DET: Learn to Solve the Tunnel Traveling Salesmen Problem using Double-Encoder Transformer

Nan Zhao, Jinshan Zhang, Paul Weng, Feng Wang, Jianwei Yin

Keywords: Combinatorial Optimization; Transformer; Deep Reinforcement Learning; Tunnel TSP

Abstract: We delve into a challenging variant of the Traveling Salesman Problem (TSP), namely tunnel TSP, which incorporates a new important constraint requiring the traversal of a prescribed set of tunnels. While traditional deep reinforcement learning (DRL) based neural TSP algorithms excel in optimizing routes without tunnel restrictions, they often struggle to achieve optimal performance in tunnel TSP due to the neglect of the crucial role of tunnel attributes during solution generation. To address this challenge, we propose a simple but effective and flexible technique, called Double-Encoder Transformer (DET), which can be seamlessly integrated into various existing autoregressive neural TSP solvers. DET processes node and tunnel location information separately and encodes them in two distinct feature spaces. Following an efficient fusion strategy, DET then integrates the encoded information from nodes and tunnels, harnessing their intricate interactions. Experimental validation demonstrates that integrating DET into existing autoregressive neural solvers significantly improves performance, enabling us to reduce the average optimality gap for tunnel TSP from 12.58% (of the previous Single-Encoder model) to 7.35%.

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: 437