Data-Driven FEM Modeling of Elastic Cables for Robotic Manipulation Tasks

The automation of flexible cables assembling tasks by industrial robots is still very challenging. Unlike rigid objects, flexible objects’ state is uncertain, and their deformation is crucial for performing correctly a task. When a human being manipulates a cable, he can have quickly an idea of its geometry and stiffness just by looking at it and shaping it by hands. In this work, we propose a method that mimics this process and makes a robot learn the geometry and stiffness of a cable. The goal is to have a simulator in which the robot can generate the cable and tune its stiffness and damping properties by comparing the simulated deformed shape with that of a cable deformed in a real scenario. At first, the geometry of the cable is obtained by a Lidar, and a FEM model is generated and imported into a simulator. The virtual cable is shaped in a plane by a robot arm, and the resulting shape is compared with that obtained by conducting the same test with a real robot. An optimization based on Bayesian inference and Gaussian process is then used to tune the stiffness and damping properties of the simulated cable. The proposed method shows the ability to generate the cable from scratch and tune its parameters to obtain the desired shape in the plane.