Managing Online Self-adaptation in Real-Time Environments

Suppose we have an autonomous aircraft flying a complex mission broken into several different phases such as takeoff, ingress, target surveillance, egress, and landing. For each mission phase, the aircraft’s control system will have prepared a plan (or controller) specifying particular actions and reactions during the phase. Now suppose that the autonomous control system onboard this aircraft is self-adaptive: that is, it can modify its own behavior (plans) to improve its performance. Why might it have to adapt? Perhaps because the mission is changed in-flight, perhaps because some aircraft equipment fails or is damaged, perhaps because the weather does not cooperate, or perhaps because its original mission plans were formed quickly and were never optimized. In any case, the aircraft’s self-adaptive control system is now facing a deliberation scheduling problem. It must decide which mission phase’s plan to try to improve via self-adaptation, how to improve that plan, and how much time to spend on that self-adaptation process itself.