Cloth Mechanical Parameter Estimation and Simulation for Optimized Robotic Manipulation

Nikolaos E. Anatoliotakis, Panagiotis Koustoumpardis, Konstantinos Moustakas; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 2612-2620

In this article a method for the estimation of cloth simulation parameters is presented. We propose a method, based on already published methods from different fields, that can successfully create the mechanical model of a cloth, based only on a single monocular video source of a cloth been held and moved in the air by two hands. We propose the use of a moving graph generation method using Scale Invariant Feature Transformation (SIFT). Having the moving graph of the real cloth as the goal, a method based on genetic algorithms was designed to produce the mechanical properties of the cloth's mechanical model. This way a simulated cloth with similar mechanical properties will be created. For our experiments a mechanical model based on Provot's mass-spring-damper (MSD) cloth model with adjustable springs and dampers was used. However, we present a method that can be easily adjusted to any particlebased cloth model. The method presented was designed to be easily applicable so as to enable the broader use of cloth models in robotized cloth manipulation tasks. The use of a cloth's digital twin, enables the major part of tuning of a robot controller to be made offline. This will significantly accelerate the tuning process, enabling the broader use of robots in more delicate cloth manipulation tasks. Finally, to prove the validity of our method we provide the results of experiments with cloths of different patterns and physical parameters.