The EU-funded ROMANDIC project is advancing robot manipulation of deformable objects, like textiles, by uniting top research institutes, industry, and policymakers to establish a Network of Excellence.
Motivation
Robot handling of deformable objects, such as textiles, is a burgeoning research area propelled by advanced robot hardware and AI. The Jožef Stefan Institute (JSI) in Slovenia, the Karlsruhe Institute of Technology in Germany and the Institut de Robòtica i Informàtica Industrial in Spain recognise the potential of the field and aim to become key players.
The EU-funded ROMANDIC project aims to bring together academic and industrial stakeholders, policymakers and regulators to establish a Network of Excellence in Robot Manipulation of Deformables. The goal is to foster and advance research with partners domestically and internationally. Additionally, the project will assist JSI in enhancing its management and leadership capabilities through knowledge transfer and the exchange of best practices.
Objective
Robot handling of deformable objects, such as textiles, is a research topic that has hit fertile ground with the advancement and accessibility of robot hardware, a change in thinking of application of robotics also outside of factories, and especially novel control and perception methodologies stemming from AI approaches. We want to combine and bring forward separate efforts on this topic by becoming a Network of Excellence (NOE) in Robot Manipulation of Deformables in the truest sense of the definition, as a network of institutions collaborating with each other to pursue excellence in a particular area.
The proposed Twinning project initiative ROMANDIC (RObot MANipulation of Deformables through dynamIC actions) will provide the means to promote this novel and exciting field of research at JSI and internationally, but more importantly, facilitate and consolidate the managing and leading capabilities of (JSI) through knowledge transfer and exchange of best practice, so that it can indeed act as an informal COE.
Concept and approach
The variability of movements when handling textiles is beyond hand-programmed robot movements. Thus, learning based methods have dominated research in fabric manipulation, often using either human demonstrations, sim-to-real transfer, or both. It has been proved that human demonstrations improve performance also in bimanual folding of garments. Given the complexity of manipulation tasks performed by robots interacting with the physical world, many robot architectures include “task and motion planning” (TAMP) methods that combine symbolic reasoning based on a discrete, abstract, relational representation of states and actions with motion planning and control based on a continuous-space, fine-grained representation.
Complex interaction with the environment and objects has driven the development of novel robotics systems specialized in manipulation. Recent studies on grasping for cloth-like objects point to the need to increase the variety of grasp types and geometries used to manipulate garments to enable more efficient dynamics. At the moment, mainly simple pinch grasps have been used, but grippers that go beyond are receiving increasing interest.
Industrial impact
The proposed project ROMANDIC pushes the limit of performance and physical intelligence because there is still no working robot setup capable of effectively manipulating deformables at near industrial levels. We have the opportunity to create a community and a technological roadmap in the field of robot handling of deformables, which will enable us to make the traditional industrial processes that handle soft deformable materials more agile and sustainable.