Learning the Partially Dynamic Travelling Salesman Problem

ICLR 2025 Conference Submission10811 Authors

Keywords: Reinforcement Learning, Graph Neural Networks, Dynamic Graphs, Dynamic Graph Neural Networks, Travelling Salesman Problem, Combinatorial Optimisation

TL;DR: Using Deep Reinforcement Learning with Graph Neural Networks to solve a dynamic variant of the Travelling Salesman Problem.

Abstract: Learning to solve the Travelling Salesman Problem (TSP) using Deep Reinforcement Learning (Deep RL) and Graph Neural Networks (GNNs) has shown promising results for small instances of the problem. We demonstrate that these methods can be extended to solve instances of a partially dynamic variant of the TSP. Solving this partially dynamic variant more effectively exploits the strengths of reinforcement learning and also presents challenges for more established methods of solving the TSP. We show the policies trained using Deep RL outperform modified versions of TSP solvers and heuristics for different distributions of dynamic vertices, including on larger instances than the policies were trained on. This shows the promise of Deep RL for solving this type of dynamic routing problem which is predicted to become of great importance as logistical services become more flexible and responsive to customer demand. Furthermore, our method is a general purpose approach to Deep RL where the problem consists of selecting items from a dynamically-evolving and arbitrarily-sized set.

Primary Area: reinforcement learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10811