SPACE: Unifying Symmetric and Asymmetric Routing Problems for Generalist Neural Solver
The paper titled 'SPACE: Unifying Symmetric and Asymmetric Routing Problems for Generalist Neural Solver' presents a new framework for addressing vehicle routing problems. The proposed SPACE framework aims to unify node representation and solution generation across symmetric and asymmetric settings. Extensive experiments demonstrate its effectiveness in achieving zero-shot generalization for various routing problem variants.
- ▪The SPACE framework defines the spatial position of each node based on relative distances to specific pivots.
- ▪It introduces a bidirectional Frechet representation using a novel furthest pivot sampling strategy.
- ▪The framework includes a weight-decomposed adaptive decoding mechanism to improve performance across different geometric settings.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.24484 (cs) [Submitted on 23 May 2026] Title:SPACE: Unifying Symmetric and Asymmetric Routing Problems for Generalist Neural Solver Authors:Rongsheng Chen, Changliang Zhou, Canhong Yu, Yuanyao Chen, Yu Zhou, Zhuo Chen, Zhenkun Wang View a PDF of the paper titled SPACE: Unifying Symmetric and Asymmetric Routing Problems for Generalist Neural Solver, by Rongsheng Chen and 6 other authors View PDF HTML (experimental) Abstract:Generalist neural routing solvers have shown great potential in solving diverse vehicle routing problems (VRPs) with a unified model.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
