Stochastic Systems Group  

Dr. Michael Schneider  Alphatech, Inc.
This talk describes some techniques for addressing the problem of controlling a sensor to collect data so as to estimate the states of objects, which are dynamically changing. The techniques described include both exact and approximate ones. In particular, classical results on multiarmed bandit scheduling characterize the optimal control for some special infinitehorizon problems. An extension of these results we have developed for finitehorizon problems is overviewed in the talk. Also included will be a discussion of the applicability of rollout approximate dynamic programming techniques to sensor management. Such approach will be illustrated with an example sensor management problem from airtoground surveillance.
Associated conference papers:
M. K. Schneider, G. L. Mealy, and F. M. Pait. Closing the loop in sensor
fusion systems: Stochastic dynamic programming approaches. In Proceedings of
the American Control Conference, 2004
R. B. Washburn, M. K. Schneider, and J. Fox. Stochastic Dynamic
Programming Based Approaches to Sensor Resource Management. In Proceedings
of the International Conference on Information Fusion, Annapolis, 2002.
Problems with this site should be emailed to jonesb@mit.edu