LFC: Combining Autonomous Agents and Automated Planning in the Multi-Agent Programming Contest

TL;DR

This paper details the strategies and implementation of a multi-agent system that successfully competed in the 2019 Multi-Agent Programming Contest, combining autonomous agents and automated planning to address complex challenges.

Contribution

It introduces a novel approach integrating autonomous agents with automated planning using JaCaMo and Fast Downward for the first time in this contest setting.

Findings

Achieved first place in the 2019 contest

Developed strategies for map exploration without global coordinates

Effectively coordinated agents to assemble complex structures

Abstract

The 2019 Multi-Agent Programming Contest introduced a new scenario, Agents Assemble, where two teams of agents move around a 2D grid and compete to assemble complex block structures. In this paper, we describe the strategies used by our team that led us to achieve first place in the contest. Our strategies tackle some of the major challenges in the 2019 contest: how to explore and build a map when agents only have access to local vision and no global coordinates; how to move around the map efficiently even though there are dynamic events that can change the cells in the grid; and how to assemble and submit complex block structures given that the opposing team may try to sabotage us. To implement our strategies, we use the multi-agent systems development platform JaCaMo to program our agents and the Fast Downward planner to plan the movement of the agent in the grid. We also provide a brief discussion of our matches in the contest and give our analysis of how our team performed in each match.