Local Optimization of MAPF solutions on Directed Graphs

TL;DR

This paper introduces an iterative local search method for multi-agent pathfinding on directed graphs, improving suboptimal solutions' quality efficiently by leveraging local dynamic programming, thus balancing feasibility and optimality.

Contribution

It presents a novel local search approach using dynamic programming to enhance rule-based MAPF solutions on directed graphs, reducing computational complexity.

Findings

Significantly improves solution quality over initial feasible solutions.

Reduces time complexity from exponential to polynomial with locality constraints.

Provides a practical method for better MAPF solutions in crowded scenarios.

Abstract

Among sub-optimal MAPF solvers, rule-based algorithms are particularly appealing since they are complete. Even in crowded scenarios, they allow finding a feasible solution that brings each agent to its target, preventing deadlock situations. However, generally, rule-based algorithms provide solutions that are much longer than the optimal one. The main contribution of this paper is the introduction of an iterative local search procedure in MAPF. We start from a feasible suboptimal solution and we perform a local search in a neighborhood of this solution, to find a shorter one. Iteratively, we repeat this procedure until the solution cannot be shortened any longer. At the end, we obtain a solution, that is still sub-optimal, but, in general, of much better quality than the initial one. We use dynamic programming for the local search procedure. Under this respect, the fact that our search is local is fundamental to reduce the time complexity of the algorithm. Indeed, if we apply a standard dynamic programming the number of explored states grows exponentially with the number of agents. As we will see, the introduction of a locality constraint allows solving the (local) dynamic programming problem in a time that grows only polynomially with respect to the number of agents.