A synergistic and extensible framework for multi-agent system verification

Recently there has been a proliferation of tools and languages for modeling multi-agent systems (MAS). Verification tools, correspondingly, have been developed to check properties of these systems. Most MAS verification tools, however, have their own input language and often specialize in one verification technology, or only support checking a specific type of property. In this work we present an extensible framework that leverages mainstream verification tools to successfully reason about various types of properties. We describe the verification of models specified in the Brahms agent modeling language to demonstrate the feasibility of our approach. We chose Brahms because it is used to model real instances of interactions between pilots, air-traffic controllers, and automated systems at NASA. Our framework takes as input a Brahms model along with a Java implementation of its semantics. We then use Java PathFinder to explore all possible behaviors of the model and, also, produce a generalized intermediate representation that encodes these behaviors. The intermediate representation is automatically transformed to the input language of mainstream model checkers, including PRISM, SPIN, and NuSMV allowing us to check different types of properties. We validate our approach on a model that contains key elements from the Air France Flight 447 accident.