Neural Deconstruction Search for Vehicle Routing Problems
About
Autoregressive construction approaches generate solutions to vehicle routing problems in a step-by-step fashion, leading to high-quality solutions that are nearing the performance achieved by handcrafted operations research techniques. In this work, we challenge the conventional paradigm of sequential solution construction and introduce an iterative search framework where solutions are instead deconstructed by a neural policy. Throughout the search, the neural policy collaborates with a simple greedy insertion algorithm to rebuild the deconstructed solutions. Our approach matches or surpasses the performance of state-of-the-art operations research methods across three challenging vehicle routing problems of various problem sizes.
Andr\'e Hottung, Paula Wong-Chung, Kevin Tierney• 2025
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Capacitated Vehicle Routing Problem | CVRP 1000 | Objective Value41.11 | 29 | |
| Capacitated Vehicle Routing Problem | CVRP500 | Objective Value36.57 | 25 | |
| Vehicle Routing Problem with Time Windows | VRPTW 500 customers | Objective Value47.94 | 11 | |
| Vehicle Routing Problem with Time Windows | VRPTW 1K customers | Objective Value87.54 | 11 | |
| Capacitated Vehicle Routing Problem | CVRP Low Capacity 500 instances (test) | Objective Value91.15 | 5 | |
| Capacitated Vehicle Routing Problem | CVRP Clustered Distribution 500 instances (test) | Objective Value44.29 | 5 | |
| Prize-collecting Vehicle Routing Problem | PCVRP 500 | Objective Value43.12 | 3 | |
| Capacitated Vehicle Routing Problem | CVRP 2000 | Objective Value56 | 3 | |
| Prize-collecting Vehicle Routing Problem | PCVRP-1000 | Objective Value80.99 | 3 | |
| Prize-collecting Vehicle Routing Problem | PCVRP 2000 | Objective Value158.1 | 3 |
Showing 10 of 11 rows