Lightweight Parallel Multi-Agent Epistemic Planning

Abstract

We study a simple version of multi-agent epistemic planning where the number of parallel steps has to be minimized. We prove that this extension of classical planning is in PSPACE. We propose an encoding in PDDL and present some experiments providing evidence that this encoding allows us to solve practical problems. The types of problems we can encode include problems in which one agent can teach another agent how to perform a task and communication problems where some information must not be revealed to some agents.