Hybrid system reachability-based analysis of dynamical agents

This paper describes a hybrid dynamical system-based approach to formalizing and mechanizing analyses of dynamical agents, i.e., situated, embodied actors that continuously respond to their environment. As an example, the paper describes a class of formalized metrics for reasoning about the relative difficulties of agent navigation in various environments -not just whether one scenario is more difficult than another, but how much more difficult a scenario might be- and presents results of relative difficulty reasoning using a specific example metric. This illustrates that qualitative or heuristic agent properties, which are commonly unformalized and imprecise, may be formalized and rigorously analyzed using this approach. The paper also discusses the potential implementation of relative difficulty metrics in meta-intelligent agents.