Multi-Vehicle Routing
Graph neural network based solution to multi-Vehicle Routing Problem.
In this collaboration research with Alelab, we seek to use a learning-based approach to vehicle routing problem. We proposed a layered graph neural network (GNN) architecture consisting of a perception GNN (pGNN) that predicts individual optimal routes, and a communication GNN (cGNN) that coordinates routes between agents. In this work, I utilized expert solutions produced by Google’s ORTools to train the pGNN with imitation learning.