Speaker:  Karen Liu, School of Interactive Computing College of Computing GeorgiaTech

URL for Speaker:  https://www.cc.gatech.edu/~karenliu/Home.html

Location:  ESB 2012

Intended Audience:  Public

Leveraging physical contacts to interact with our surroundings is an essential skill to achieve any physical task, but contact-rich, dynamically changing environments often create significant challenges to autonomous robotic locomotion and manipulation. Unexpected slippage or loss of contact can cause a balance controller to fail during locomotion, incidental contacts with unseen obstacles can disrupt a manipulator during a pick-and-place task, and large impulse induced by contacts can result in irreparable damage to the robot hardware. While there exists computationally tractable contact models to aid the development of robust contro policies, the discontinuities inherent in the contact phenomenon introduce non-differentiability in the equations of motion, rendering traditional approaches to optimal control ineffective. In this talk, I will show that, with intelligent contact control and planning algorithms, the challenge of handling contact can become a solution. The first part of the talk focuses on a model-based approach to controlling a deformable robot for locomotion. The control algorithm leverages both static and dynamic contact friction by solving an optimization with non-differentiable linear complementarity constraints efficiently. The second part of the talk focuses on a model-free reinforcement learning approach to minimizing the damage of humanoid falls. We formulate the control problem as a Markov Decision Process that solves for a contact sequence with the ground such that the maximal impulse incurred during the fall is minimized. Lastly, I will mention some work we have done in the area of data-driven haptic perception for robot-assisted dressing tasks.