Motivation

Manipulating deformable linear objects (DLOs) is a complex yet essential task for industrial applications like cable routing and hot-wire cutting. Combining offline learned interaction models with online adaptation schemes has shown promise, but existing approaches still suffer from slow and unreliable execution. We propose a real-time gradient-based model predictive control (MPC) approach for 3D bimanual cable shaping, yielding improved performance and fast execution by effectively utilizing a learned interaction model.

Method

Results