Manipulation Skills that I Wish My Robots Had
Prof. Alberto Rodriguez
Department of Mechanical Engineering, M.I.T.
My main goal in this talk is to discuss capabilities that I believe are essential to develop autonomous robotic manipulation systems, systems with the ability to understand, control and embrace [frictional] contact forces. To describe the challenges of planning and control through friction, I’ll introduce the pusher-slider system, a fundamental problem in manipulation analogous to the inverted pendulum in classical controls, subject to underactuated, hybrid, and noisy dynamics. I’ll demonstrate real-time solutions by combining quasi-static approximations of the dynamics, a model predictive control framework, and mixed-integer programming. I’ll then show how this understanding of frictional dynamics can be applied to the problem of controlling the pose of a grasped object by exploiting contacts with the environment, in the context of assembly automation. Finally, in a more unstructured task, I’ll brief on recent work by Team MIT-Princeton in the Amazon Robotics Challenge to develop a robotic pick-and-place system capable of grasping and recognizing novel objects in cluttered environments.