AI-Driven Vehicle Routing Optimization: A Hybrid Model Using GNN, PPO, Tabu Search, and Agentic Intelligence

The Vehicle Routing Problem (VRP) remains a central challenge in logistics optimization, requiring efficient solutions that balance operational cost, service quality, and sustainability. Traditional heuristic and metaheuristic approaches achieve reasonable results but struggle to generalize in large-scale, dynamic environments. Recent advances in machine learning and reinforcement learning have opened new opportunities, yet standalone methods still face limitations in adaptability and semantic reasoning. This paper presents a hybrid framework that integrates Graph Neural Networks (GNN) for spatial representation, Proximal Policy Optimization (PPO) for sequential decision-making, Tabu Search for local refinement, and an Agentic Large Language Model (LLM) for high-level reasoning and constraint re-weighting. Experiments conducted on realistic VRP instances with sustainability-aware objectives—including distance, fuel consumption, and on-time delivery—demonstrate that the proposed architecture outperforms classical heuristics and pure learning models. Results show consistent improvements across operational and environmental metrics, highlighting the potential of agentic hybrid AI to support next-generation, sustainable transport management systems.