Fast and interactive inverse simulation supporting human-robot deformable object manipulation *

Abstract

The problem of shaping soft objects is widespread in industrial, medical, and household settings. Hence, robotic Deformable Object Manipulation (DOM) is a field of research that has recently emerged to improve robotic systems’ ability to handle such objects. Indeed, human-robot collaboration is also relevant to applications featuring soft objects, since the decision-making and dexterity of a human operator are currently beyond reach.Our aim is to assess the feasibility of using fast finite element inverse simulation in collaborative shaping tasks. To this end, we propose a computationally efficient method for controlling the shape of an object grasped at both ends. In our experimental setup, a leader robot moves freely along unplanned trajectories, while a controlled robot maintains the desired shape despite these perturbations. We reach an update frequency of 20 Hz for the inverse simulation, with average steady-state shape errors below 10 mm in most cases. Those satisfying results allow us to envision our next milestone: the deployment of the inverse simulation in real human-robot shaping tasks.