Multi-Depot Vehicle Routing Problems: A Narrative Review of Problem Variants, Algorithms, and Benchmarking Practices (2021–2026)
This narrative review systematically analyzes 100 studies on the Multi-Depot Vehicle Routing Problem (MDVRP) published between 2021 and 2026, highlighting the field's evolution toward complex real-world constraints like time windows and sustainability, the dominance of hybrid metaheuristics alongside emerging learning-based methods, and the inconsistent state of benchmarking practices.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a city where goods must be delivered to thousands of homes, but the delivery trucks do not all start from a single central warehouse. Instead, they launch from a dozen different depots scattered across the region, each with its own fleet of vehicles. The challenge is to decide which truck leaves from which depot, which customers it should visit, and in what order, all while ensuring no vehicle is overloaded, every customer is served within their requested time window, and the total distance traveled is as short as possible. This is the Multi-Depot Vehicle Routing Problem, a complex logistical puzzle that sits at the heart of modern supply chains. It is far more difficult than the simpler version where all trucks start from one place, because the system must now solve two problems at once: assigning customers to the right depots and then plotting the most efficient paths for the vehicles. As the world moves toward faster deliveries, electric fleets, and more sustainable operations, this problem has become increasingly critical for businesses and cities alike.
A team of researchers from the Chittagong University of Engineering & Technology recently took a deep look at how scientists and engineers have been tackling this challenge over the last five years. They examined one hundred studies published between 2021 and 2026 to understand how the field has evolved, what new problems are being solved, and how researchers are testing their solutions. Their review reveals a field in rapid transition. While the core goal remains minimizing costs and travel distance, the real-world constraints have multiplied. Today's models must account for electric vehicles that need to stop for charging, strict delivery windows that customers demand, and the coordination of multiple depots working together. The researchers found that time windows are now a standard feature in nearly sixty percent of the studies, and the inclusion of electric and green routing has grown significantly, reflecting a global shift toward sustainability.
The way researchers solve these problems has also shifted. For decades, the primary tool has been metaheuristics, a class of smart, trial-and-error algorithms that search for good solutions without guaranteeing the absolute perfect one. These methods remain the dominant approach, appearing in eighty-three percent of the studies reviewed. However, the landscape is changing. There is a noticeable rise in hybrid frameworks, where traditional algorithms are combined with exact mathematical methods or new learning-based techniques. In particular, methods that use machine learning, such as deep reinforcement learning and neural networks, are gaining traction. These learning-based approaches, which make up about seventeen percent of the recent work, offer the promise of making decisions quickly once they are trained, though they are often used to support rather than replace the established optimization engines.
Despite these advances, the researchers identified a significant hurdle in how progress is measured. To know if a new algorithm is truly better, scientists need to test it against the same set of problems, known as benchmarks. The review found that while the classic benchmark sets created years ago are still used, they are no longer enough. These older datasets were designed for simple trucks and single depots; they cannot capture the complexity of electric fleets, drone deliveries, or dynamic traffic conditions. Consequently, many recent studies have created their own custom datasets or modified existing ones to fit their specific needs. While this allows for more realistic testing, it has created a fragmented landscape where different researchers use different test cases, making it difficult to compare results directly. The authors note that fewer than half of the studies rely on the classical shared references, and many of the new, realistic datasets are not made public, which limits the ability of others to verify or build upon the work.
The review also highlights how the problems themselves have become more specialized. Beyond the basic task of moving goods, researchers are now modeling scenarios involving cold-chain logistics for perishable foods, hazardous material transport, and even the coordination of ground vehicles with drones. In these complex setups, constraints like battery capacity, charging times, and synchronization between different types of vehicles become critical. The studies show that these real-world details are no longer optional add-ons but are central to the models. Yet, the paper suggests that while the mathematical models are becoming more sophisticated, the community still lacks a unified way to test them. The authors conclude that for the field to move forward effectively, there is an urgent need for standardized, publicly available datasets that reflect these modern, feature-rich challenges. Without this common ground, the impressive advances in algorithms and modeling may remain isolated, unable to be fully compared or integrated into the next generation of logistics systems.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.