University of Minnesota
University Relations
http://www.umn.edu/urelate
612-624-6868
myU OneStop


Go to unit's home.

Home | Seminars and Symposia | Past seminars/symposia: Wednesday, February 11, 2004

DTC Seminar Series

Active Sensing for Terrain Classification on an Agile Robot

by

Richard Voyles
Computer Science and Engineering
University of Minnesota

Wednesday, February 11, 2004
1:00 pm

402 Walter Library

Richard Voyles

Download slides (pdf 2.6 MB) Robotic applications in urban search-and-rescue and planetary exploration involve unknown terrains with unknown characteristics. Both require a high degree of adaptability for successful locomotion. But before one can adapt, information must be gleaned from the environment to drive the adaptation. A novel terrain classification strategy for our legged robot, "TerminatorBot," to drive adaptation of the gait to terrain conditions has been developed. Using a forward-looking camera, a measure of "gait bounce" that encodes the characteristics of interaction between limb and terrain during locomotion is derived. Based on the gait bounce signal, the current terrain upon which the robot is crawling is classified from a small set of previously observed terrains. The classification scheme is a two-tiered technique based on discriminant analysis and hidden Markov models. The gait cycle of the robot is controlled during locomotion to segment the sensory data into "epochs" to make use of the fact. These epochs represent the lifting of the body, the forward motion, and the retraction of the legs, which are classified individually using discriminant analysis. Since these classifications are neither exact, nor independent, discrete hidden Markov models on the resulting symbols to improve classification accuracy and robustness are used. Using this meta-classifier, Professor Voyles's students have achieved terrain classification rates averaging 84% on five man-made terrains over just one gait cycle.

Richard Voyles presenting

This work is part of a large body of work that will form the basis of a proposed NSF Center on Safety, Security, and Rescue Research. Professor Voyles will preface this talk with a description of the vision of this proposed Center and its operation. He will then describe the terrain classification project described above.

 

Richard Voyles received his Ph.D. in the School of Computer Science at Carnegie Mellon University in 1997. Currently, he is assistant professor in Computer Science and Engineering at the University of Minnesota and an associate member of the graduate faculties of Electrical Engineering, Computer Engineering and Software Engineering. His primary research interests are in the fields of robotics and multi-agent systems. Specifically, he works on miniature, resource-constrained robots, multi-agent teams of robots, mobile manipulators, programming robots by human demonstration, skill learning, and haptic sensing and display.