Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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

TaskDatasetResultRank
Capacitated Vehicle Routing ProblemCVRP 1000
Objective Value41.11
29
Capacitated Vehicle Routing ProblemCVRP500
Objective Value36.57
25
Vehicle Routing Problem with Time WindowsVRPTW 500 customers
Objective Value47.94
11
Vehicle Routing Problem with Time WindowsVRPTW 1K customers
Objective Value87.54
11
Capacitated Vehicle Routing ProblemCVRP Low Capacity 500 instances (test)
Objective Value91.15
5
Capacitated Vehicle Routing ProblemCVRP Clustered Distribution 500 instances (test)
Objective Value44.29
5
Prize-collecting Vehicle Routing ProblemPCVRP 500
Objective Value43.12
3
Capacitated Vehicle Routing ProblemCVRP 2000
Objective Value56
3
Prize-collecting Vehicle Routing ProblemPCVRP-1000
Objective Value80.99
3
Prize-collecting Vehicle Routing ProblemPCVRP 2000
Objective Value158.1
3
Showing 10 of 11 rows

Other info

Follow for update