Gradient-Type Minimization Methods for Initializing State Variables in GPS/Ins Integration

Abstract

A common problem in the filter-based processing of GPS/INS data is incorrect initial state and covariance estimates in the time before filter convergence. These problems are especially acute when the initialization time is relatively short – say, less than a minute. Moreover, lower quality inertial systems have large initial angular errors that are often outside the linearity assumptions in the linear filter methods. In the past, means to compensate for this have been to try and improve the angular models for large initial errors, or to use adaptive filtering methods. This paper describes an alternative to linear filtering methods for processing a combination of IMU data and GPS data when there is only a limited time to estimate state variables. We present experimental data from minivan tests and from a parachute drop.