Adaptive, Intelligent Control of Embedded Applications

The paper describes an adaptive intelligent controller architecture (GRAVA) for an image understanding system. The controller architecture is based on the concept of selfadaptive software, an approach that has seen significant effort in the past few years. Self-adaptive software involves synthesizing code that monitors, diagnoses and repairs behavioral deviations from explicitly stored and managed goal and method information. There have been several self-adaptive systems produced in recent years, but none of them has yet seriously addressed the issue of behavioral guarantees (including GRAVA). In addition to describing GRAVA, this paper proposes a new mechanism for providing assurance of stability in self-adaptive architectures. That mechanism is a process reflection layer to reason about the behavior of the program correction parts of the self-adaptive system. This mechanism will limit the scope in time and severity of deviations from nominal behavior, reduce or eliminate unstable corrections, and provide design time assurance of these behavioral characteristics.