Towards High-Level Programming for Distributed Problem Solving

We present a new approach to distributed problem solving based on high-level program execution. While this technique has proven itself for single-agent systems based on the Golog language, several challenges are encountered when moving to a multi-agent setting. Key to our approach is a better representation of the dynamics of multi-agent teams by means of the following features: a robust combination of true concurrency of actions with the interleaved concurrency of ConGolog; an explicit notion of time to assist coordination; and semantic support for predictable exogenous actions (also called "natural actions"). The result is MIndiGolog, a new Golog variant suitable for distributed, cooperative execution by a multi-agent team.