Physics-Informed Graph Learning for Shape Prediction in Robot Manipulate of Deformable Linear Objects

• A novel Physics-Informed Graph Neural Network (PIGNN) integrates physical laws into the loss function to accurately predict complex deformations of deformable linear objects (DLO) such as fibers, addressing nonlinear behaviors like bending and torsion. • The method leverages GNNs to learn an initial deformation model, mitigating data scarcity issues and improving prediction accuracy even with limited training data. • Experimental results show that PIGNN significantly reduces both execution time and prediction error compared to traditional methods, making it suitable for real-time robotic manipulation of DLO. • The approach demonstrates strong potential for applications in robotics, medical devices, aerospace, and manufacturing, particularly in tasks like wire assembly and surgical suturing.