Robotic manipulation of cloth-like deformable objects

The manipulation of cloth-like deformable objects represents a challenging problem. Clothes are characterized by having shape, appearance, and other mechanical and visual properties to vary due to previous manipulations or external effects. In this thesis the problem of grasping cloth-like objects is addressed. This dissertation is part of a project where a mobile robot is involved in the loading and unloading of clothes for the end-of-line tests of washing machines. In particular, the thesis focuses on estimating target poses that would result in reliable grasping operations for a robotic arm. The processing of input pointclouds coming from a 3D camera is performed in order to develop strategies and algorithms for grasping clothes both from a bin randomly placed nearby the robot but also for clothes placed inside a drum or across its opening door. The structure of the developed algorithms is organized into three layers. In addition, the problem of avoiding collisions is analyzed, in particular inside the drum where the plastic paddles are identified with this purpose. The PointCloud library along with the Eigen library are utilized to perform the processing of the pointclouds. Chapter 2 focuses on the grasping of clothes from an external bin while Chapter 3 describes how the paddles in the washing machine are localized. Chapter 4 provides solutions for the grasping of clothes placed inside the drum. Chapter 5 addresses the problem of detecting cloths along the drum opening. Chapter 6 shows a possible application of the algorithms described in Chapter 2 and 5 on a real robot employing tools like ROS, Moveit! and a behavior three as task manger.