A Decade of Progress in the Capacitated Vehicle Routing Problem: Variants, Metaheuristics, Machine Learning, and Applications

The Capacitated Vehicle Routing Problem (CVRP) is fundamental to logistics optimization and central to improving routing efficiency across industries. Metaheuristics have been pivotal in solving the CVRP and its variants over the last decade. Despite extensive work on vehicle routing, a focused, comprehensive review of CVRP and its variants—especially solutions based on metaheuristics—has remained limited. This paper presents a review of CVRP developments from 2014 to 2024, highlighting how metaheuristic algorithms are designed and integrated with other methods to address emerging challenges. By examining 220 research articles, we trace the evolution of 25 CVRP variants, spanning urban last-mile delivery to disaster response. Our analysis shows a clear post-2020 shift: about 69% of new variants target eco-friendly goals or incorporate technological/structural elements. We also assess the most-used metaheuristics—Genetic Algorithms (19.1%), Ant Colony Optimization (10.1%), and Simulated Annealing (7.9%)—and discuss their hybridizations and links to Machine Learning (ML). A growing trend is to embed data-driven ML within metaheuristic frameworks for improved initialization, adaptive operator choice, and policy learning, thereby improving adaptability in real-world settings. Overall, this review consolidates current insights, identifies gaps, and outlines future directions to advance robust, scalable, and sustainable CVRP solutions.