The integration between Inertial Navigation Systems (INS) and Global Positioning System (GPS) based on Kalman Filter (KF) has been widely used. In INS/GPS systems, KE utilizes linearized dynamic models for INS errors modeling and GPS measurements. If Low-Cost MEMS-based inertial sensors with complex stochastic error nonlinearity are used, performance degrades significantly during short periods of GPS-outages due to the approximation introduced in the linearized INS errors dynamic model. This paper proposes a nonlinear data-driven INS-errors modeling based on Gaussian Process Regression (GPR). During reliable GPS availability, the correct vehicle state, sensors measurements, and INS output deviations from GPS measurements are collected. During GPS-outages, GPR is applied to recently collected data set to predict INS deviations from GPS. The predicted INS deviations are then fed to KF as a virtual update to estimate all INS errors. The proposed technique was tested with a low-cost Reduced Inertial Sensors System (RISS) for land-vehicles in which the vehicle odometer is used along with inertial sensors. Real road experiments on two different trajectories showed significant improvements during long GPS-outages.

Additional Metadata
Conference Institute of Navigation International Technical Meeting 2012, ITM 2012
Atia, M, Noureldin, A. (Aboelmagd), & Korenberg, M. (Michael). (2012). Enhanced kalman filter for RISS/GPS integrated navigation using gaussian process regression. In Institute of Navigation International Technical Meeting 2012, ITM 2012 (pp. 1148–1156).