A GPU-Accelerated Hybrid Method for a Class of Multi-Depot Vehicle Routing Problems
New algorithm solves large-scale logistics routing 10x faster using GPU tensors.
A team led by Zhenyu Lei and Jin-Kao Hao has unveiled a new algorithm that uses GPU acceleration to solve multi-depot vehicle routing problems (MDVRPs) far more efficiently than existing methods. The hybrid approach combines a learning-driven, diversity-controlled route-exchange crossover with a multi-depot feasible-and-infeasible search framework guided by a multi-penalty evaluation function. Two dedicated depot-related local search operators further enhance the search in multi-depot settings. To boost scalability, the researchers developed an enhanced version that leverages tensor-based GPU acceleration alongside a novel multi-move update strategy, allowing simultaneous evaluation of many route changes.
Extensive experiments across three MDVRP benchmark variants demonstrate that the algorithm is highly competitive with state-of-the-art solvers, particularly on large-scale instances where traditional methods struggle. By harnessing parallel GPU compute, the method can explore solution spaces that were previously computationally prohibitive. This could lead to significant efficiency gains in logistics, supply chain management, and delivery routing—industries where even small percentage improvements translate into massive cost savings. The paper is available on arXiv under reference 2605.05208.
- Integrates learning-driven crossover with multi-penalty search for MDVRPs
- Tensor-based GPU acceleration and multi-move update strategy for scalability
- Outperforms state-of-the-art methods on three MDVRP variants, especially large-scale instances
Why It Matters
Faster, scalable routing optimization unlocks major cost savings in logistics, delivery, and supply chain operations.