Mitigating Byzantine Failures in Multi-agent based Dependable and Adaptable Avionics Software

With increase in air traffic year on year and recent advancements and proliferation of unmanned aircraft systems, there is an increased drive and pressure to provide efficient and collaborative systems that leverage actionable information available on these ariel platforms. Adaptive systems forming the backbone of such platforms and adaptivity engineered through intelligent software agents is an attractive proposition as these ariel platforms operate in increasingly connected environments along with a number of other onboard systems collaboratively. This environment poses risks and a new set of problems wherein the adaptive system engineered with agents could fail, since some of these agents could be malicious. Byzantine faults form an important class of problems, contributing to these failures. Since the domain of avionics encompasses mission, safety, and flight critical systems, we provide one scheme that forms the basic block providing fault tolerance for byzantine failures caused by a single entity. A second scheme uses concepts from the signal-processing domain to over-come byzantine failures that involve multiple masquerading agents. We discuss the incorporation of these schemes in an exemplar adaptive flight planning system and provide reasoning for their correctness.