Heterogeneous GNN for Multi-Agent Path Planning Roadmaps

A scalable heterogeneous Graph Neural Network framework automates the generation and evaluation of shared multi-agent roadmaps. By training on expert trajectories, the GNN learns to prune redundant nodes and edges while reasoning over global connectivity and inter-agent interactions.