Decomposition of Multi-Agent Planning under Distributed Motion and Task LTL Specifications

TL;DR

This paper presents a two-phase automata-based approach for efficient multi-agent planning under complex temporal logic specifications, decoupling local behavior constraints from collaborative task requirements to reduce computational complexity.

Contribution

It introduces a novel decentralized planning method that separates local and collaborative specifications, enabling scalable multi-agent planning with guaranteed satisfaction.

Findings

Reduces computational complexity of multi-agent planning

Guarantees satisfaction of combined specifications

Demonstrates practical benefits through an illustrative example

Abstract

The aim of this work is to introduce an efficient procedure for discrete multi-agent planning under local complex temporal logic behavior specifications. While the first part of an agent's behavior specification constraints the agent's trace and is independent, the second part of the specification expresses the agent's tasks in terms of the services to be provided along the trace and may impose requests for the other agents' collaborations. To fight the extreme computational complexity of centralized multi-agent planning, we propose a two-phase automata-based solution, where we systematically decouple the planning procedure for the two types of specifications. At first, we only consider the former specifications in a fully decentralized way and we compactly represent each agents' admissible traces by abstracting away the states that are insignificant for the satisfaction of their latter specifications. Second, the synchronized planning procedure uses only the compact representations. The satisfaction of the overall specification is guaranteed by construction for each agent. An illustrative example demonstrating the practical benefits of the solution is included.