This dissertation takes a step to increase the abilities of robots to manipulate deformable objects, like ropes and cables. I focus on two problems, our inability to accurately predict in all scenarios how a DOO will move or deform, and the cost of collecting training data in robotic manipulation. I present a system that can quickly learn to manipulation the DOO despite limited data and imperfect predictions, as well as plan grasps and regrasps when the robot gets stuck.

Chair: Dmitry Berenson
Meeting password: rope