COAgents: Multi-Agent Framework to Learn and Navigate Routing Problems Search Space

Oleksandr Yakovenko, Mahdi Mostajabdaveh, Cheikh Ahmed, Abdullah Ali Sivas, Xiaorui Li, Zirui Zhou, Mao Kun

arXiv:2605.20618 · 2026-05-22 공개 · arXiv · PDF

multi-agent-framework vehicle-routing-problems search-space-navigation partial-search-graph node-selection-agent move-selection-agent jump-agent cvrp

Abstract

Although Vehicle Routing Problems (VRP) are essential to many real-world systems, they remain computationally intractable at scale due to their combinatorial complexity. Traditional heuristics rely on handcrafted rules for local improvements and occasional \textit{jumps} to escape local minima, but often struggle to generalize across diverse instances. We introduce \textbf{COAgents}, a cooperative multi-agent framework that models the search process as a graph: nodes represent solutions, and edges correspond to either local refinements or large perturbations for diversification (i.e., jumps). A \textit{Partial Search Graph} (PSG) is dynamically constructed during search, enabling COAgents to train a Node Selection Agent and a Move Selection Agent to guide intensification, and a Jump Agent to trigger well-timed explorations of new regions. Unlike end-to-end learning approaches, COAgents cleanly separates problem-agnostic search control from compact domain-specific encoding, facilitating adaptability across tasks. Extensive experiments on the CVRP and VRPTW benchmarks show that COAgents remains competitive with several learn-to-search baselines on CVRP and sets a new state of the art among learning-based methods on the more challenging VRPTW instances, reducing the gap to the best-known solutions by 14\% at $N\!=\!100$ and 44\% at $N\!=\!50$ relative to the strongest neural solver (POMO), and by 21\% and 40\% respectively relative to ALNS. Code is available at https://github.com/mahdims/COAgents.

한국어 요약

한 줄 요약

**[Vehicle Routing / Multi-Agent RL]** COAgents는 search 공간을 그래프로 모델링하고 Node·Move·Jump 3 에이전트가 협력 탐색, VRPTW에서 학습 기반 방법 SOTA, N=100 14%·N=50 44%로 POMO 대비 best-known 격차 단축.

핵심 기여도

핵심 아이디어

VRP 등 조합 최적화의 학습 기반 가속은 종단 솔버 학습 대신, search 공간 자체를 그래프로 모델링하고 intensification과 diversification을 별도 학습 모듈로 분리해 problem-agnostic·도메인-specific 인코딩을 깔끔 분리하는 것이 일반화 핵심이다.

기술적 접근법

주요 결과

의의 및 한계

**의의**: VRP의 학습 기반 search에 멀티에이전트 분업 원리 적용, intensification/diversification 분리로 일반화 향상, VRPTW SOTA로 실용 가치, problem-agnostic search control의 다른 조합 문제 확장 잠재. **한계**: CVRP·VRPTW 중심으로 다른 routing variant(pickup-and-delivery, multi-depot)는 추가 검증, PSG 동적 구성의 메모리 비용, 매우 큰 instance 일반화는 후속.

실용적 활용