UniteFormer: Unifying Node and Edge Modalities in Transformers for Vehicle Routing Problems

The PyTorch Implementation of NeurIPS 2025 --"UniteFormer: Unifying Node and Edge Modalities in Transformers for Vehicle Routing Problems."

Neural solvers for the Vehicle Routing Problem (VRP) have typically relied on either node or edge inputs, limiting their flexibility and generalization in real-world scenarios. We propose UniteFormer, a unified neural solver that supports node only, edge-only, and hybrid input types through a single model trained via joint edge-node modalities. UniteFormer introduces: (1) a mixed encoder that integrates graph convolutional networks and attention mechanisms to collaboratively process node and edge features, capturing cross-modal interactions between them; and (2) a parallel decoder enhanced with query mapping and a feed-forward layer for improved representation. The model is trained with REINFORCE by randomly sampling input types across batches. These results underscore UniteFormer’s ability to handle diverse input modalities and its strong potential to improve performance across various VRP tasks.

Overview

Dependencies

Python >= 3.8
Pytorch >= 2.0.1
numpy==1.24.4
matplotlib==3.5.2 
tqdm==4.67.1

Download datasets and models

Download datasets and models from Hugging Face.

Unzip UF-TSP-results.zip and UF-CVRP-results.zip, and organize the files in the project directory as follows:

UniteFormer
├─ UF-TSP
│  ├─ data
│  └─ train_models
└─ UF-CVRP
   ├─ data
   └─ train_models

Citation

Acknowledgments

• https://github.com/wouterkool/attention-learn-to-route
• https://github.com/yd-kwon/POMO
• https://github.com/yd-kwon/MatNet