Exact Search Path Planning with Graph Neural Networks

We propose a Graph Neural Network (GNN) based branch-and-bound algorithm for solving path search optimization problems. By modeling the problem as a mixed-integer linear program and employing a GNN-supervised branching policy, we significantly improve solving efficiency. Experiments demonstrate that our method achieves balanced performance in wall time and explored nodes compared to traditional methods, being 49.8% faster than vanillafullstrong and showing good generalization ability across different problem scales. The analysis confirms the method's advantages in computational efficiency and stability.