Learning to solve the Skill Vehicle Routing Problem with Deep Reinforcement Learning

Nayeli Gast Zepeda, André Hottung, Kevin Tierney

Confirmation: I have read and agree with the workshop's policy on behalf of myself and my co-authors.

Tracks: Special Session 1: (Deep) Reinforcement Learning in OR Optimization

Keywords: Neural Combinatorial Optimization, Deep Reinforcement Learning, Routing Problems

TL;DR: We show that the skill VRP, a resource constrained problem, is a challenging problem to learn for neural solvers due to its constraints making it difficult to find feasible solutions.

Abstract: Neural combinatorial optimization has proven effective in solving various simple routing problems including the traveling salesperson problem and the vehicle routing problem (VRP). However, real-world routing scenarios are usually significantly more complex, often requiring sophisticated methods to find even a single feasible solution. In this work, we apply neural combinatorial optimization to the more challenging skill VRP, where routes must be constructed for technicians with diverse skill sets while adhering to customer time windows. Due to the limited number of available technicians, finding feasible solutions is usually very challenging. We evaluate several state-of-the-art learning-based approaches on the skill VRP and explore different reward shaping techniques to penalize infeasible solutions during training. Our findings show that while most approaches can effectively solve instances with 20 customers, all approaches struggle to reliably find feasible solutions for instances with 50 customers.

Submission Number: 74