COAgents: Multi-Agent Framework to Learn and Navigate Routing Problems Search Space
The article introduces COAgents, a multi-agent framework designed to address the complexities of Vehicle Routing Problems (VRP). This framework utilizes a cooperative approach to model the search process as a graph, enhancing adaptability across various tasks. Experimental results demonstrate that COAgents achieves competitive performance on benchmark instances, setting new records in learning-based methods for challenging VRPTW cases.
- ▪COAgents models the search process for routing problems as a graph with nodes representing solutions and edges for local refinements or large perturbations.
- ▪The framework includes a Partial Search Graph that dynamically constructs during the search to guide solution selection and exploration.
- ▪COAgents has shown to reduce the gap to the best-known solutions by significant margins on various benchmark instances.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.20618 (cs) [Submitted on 20 May 2026] Title:COAgents: Multi-Agent Framework to Learn and Navigate Routing Problems Search Space Authors:Oleksandr Yakovenko, Mahdi Mostajabdaveh, Cheikh Ahmed, Abdullah Ali Sivas, Xiaorui Li, Zirui Zhou, Mao Kun View a PDF of the paper titled COAgents: Multi-Agent Framework to Learn and Navigate Routing Problems Search Space, by Oleksandr Yakovenko and 6 other authors View PDF HTML (experimental) Abstract:Although Vehicle Routing Problems (VRP) are essential to many real-world systems, they remain computationally intractable at scale due to their combinatorial complexity.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
