Modeling and Information Processing

TAP - Advanced Particulate Techniques

Objective

To design and evaluate robust non-linear particle filtering methods for navigation.

Onera's contribution

In certain contexts in which the measurement is strongly non-linear, traditional filters like the Kalman filter are ineffective. Non-linear filters, called particle filters, estimate the state via an approximation of the state's density (the target's candidate positions) on the basis of Monte Carlo samplings. These filters are particularly attractive because they are not very sensitive to the state's dimension. However, the academic versions of this filter are not very robust. Since 2000, Onera has been developing particle filter algorithms robust to measurement noise and the modeling of the dynamics. These new filters represent the state's density by an adapted functional decomposition. They have been theoretically validated (convergence) and the algorithms have been accelerated in order to make them useable in real time. They have been applied to navigation by correlation of terrain (altimetry, gravimetry) and to GPS (Global Positioning System) in severe contexts where traditional methods are inoperative.

Illustrations

• Resetting navigation
• Trajectory calculation
• GPS reception

Last Update: 11 October 2006 - © ONERA 2009 - Terms of use