Abstract
This paper presents a classical optimization approach for solving the Vehicle Routing Problem using Google OR-Tools, with a focus on real-case urban logistics applications. Leveraging an real world dataset comprising delivery locations and one depot. The study formulates the VRP as a combinatorial optimization problem and computes realistic inter-node distances using the Open Source Routing Machine API. Two distinct search solutions are implemented: a constructive heuristic based on a greedy algorithm and an enhanced solution refinement employing metaheuristic with a time-constrained search. Comparative experiments indicate that the OR-Tools approach significantly reduces total travel distance and computation time compared to manually constructed routes, thereby demonstrating its practical viability for vehicle routing problem and fleet management. The study also discusses the scalability of this framework for multi-vehicle scenarios and the integration of dynamic constraints, which is essential for adaptive urban logistics in rapidly changing environments.Comparative experiments illustrate the practical benefits of the OR-Tools approach. In Experiment 1, covering 30 delivery points, the manually built route measured 103.3 km, whereas the basic OR-Tools solution reduced this to 96.1 km, an improvement of 7.2 km (approximately 7% reduction) computed in just 0.02 seconds. A fine-tuned configuration further decreased the route to 95.3 km (an 8 km reduction relative to the manual route) within 1 second. In Experiment 2, involving 54 delivery points, the manual route was 127.6 km long. Here, the basic OR-Tools solution produced a route of 123.8 km (3.8 km or roughly 3% shorter), and the fine-tuned solver achieved an optimized route of 120.8 km—reducing the distance by 6.8 km (about 5.3%) when given a 30-second computation limit. Overall, this framework demonstrates that classical optimization can deliver transparent, computationally efficient, and scalable solutions for vehicle routing and fleet management, offering a robust alternative to manual planning and more complex machine learning-based approaches in dynamic urban logistics environments.