Abstract

The continued deployment of systems in settings that are unpredictable, rapidly changing, and increasingly removed from human access drives the development of self-adaptive software: these are software systems that are capable of changing the way they behave in response to changes in their environment. The development of this class of software, however, is made challenging by difficulties in designing self-adaptive behavior coherently and consistently, accounting for adaptive needs, and perceiving adaptations as they take place.

Our approach to addressing these challenges is architecture-based: the adaptation process is rooted in explicit architectural models and adaptation policies that model and implement changes through architectural means. This approach to policy-based architectural adaptation management (PBAAM) addresses the challenges of developing self-adaptive systems and improves on other architecture-based approaches. The key features of the approach are: using architectural models to provide a consistent foundation, adopting adaptation policies that can be modified at runtime, and providing explicit support for the recording and visualization of adaptations.

Evaluation of the approach and our claims is conducted through case studies focused on the robotics domain. In both simulated as well as physical robotic platforms, we show how our approach is feasible and beneficial to component reuse and expressiveness while not imposing a prohibitive computational overhead. The contributions of this research include: an organizational framework useful in analyzing the current state of the art in architecture-based self-adaptive software, an architectural style for robotic systems that fosters adaptability, a specific approach to building architecture-based self-adaptive systems supported by extensible notations and tools, and a demonstration of novel self-adaptive capabilities in our case study domains.