Magnetometer Aided Navigation Filters for Improved Observability and Estimation on Ground Vehicles

Abstract

Two modified extended Kalman filters, called the MGVA and MVVA filters, have been developed which combine the attitude determination capabilities of magnetometers with the standard extended Kalman filter. The standard extended Kalman filter is known to experience observability problems during driving which does not provide enough excitation to its sensors. The MGVA and MVVA filters attempt to remedy these problems by providing an attitude solution to the filter, via a process that utilizes magnetometer measurements. In order to show the effectiveness of the MGVA and MVVA filters, the filters were tested both in simulation, and experimentally for low excitation trajectories. It is shown in simulation that the two filters are able to improve attitude and accelerometer bias estimation in situations where the standard filter experiences estimation error. In addition to this, it is shown that both modified filters increase the observability rank of the system such that it is full rank. Experimental testing of the two modified filters reveals that both the MGVA and MVVA filters provide improved attitude and accelerometer bias estimation during low-excitation trajectories.

The MGVA and MVVA filters are also compared to a heading constrained filter that has been studied in the past. It is shown that the heading constrained filter produces more accurate estimates of heading than both of the modified filters. As a result, a further modification is implemented on both the MGVA and MVVA filters, and improved performance is shown for the MGVA filter. Finally, the MGVA and MVVA filters were tested on both simulated and experimental dynamic trajectories. This was done in order to test whether or not the filters are useful under typical trajectories that a ground vehicle might drive, not just low excitation trajectories. It is shown that under these circumstances the MGVA filter is suffers from the increased vehicle dynamics. In contrast, it is also shown that the MVVA filter is, once again, able to estimate attitude and accelerometer bias well. It is concluded from these tests that the MVVA filter is, in general preferable to the MGVA filter for more dynamic trajectories.