Abstract:
This paper addresses the robust estimation of a covariance matrix to express
uncertainty when fusing information from multiple sensors. This is a problem of
interest in multiple domains and applications, namely, in robotics. This paper
discusses the use of estimators using explicit measurements from the sensors
involved versus estimators using only covariance estimates from the sensor
models and navigation systems. Covariance intersection and a class of orthogonal
Gnanadesikan-Kettenring estimators are compared using the 2-norm of the
estimates. A Monte Carlo simulation of a typical mapping experiment leads to
conclude that covariance estimation systems with a hybrid of the two estimators
may yield the best results.
