Geometric Extended Kalman Filter With Dual Robust Kernels for Integrated Navigation

Abstract

Accurate and robust state estimation, especially for attitude, velocity, and position, is crucial for robotics and autonomous navigation systems. This paper introduces a novel robust navigation framework combining a geometric extended Kalman filter (GEKF) with dual robust kernels (DRK). The GEKF reduces process model uncertainty by embedding navigation states into an extended special Euclidean group, addressing frame mismatch in error definition, and deriving a process model weakly dependent on attitude. The DRK technique enhances measurement update robustness by integrating convex and non-convex kernels, with the convex kernel ensuring fast convergence and the non-convex kernel suppressing outliers. Experimental results of the fusion between inertial navigation system (INS) and global navigation satellite system (GNSS) demonstrate that the proposed framework outperforms advanced methods in convergence speed, estimation accuracy, and robustness. This approach provides an innovative solution to improving navigation performance in complex environments.