Multi-robot Manipulation of Deformable Objects - Leveraging Interactive Simulation for Planning and Control – Mandolin

The project asks how multiple collaborative robots can shape deformable objects in 3D by optimizing grasping point location and controlling deformation in a closed loop, using interactive inverse finite element simulation.
Deformable object manipulation (DOM) is a valuable skill for robots in both industrial and household contexts. Examples of manipulated objects are clothes, cables, seals, rubber parts, or food. When performing these manipulation tasks, it is difficult to model the deformation pattern of the object, manipulate the object during the whole task without damage, and reach the desired position and shape. For real-world applications, these difficulties should be addressed while considering the possibility of perturbations and contact with the environment.
To tackle this topic, we propose an interdisciplinary approach involving perception, robot control, mechanical simulation, numerical methods, and robot system integration. We will (i) plan the optimal shape and grasp location on the deformable object, (ii) control the robots manipulating the object to impose the desired deformation, and (iii) perform real-time simulation of the deformation to predict object deformation. Results are meant to allow automation of tasks involving flexible objects, while still using conventional robots, sensors, and control methods. In that sense, the applicability of results will be shown using real-world use cases of manipulation in a multi-robot system.
The Mandolin project is structured around four technical work packages addressing deformation planning, shape control, interactive inverse finite element simulation, and validation on a multi-robot system. A consortium of experts led by LCFC (Metz) and composed of Institut Pascal (Clermont-Fd) and iCube (Strasbourg) will accomplish this work.