Yifan Guo, Zhongqiang Ren, Chen Wang. "iMTSP: Solving Min-Max Multiple Traveling Salesman Problem with Imperative Learning," IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024. Download PDF
iMTSP is a novel self-supervised framework to solve the multiply traveling salesmen problem (MTSP). By introducing a surrogate network as a control variate, iMTSP can efficiently train the allocation network through the non-differentiable TSP solver and the discrete decision space.
We formulate the MTSP as a bilevel optimization problem where the upper-level optimization decomposes MTSP into several small-scale TSPs, and the lower-level speedily solves these TSPs with a heuristic solver.
The figures demonstrate the advantages of iMTSP on two specific MTSP instances. We compare iMTSP with a reinforcement learning(RL)-based approach and Google OR-Tools routes module.
We also explicitly record the history of gradient variance of iMTSP and the RL baseline. Our method converges 20 times faster than the baseline with the help of the surrogate network.python == 3.8.15
PyTorch == 1.13.1
torch-geometric == 2.2.0
wandb == 0.15.2
ortools == 9.5.2237
Open data_generator.py, and modify the flag (training, validation, or testing), the number of nodes, and the batch size as needed. Run the script to generate data.
The validation process is embedded in the training files.
train.py contains the learning-based baseline.
my_train.py contains the proposed approach.
test.py
Trained models: Please refer to here for trained allocation networks and corresponding surrogate networks.
More details about this work are available in this paper.
@inproceedings{guo2024imtsp,
title = {{iMTSP}: Solving Min-Max Multiple Traveling Salesman Problem with Imperative Learning},
author = {Guo, Yifan and Ren, Zhongqiang and Wang, Chen},
booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2024},
}