Accelerating Vehicle Routing via AI-Initialized Genetic Algorithms

Ido Greenberg, Piotr Sielski, Hugo Linsenmaier, Rajesh Gandham, Shie Mannor, Alex Fender, Gal Chechik, Eli Meirom

Keywords: Vehicle Routing Problem, Traveling Salesperson Problem, Reinforcement Learning, Genetic Algorithms, Combinatorial Optimization, Initialization, Capacity Regularization

Abstract: This work introduces a novel hybrid method combining Reinforcement Learning and Genetic Algorithms to solve Vehicle Routing Problems (VRP). While Machine Learning approaches have been extensively applied to VRP, they have struggled to surpass state-of-the-art optimization methods. Our approach bridges this gap by leveraging the strengths of both ML and traditional optimization techniques. We also design a novel regularization method for learning to minimize the number of vehicles (in addition to travel distance), as required in many applications. Our method presents state-of-the-art solution costs given limited optimization time budgets, and scales to hundreds of locations within seconds.

Submission Number: 25