Robotic Manipulation with Physics-Based Skills

Mehmet Dogar - Hiring Candidate - Research Associate, Computer Science and Artificial intelligence Lab, Distributed Robotics Group, MI

March 10, 2015, 10 a.m. - March 10, 2015, 11:30 a.m.

McConnell 437

There are striking differences between human manipulation capabilities and state-of-the-art robotic manipulation. First, we humans are capable of a wide variety of manipulation skills, e.g. pushing, pulling, throwing, and tumbling the objects around us, whereas robotic manipulation is extremely limited, mostly to grasping actions. Second, we humans can plan and execute long sequences of collaborative manipulation actions to reach a high-level goal, such as assembling a piece of furniture, whereas robots are limited in their planning and execution capabilities constraining them to short sequences of simple pick-and-place operations. In this talk I will present my research addressing these limitations of robotic manipulation. First, I will focus on using physics-based reasoning to add to a robot’s repertoire of manipulation skills. Particularly, I will talk about using physics-based pushing actions. I will present an algorithm to compute the "capture region" of an object, which represents the set of object poses that a pushing action can funnel into the hand. I will then present another algorithm which uses these capture regions to plan push-grasping actions in cluttered environments. I will show that these physics-based primitives enables the robot to perform manipulation under high uncertainty and clutter. Second, I will talk about planning long sequences of collaborative manipulation actions for multi-robot assembly operations. I will formulate the question as a constraint satisfaction problem, show that naive solutions are computationally intractable, and present a planning algorithm which uses “regrasping” actions to divide the constraint graph into smaller components. The algorithm enables a team of robots to plan long sequences of collaborative assembly operations. I will conclude by outlining my future plan on how to make robots more capable and intelligent manipulators of the physical world. Mehmet Dogar is a Postdoctoral Researcher at the Distributed Robotics Group at MIT CSAIL. His research focuses on using physics-based predictions in robotic manipulation which enables robots to accomplish useful tasks in dynamic and cluttered human environments. He received his Ph.D. from the Robotics Institute at Carnegie Mellon University in August 2013. He received his B.S. degree in Computer Engineering from the Middle East Technical University, Turkey. ALL ARE WELCOME