2009 Meetings

December 15, 2009: "Automated Antenna Design and Optimization" by Dr. Jason Lohn, Carnegie Mellon University West

Abstract

Current methods of designing and optimizing antennas by hand are time and labor intensive, limit complexity, and require significant expertise and experience. Evolutionary design techniques can overcome these limitations by searching the design space and automatically finding effective solutions that would ordinarily not be found. In this talk, we present automated antenna design and optimization methods based on evolutionary algorithms. We present evolved antennas for a variety of aerospace applications, focusing on a project that produced antennas that flew on NASA's Space Technology 5 (ST5) mission. We discuss the software tools we developed to automate the design of these evolved antennas which are the first ever artificially-evolved objects to fly in space.

Biography

Dr. Lohn is a Sr. Systems Scientist at Carnegie Mellon, Silicon Valley Campus, and recently co-founded a startup to commercialize his automated antenna design technology. Previously he led Evolvable Systems research at NASA Ames Research Center, worked at Google, held a Visiting Scholar appointment in the Computer Science Department at Stanford University, and worked at IBM. He received his MS and PhD in Electrical Engineering from the University of Maryland at College Park, and his BS in Electrical Engineering from Lehigh University. He has over 50 technical publications and his work has been featured in Wired magazine, MIT Tech Review, and Popular Science. Dr. Lohn is a member of the IEEE, ACM, and Sigma Xi. He was a co-founder and co-chair of six NASA/DoD Conferences on Evolvable Hardware, and serves as an Associate Editor of IEEE Transactions on Evolutionary Computation.


July 30, 2009: "A Simulated Annealing and Gain Scheduling Approach" by Dr. Charles Jorgensen, NASA Ames Research Center

Abstract

There are many dimensions across which human communication occurs. The NASA Neuro Engineering Laboratory has been involved in basic research studying alternatives to standard auditory communication for both pilots and astronauts. This talk will present two such approaches one involving communication directly from the human nervous system called subvocal speech and the other a new multimodal technique for graphically understanding content changes in analog signal sources. Short demonstration film clips will be presented showing non vocal robotic control, high noise environment communication, and fault recognition of sensor data from the space shuttle and Ares 1 launch vehicle.
For more information visit: http://ti.arc.nasa.gov/people/hornby/evo_gaits/evo_gaits.html

Biography

Dr. Jorgensen received his Ph.D. in Mathematical Psychology from the University of Colorado Boulder in 1973. He is a senior IEEE member who has worked in academia, industry, and government including Carnegie Mellon, the US Army Research Institute, Martin Marietta, and Oak Ridge National Laboratories. He has been with NASA Ames Research Center since 1990 and is presently Chief Scientist of the NASA Neuro Engineering Laboratory. Dr. Jorgensen has over 180 publications and over 19 patents in aeronautics, neural engineering, and robotics. He is recipient of numerous awards including the American Nuclear Society and Department of Defense. He holds three NASA scientific medals including the Outstanding Engineering Achievement Medal in 1995 for neural network Controllers, NASA Exceptional Achievement medal for his work in Aeronautics in 1998, and the NASA Exceptional Service Medal for outstanding contributions to neural computing in 2001. His 2006 research on subvocal speech was a world finalist for the Saatchi & Saatchi prize for world changing ideas. His current research interests include electromyography machine interfaces, computing architectures based on biological dynamical systems, biometrics, and graphic visualization.


April 8, 2009: "Using an Evolutionary Algorithm to Evolve Dynamic Gaits for Sony's Aibo" by Dr. Greg Hornby, Carnegie Mellon Silicon Valley at Moffett Field

Abstract

A challenging task that must be accomplished for every legged robot is creating the walking and running behaviors needed for it to move. In this talk I will present the system for autonomously evolving dynamic gaits used for two of Sony's quadruped robots. The developed Evolutionary Algorithm runs onboard the robot and uses the robot's sensors to compute the quality of a gait without assistance from the experimenter. First I will show the evolution of a pace and trot gait on the OPEN-R prototype robot. With the fastest gait, the robot moves at over 10m/min., which is more than forty body-lengths/min. While these first gaits are somewhat sensitive to the robot and environment in which they are evolved, I then show the evolution of robust dynamic gaits, one of which was used on the ERS-110, the first consumer version of AIBO.

Biography

Greg Hornby was a visiting researcher at Sony's Digital Creatures Laboratory from 1998 to 1999, received his Ph.D. in Computer Science from Brandeis University in 2002, and is currently a Senior Scientist with UC Santa Cruz at NASA Ames Research Center. His interests include Evolutionary Algorithms (EAs), Computer-Automated Design, Generative Representations and Complexity Measures. In addition to creating the EA that evolved antenna for the ST-5 spacecraft, he created the EA that autonomously evolved the dynamic gait used on Sony's AIBO robot and also developed the Age-Layered Population Structure (ALPS) EA for reducing the problem of premature convergence.


March 11, 2009: "Control on Landscapes with Local Minima and Flat Regions: A Simulated Annealing and Gain Scheduling Approach" by Dr. Abe Ishihara, Carnegie Mellon University, Silicon Valley

Abstract

Convergence of the backpropagation algorithm and its variants highly depend on the shape of the landscape. It is well known that multilayer neural networks employing sigmoid-like nonlinearities in the hidden layers result in landscapes with local minima. Another type of local minima, in which we term "flat regions" arise due to the fact that the derivative of sigmoid like activation functions tends exponentially to zero. In this talk we propose a new control method to deal with these issues. This method employs simulated annealing and a gain-scheduled learning rate. Using a Ito calculus, we are able to prove that this method results in a closed loop system that is semi-globally uniformly bounded in expected value. Furthermore, we illustrate on a simple example how the proposed method escapes local minima and flat regions, whereas a conventional neural network controller gets "stuck". Extensions to the control of the multi-link robotic manipulator are discussed and simulations are presented.

Biography

Dr. Ishihara is currently a research scientist at Carnegie Mellon Silicon Valley. He received the B.S. degree in Electrical Engineering from Rensselaer Polytechnic Institute, and the M.S. and Ph.D. degrees in Aeronautics and Astronautics from Stanford University. Prior to joining Carnegie Mellon, he worked on the adaptive control of drug delivery systems, reconfigurable flight control, and neural network control of robot manipulators. Additionally, he developed techniques for measuring human motor learning in unknown environments using surface electromyography (EMG). He was the recipient of the NIH National Research Service Award (NRSA) training grant. His current research interests include simulated annealing for nonlinear control and stability margin estimation for adaptive flight control.