Simulation and Statistical Model Checking of Logic-Based Multi-Agent System Models

Abstract

This thesis presents SALMA (Simulation and Analysis of Logic-Based MultiAgent Models), a new approach for simulation and statistical model checking of multi-agent system models.Statistical model checking is a relatively new branch of model-based approximative verification methods that help to overcome the well-known scalability problems of exact model checking. In contrast to existing solutions, SALMA specifies the mechanisms of the simulated system by means of logical axioms based upon the well-established situation calculus. Leveraging the resulting first-order logic structure of the system model, the simulation is coupled with a statistical model-checker that uses a first-order variant of time-bounded linear temporal logic (LTL) for describing properties. This is combined with a procedural and process-based language for describing agent behavior. Together, these parts create a very expressive framework for modeling and verification that allows direct fine-grained reasoning about the agents' interaction with each other and with their (physical) environment.SALMA extends the classical situation calculus and linear temporal logic (LTL) with means to address the specific requirements of multi-agent simulation models. In particular, cyber-physical domains are considered where the agents interact with their physical environment. Among other things, the thesis describes a generic situation calculus axiomatization that encompasses sensing and information transfer in multi agent systems, for instance sensor measurements or inter-agent messages. The proposed model explicitly accounts for real-time constraints and stochastic effects that are inevitable in cyber-physical systems.In order to make SALMA's statistical model checking facilities usable also for more complex problems, a mechanism for the efficient on-the-fly evaluation of first-order LTL properties was developed. In particular, the presented algorithm uses an interval-based representation of the formula evaluation state together with several other optimization techniques to avoid unnecessary computation.Altogether, the goal of this thesis was to create an approach for simulation and statistical model checking of multi-agent systems that builds upon well-proven logical and statistical foundations, but at the same time takes a pragmatic software engineering perspective that considers factors like usability, scalability, and extensibility. In fact, experience gained during several small to mid-sized experiments that are presented in this thesis suggest that the SALMA approach seems to be able to live up to these expectations.