Run-time Reasoning from Uncertain Observations with Subjective Logic in Multi-Agent Self-Adaptive Cyber-Physical Systems

Modern society has become increasingly reliant on the omnipresent cyber-physical systems (CPSs), making it paramount that the contemporary autonomous and decentralized CPSs (e.g., robots, drones and self-driving cars) act reliably despite their exposure to a variety of run-time uncertainties. The sources of uncertainties could be internal, i. e., originating from the systems themselves, or external-unpredictable environments. Self-adaptive CPSs (SACPSs) modify their behavior or structure at run-time in response to the uncertainties mentioned above. The adaptation relies on gained knowledge from the observations that the SACPSs make during their operation. As a result, to build the knowledge, the need for run-time observations aggregation and reasoning emerges since the observations made by decentralized CPSs are uncertain, partial, and potentially conflicting. In response, in this paper, we propose a novel methodological approach for deriving or aggregating knowledge from uncertain observations in SACPSs utilizing the Subjective Logic. The effectiveness of the proposed approach is demonstrated through extensive evaluation on an in-house, multi-agent system from the robotics domain.