 |
Learning to Solve Stochastic Multi-Agent Path Finding
1 : Centre d'Énseignement et de Recherche en Mathématiques et Calcul Scientifique
(CERMICS)
Ecole des Ponts ParisTech
In large railway networks, real-time traffic management is essential to minimize disruptions and maximize punctuality. We propose a novel approach to tackle the Multi-Agent Path Finding problem, using the AIcrowd Flatland challenge as a testing ground. By leveraging machine learning inside simple combinatorial procedures such as prioritized planning, we provide a principled way to make better heuristic decisions and anticipate delay propagation.
