Map-Distribution Mixture-of-Experts for Multi-Agent Pathfinding

Key points

• In contrast to prior studies focusing primarily on individual policy design, this study explores a meta-level Mixture-of-Experts (MoE) approach to learning-based MAPF.1
• Specifically, given a collection of compact yet high-performing experts specialised for different map distributions, our architecture learns a router that selectively fuses their outputs according to each agent's local situation.1
• We instantiate and validate this MoE approach with a commonly used graph neural network-based policy as well as a newly developed recursive action-token refinement model.1
• At the same time, our findings reveal that appropriately trained, parameter-matched monolithic models still outperform MoE, highlighting a nuanced role for map-distribution MoE for MAPF.1

Sources (1)

• [1]Map-Distribution Mixture-of-Experts for Multi-Agent Pathfinding
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 08:17 AM

  Designing capable neural policies for multi-agent pathfinding (MAPF), which map local observations to action distributions, represents an important challenge in multi-agent learning, with practical relevance to scalable collective automation problems such as warehouse logistics.

  In contrast to prior studies focusing primarily on individual policy design, this study explores a meta-level Mixture-of-Experts (MoE) approach to learning-based MAPF.

Extractive summary: sentences quoted from the sources.