Minimizing the Oracle Problem for Self-Adaptive Cyber-Physical Systems

Cyber-physical systems (CPS), particularly those within the safety-critical domain, must continually ensure that their requirements are being satisfied. To support online requirements satisfaction, a CPS can be modeled with self-adaptive characteristics (SA-CPS) to enable the system to self-reconfigure as uncertainty manifests in both the environment and the system itself. For instance, unexpected weather conditions, misunderstood requirements, or system hardware failures may lead to observed behavioral faults and/or requirements violations, thereby necessitating a change of system state to mitigate the problem. However, an SA-CPS may adapt into an unknown state to resolve such uncertainties, where the new system state may not have been properly or fully validated or verified at design time. Therefore, additional techniques are necessary to enhance run-time assurance.

Specifically, we aim to introduce adaptive capabilities into the test oracles that guide run-time validation of SA-CPSs. An oracle can guide validation activities by providing feedback on system output with respect to encountered inputs. However, such oracles are often derived at design time where not all system states are known. Therefore, this project will investigate (1) how oracles can adapt alongside the SA-CPS adaptation engine, (2) enable run-time verification of oracles, and (3) maintain and update traceability links between oracles and system artifacts at run time. To validate these research questions, we are developing the Cognitive Assisted Living (CAL) framework, a real-world experimental testbed, modeled as an SA-CPS, for providing long-term support of early-stage Alzheimer's patients, their caregivers, and their family members.