Boston University
Biography
Pierre E. Dupont received the B.S., M.S. and Ph.D. degrees
in mechanical engineering from Rensselaer Polytechnic Institute
in 1982, 1984 and 1988, respectively, with a specialization in
robot kinematics. From 1988 to 1990, he was a Postdoctoral Fellow
at Harvard University with the Center for Intelligent Control
Systems. In 1990, he joined the Aerospace and Mechanical Engineering
Department at Boston University, where he is currently an Associate
Professor. His research interests include manipulation, system
modeling, model reduction and control theory.
Abstract
In this talk, our framework for machine perception is applied to
modeling geometric properties of objects manipulated by a telerobotic
system. In the context of our perception paradigm, an integrated solution
approach to the two subproblems of data segmentation and property
estimation is presented. The theme of this modeling approach is to combine
physics-based models of object contact states with statistics-based estimation
techniques to account for sensor noise and model uncertainty. The
physical contact models describe mathematically the possible interactions
between the robot and a manipulated object, or between one object and
another. The contact models are functions of the robot sensor data and
the properties to be estimated. A task description, based on a network
of contact states and permissible transitions, is assumed given. A
multiple model estimation approach is employed for data segmentation. In
this approach, the contact state network is used to select a set of
possible contact states at each time step. Property estimates are
then computed for each contact state model in the set. The statistical
properties of these estimates determine which contact state(s) is
(are) valid. As an illustration, the successful estimation of local
geometric properties is described for 3-D peg insertion on a tabletop
teleoperator system. In this case, the robot sensor data consists of
joint angles and the contact models correspond to kinematic constraint equations.
The properties estimated are the peg and hole dimensions as well as
the hole location.
Related Publications
T. Debus, P. Dupont and R. Howe "Automatic Identification of Local Geometric Properties During Teleoperation," Proceedings of the 2000 IEEE International Conference on Robotics and Automation, San Francisco, April 22-28, to appear.
pdf
Debus, T., Dupont, P., and Howe, R. "Automatic Property Identification via Parameterized Constraints." Proceedings of the IEEE International Conference on Robotics and Automation, Detroit, MI, May 1999.
pdf
Dupont, P., Schulteis, T., and Howe, R. "Experimental Identification of Kinematic Constraints." Proceedings of the IEEE International Conference on Robotics and Automation, Albuquerque, New Mexico, April, 2677-2682.
pdf
Links
Pierre Dupont's web site
Contact
pierre@bu.edu