Exploiting Dynamics with Whole-Body Control for Stability and Mobility

27 May
Friday, 05/27/2011 10:30am to 12:30pm
Ph.D. Dissertation Proposal Defense

Scott Kuindersma

Computer Science Building, Room 151

I propose a research program to develop solutions to novel control problems in mobile manipulation. In particular, I intend to focus on the problem of upper body coordination for maintaining stability using the uBot, a dynamically balancing mobile manipulator. This problem is challenging from a control perspective because the system is nonlinear, underactuated, high-dimensional, and successful policies are likely to be nonlinear. Furthermore, the problem is interesting from a behavioral perspective since the recruitment of multiple bodily resources to meet critical mobility objectives such as balance and fall recovery is a ubiquitous behavioral strategy in nature.

The biomechanics community has recently made progress toward understanding the role of upper extremities in human stability maintenance, but comparatively little work has been done exploring this type of control in humanoid robots. I intend to advance the state of the art by applying modern machine learning and optimal control techniques and insights from the study of human motion to develop and analyze this type of control on a robot. In turn, I expect a computational implementation may lead to new insights into nonlinear control algorithms and contribute to the understanding of this type of behavior in animals.

I present initial results that demonstrate the mechanical feasibility of whole-body stability control in the face of significant impacts using the uBot-5. I also describe a simulated model of the system and provide initial results demonstrating stabilizing arm motions discovered via direct numerical optimal control. Future experiments on both the uBot and NASA Robonaut 2 platform are discussed.

Advisors: Roderic Grupen & Andrew Barto