Acquisition of Cooperative Behavior in Multi-Agent Path Finding

Haruto sugawara, Hiroyuki Toda

Keywords: Multi-agent path finding, Reinforcement learning, Curriculum learning

TL;DR: We improve cooperative behavior in multi-agent pathfinding by modulating how much each agent’s reward is influenced by others.

Abstract:

The Multi-Agent Path Finding (MAPF) problem involves planning collision-free paths for multiple agents traversing from initial to designated positions.

Reinforcement learningbased approaches have recently gained attention, demonstrating effective path planning in complex environments under decentralized control.

However, these methods encounter a fundamental limitation: individual reward maximization by each agent results in inter-agent interference, degrading performance.

This research addresses reward design in reinforcement learning-based MAPF to facilitate cooperative behavior. Our key idea is to incorporate other agents’ reward influence into individual reward functions and systematically modulate this influence to enhance cooperation acquisition. Through comparative evaluation against existing methodologies, we demonstrate our approach achieves improved performance.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 10