Jay Kamat

Cloth-like deformable objects, such as clothing items and household objects like bed sheets and blankets, are ubiquitous in our daily lives. Enabling robots to efficiently interact with such objects could greatly benefit industries like healthcare and fashion. However, these objects present unique challenges not encountered with rigid objects. Some of these challenges include accurately reasoning on the cloth's shape and state, and decision on how to grasp and manipulate. These tasks are complex due to the inherent complexity of the configuration space of cloth, how it interacts with the environment as well as the high variability of physical properties of different fabrics, such as stiffness, elasticity, and friction, that affect the ability of successfully executing a learned skill.

During my PhD, I aim to address two major challenges: (1) developing a robust and efficient representation for cloth-like objects that enables reasoning on its configuration space on both the high and the low levels and (2) implementing methods for grasping and manipulating these objects to achieve desired configurations based on the robot reasoning of its configuration space. The goal of my research is to draw on concepts from Mathematics, Physics, and Machine Learning to create robust and efficient manipulation methods for folding, unfolding, flattening, (un)stacking and other daily tasks that are more explainable and robust than current approaches.