Brian Bittner

Ph.D Pre-candidate
Robotics

Education


B.S. Mechanical Engineering, Carnegie Mellon University, 2016

I am a second year Robotics Ph.D. student at the University of Michigan. I'm interested in reduced-order control architectures. What structure in a system's dynamics or geometric composition might enable sensible ways to build up mechanical complexity while maintaining physically intuitive control schemes?

My current work involves development of a geometric gait optimization algorithm. A robot learns its local mechanical connection while being controlled about an input gait.

Building this local connection (about the gait) gives the robot the physical information it needs to improve its performance. Obtaining a global mechanical connection would require exhaustively testing the robot in all possible configurations. Building a local linearized connection about a gait provides exponential experimental savings as mechanical complexity increases.

This type of scalability allows us to apply the technique on complex robotic platforms, such as this 9-link robotic snake. The snake takes 30 strokes given an input gait, builds a local connection, and finds a gait adjustment which provides a higher efficiency gait. The final gait provides a locally optimal gait for energetically-efficient swimming.


This work is strongly aided by insights from our collaborators at the LRAM at Oregon State.