Manipulating Deformable Objects with a Dual-arm Robot

Shape visual servoing of a tether cable from parabolic features

“…Since these tasks are still mainly performed manually, their automation by robots has a great potential [8], [7], [6], [23]. In particular, two robot arms are used in [1] for manipulating a flexible beam in order to adapt its shape before an assembly operation. Accurately and automatically controlling the shape of a suspended cable can also be useful for Unmanned Aerial Vehicle (UAV) applications such as managing the shape of the power cable of tethered UAVs.…”

“…Experiments presented in this paper were carried out thanks to a platform of the Robotex 2.0 French research infrastructure 1 L. Smolentsev, A. Krupa, and F. Chaumette are with Inria, Univ Rennes, CNRS, IRISA, Campus de Beaulieu, 35042 Rennes, France Firstname.Name@inria.fr into two families: model-based and model-free approaches. In the model-free branch, the DLO is usually modeled as a set of points [1], [10], [19] or a contour [14] that can be represented by geometrical features. The common way to control the object deformation is by minimizing the error between the current measure of these features and their desired values corresponding to the targeted shape.…”

“…The common way to control the object deformation is by minimizing the error between the current measure of these features and their desired values corresponding to the targeted shape. In [1] the considered features are the 3D positions of markers attached on the DLO while in [10] a set of 3D equidistant points lying on a B-spline curve fitting the shape is considered. In [14] four different cases of features are studied: contour polynomial spline coefficients, Bezier and Nurbs splines control points, and Fourier coefficients.…”

“…In [14] four different cases of features are studied: contour polynomial spline coefficients, Bezier and Nurbs splines control points, and Fourier coefficients. To relate the variation of these features to the velocities of the robot end-effector that interacts with the DLO, [1] propose to estimate an approximate shape Jacobian matrix during an offline identification procedure that consists in deforming the DLO by applying open-loop step motions on each robot DOF and measuring the induced displacements of the attached markers. The shape Jacobian matrix can also be estimated online from visual data by using a Kalman filter [14] or by minimizing a quadratic objective on a sliding time windows [10].…”