Land vehicle navigation filtering for a GPS/dead-reckoning system

This paper addresses the implementation of a cascade of Kalman filters for a Dead-Reckoning (DR) navigation system. The DR navigation solution, i.e., heading and position, is normally obtained from a magnetic compass and wheel encoders providing direction and distance traveled respectively. Generated directly from the outputs of these sensors, the DR navigation solution would be corrupted with errors associated with the compass, e.g., installation and output transients, and with the wheel encoders, e.g., variation in tire pressures and wheel slippage. Improvements to the short term DR navigation data is achieved by implementing a compass pre-filtering algorithm that smooths through transients in the compass data using the compass heading and the output differences from dual wheel encoders as observations in this pre-filter. The smoothed compass data and the dual wheel encoders' data are then used to generate the DR navigation solution. Long term improvements to the DR navigation solution are achieved by estimating errors associated with the compass' installation and the vehicle's wheel encoders by implementing a navigation filter algorithm including compass bore-sight and wheel scale-factor errors using GPS positions as observations and using the compass pre-filter's estimate of vehicle heading rate to tune the navigation filter. This tuning provides a means to negate the navigation filter algorithm's tendency to attribute errors to bore-sight and scale-factor errors, errors that actually result from wheel slippage. The outputs of the navigation filter are then used to correct the DR navigation solution in an open-loop architecture. Actual recorded data is used to illustrate the benefits of the approach described above.