Graph Neural Network-Based Motion Planning for Mobile Robots
TL;DR
Major GNN architectures are reviewed, including Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), GraphSAGE, and Message Passing Neural Networks (MPNNs), and their applications in warehouse automation, autonomous vehicles, service robots, drones, and search-and-rescue missions are discussed.
Abstract
Motion planning is a fundamental challenge in autonomous mobile robotics, requiring robots to navigate efficiently in dynamic and uncertain environments. Traditional algorithms such as Dijkstra, A*, RRT, and PRM perform well in structured settings but often struggle with moving obstacles, high computational costs, and limited adaptability. Recent advances in Artificial Intelligence, particularly Graph Neural Networks (GNNs), provide a more effective solution by representing robotic environments as graphs, where nodes denote robot states or waypoints and edges represent feasible movements. Through message passing and graph-based learning, GNNs capture both local and global spatial relationships, enabling efficient path planning, obstacle avoidance, and trajectory optimization in complex environments. This paper reviews major GNN architectures, including Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), GraphSAGE, and Message Passing Neural Networks (MPNNs), and discusses their applications in warehouse automation, autonomous vehicles, service robots, drones, and search-and-rescue missions. It also highlights current challenges such as limited interpretability, high training costs, insufficient benchmark datasets, and real-time deployment constraints. Overall, GNNs represent a promising direction for scalable, adaptive, and intelligent robotic motion planning.