Offline-Online Learning of Deformation Model for Cable Manipulation With Graph Neural Networks

Manipulating deformable linear objects by robots has a wide range of applications, e.g., manufacturing and medical surgery. To complete such tasks, an accurate dynamics model for predicting the deformation is critical robust control. In this letter, we deal with challenge proposing hybrid offline-online method to learn cables in data-efficient manner. offline phase, adopt Graph Neural Network (GNN) purely from simulation data. Then residual learned real-time bridge sim-to-real gap. The then utilized as constraint trust region based Model Predictive Controller (MPC) calculate optimal robot movements. online learning MPC run closed-loop manner robustly accomplish task. Finally, comparative results existing methods are provided quantitatively show effectiveness robustness.