         
|
GRASP Lab Seminar 2004-2005
November 5, 11:00 AM, Levine Hall 307.
Noah Cowan
Department of Mechanical Engineering, Johns Hopkins
Sensor-Based Locomotion Control in Animals and Machines.
Abstract: Animals execute split-second maneuvers to avoid obstacles, catch prey and evade predators amidst myriad information from thousands of sensors such as vision and touch. We tackle the challenge of closing sensory feedback loops in robotics by deriving inspiration from a variety of natural sensor-based control systems. First, we observed cockroaches, Periplaneta americana, executing a high speed wall following behavior in which their antennae serves as the primary feedback sensors. Based on our observations, we constructed a dynamical model of this behavior and designed a new class of tactile robotic sensors, similar in structure and function to cockroach antennae. Our tactile sensors enable several approaches to robotic wall following and "tactile map" building which will be presented. More recently, we modeled the closed-loop, whole-animal sensorimotor performance in weakly electric fish. Using specialized electrosensory feedback, the fish attempt to remain centered within a pen-plotter driven "shuttle" with two parallel walls. The shuttle is driven by a function generator, enabling us to measure each fish's closed-loop "Bode" response to sinusoidal shuttle motions. Assuming a second order model, the response appeared to be either critically- or slightly over-damped. The next step will be to investigate how these fish fuse sensory feedback from vision, tactile sensing and electrical sensation.
Biography: Noah Cowan received his Ph.D. in 2001 from
the University of Michigan, studying vision-based control. Subsequently,
he completed a postdoc at UC Berkeley, where he studied biomechanics and
control in cockroach locomotion, as well as machine learning. Last year,
Dr. Cowan joined the Mechanical Engineering Department at the Johns Hopkins
University as an Assistant Professor. His research interests include sensor-based
locomotion control, biomechanics and computational neuroscience. For more
information, please visit his website: http://limbs.me.jhu.edu/~ncowan/
