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

2026-05-20 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

COAgents is a cooperative multi-agent framework designed to address the computational intractability of Vehicle Routing Problems (VRPs) at scale. It models the VRP search process as a graph where nodes are solutions and edges represent local refinements or large perturbations. The framework dynamically constructs a Partial Search Graph (PSG) to train a Node Selection Agent and a Move Selection Agent for intensification, alongside a Jump Agent for well-timed exploration. Unlike end-to-end learning, COAgents separates problem-agnostic search control from domain-specific encoding, improving adaptability. Experiments on CVRP and VRPTW benchmarks show COAgents is competitive on CVRP and achieves a new state of the art for learning-based methods on VRPTW. It reduces the gap to best-known solutions by 14% at N=100 and 44% at N=50 relative to POMO, and by 21% and 40% respectively relative to ALNS.

Key takeaway

For Machine Learning Engineers developing solutions for complex Vehicle Routing Problems, COAgents offers a significant advancement. Its multi-agent framework, which achieves a new state of the art on VRPTW by reducing the gap to best-known solutions by up to 44%, suggests you should explore hybrid learning-based approaches. Consider integrating specialized agents for search control and diversification into your optimization models to improve scalability and generalization across diverse instances.

Key insights

COAgents uses cooperative agents and a dynamic search graph to efficiently navigate complex routing problem spaces.

Principles

• Model search as a graph of solutions and transitions.
• Separate search control from domain encoding for adaptability.
• Employ specialized agents for intensification and diversification.

Method

COAgents dynamically constructs a Partial Search Graph (PSG) during search. It trains a Node Selection Agent and a Move Selection Agent for intensification, and a Jump Agent for exploration, guiding the search process.

In practice

• Apply multi-agent systems to combinatorial optimization.
• Use graph-based search for VRPs.
• Integrate intensification and diversification agents.

Topics

• Vehicle Routing Problems
• Multi-Agent Systems
• Combinatorial Optimization
• Search Algorithms
• Reinforcement Learning
• VRPTW Benchmark

Code references

• mahdims/COAgents

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• LLM-Driven Co-Evolutionary Automated Heuristic Design for Bi-Component Coupled Combinatorial Optimization — CoEvo-AHD uses LLMs and co-evolution to design heuristics for coupled combinatorial optimization problems, improving solution quality.
• RACL: Reasoning-Agent Control Layers for Continuous Metaheuristic Learning — RACL uses a reasoning agent to dynamically control metaheuristic search behavior for improved optimization.
• Mango: Multi-Agent Web Navigation via Global-View Optimization — Global website structure analysis and adaptive URL prioritization significantly improve web navigation efficiency for LLM agents.
• AI Agents of the Week: Papers You Should Know About — Autonomous AI agents are evolving rapidly, showing both advanced capabilities and significant challenges in real-world and complex tasks.
• Bilevel Optimization of Agent Skills via Monte Carlo Tree Search — Bilevel optimization with MCTS and LLMs effectively refines LLM agent skills by separating structure search from content refinement.
• HMACE: Heterogeneous Multi-Agent Collaborative Evolution for Combinatorial Optimization — Heterogeneous multi-agent collaboration with specialized roles and memory-guided reflection enhances heuristic discovery for COPs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.