaluation. Autonomous systems have different levels of autonomy (LoAs), depending on their AI technology and the autonomous capabilities they provide. One challenge is to decide on the suitable LoAs for SoAS. Furthermore, uncertainty in AI algorithms’ outputs may lead to undesirable emergent behaviors for SoAS. Therefore, another challenge is analyzing emergent behaviors, identifying root causes, and preventing them in the future. To address these challenges, this paper provides a methodology that employs Bayesian networks and machine learning to conduct a stochastic trade study analysis of SoAS with different LoAs. This methodology provides the following: (1) predictive analysis which is a bottom-up approach to predicting potential undesirable emergent behaviors by examining various scenarios and (2) prescriptive analysis which is a top-down approach to identifying root causes of a possible undesirable emergent behavior. We present a hypothetical example of a search-and-rescue SoS to demonstrate the implementation and effectiveness of the proposed methodology in evaluating the integration of various LoAs into an existing SoS.