GPS-INS state estimation for multi-robot systems with computational resource constraints

Abstract

A decoupled Kalman Filter for GPS-INS sensor fusion is developed for a high-speed multi-robot system with computational resource constraints. An eighth-order filter describing system and bias dynamics is decoupled into four second-order filters. Process and measurement noise statistics and first-order bias dynamics are derived from experimental data. The decoupled filter reduces computation time by a factor of seven over the coupled filter, enabling real-time implementation on an inexpensive processor at the required control update rate of 20 Hz. The decoupled filter is evaluated through simulation and experiments and provides sub-meter position error for over a minute, an order of magnitude improvement over GPS alone.