An autonomic model-driven architecture to support runtime adaptation in swarm behavior

Abstract

The use of unmanned vehicles in swarms requires significant runtime adaptation within the managing software due to the unpredictability of the environment it operates. This is compounded by rapid context changes occurring within the software elevating its operational complexity to a magnitude that renders them infeasible for humans to effectively manage. Our approach to addressing this challenge is model-driven self-adaptation using autonomic methods. This work extends and refines ongoing work on an unmanned vehicle swarm platform based on probabilistic finite state machines as behavioral runtime models and the formation of subswarms in the context of communication constrained search. We present the architecture of our work in progress as a reflection mechanism controlling short and long-term adaptive behavior. We realize short-term behavior change by the continuous transformation of structural models at runtime. To validate the architecture’s autonomic properties, we provide a walkthrough of an indicative scenario pertaining to swarm resilience as proof of principle of the architecture’s ability to dynamically replan under element failure.