Robust State Estimation in Atomic Inertial Navigation Systems: A Comparative Study of CI-RKF and MKMC-EKF Filters

Abstract Atomic inertial sensors offer exceptional long-term stability and ultra-low bias drift, but maintaining consistent navigation performance in dynamic environments remains challenging owing to non-Gaussian noise, sensor nonlinearities, and unmodeled dynamics. This study presents INS-ATOMIC, an autonomous inertial navigation system based solely on atomic gyroscopes and accelerometers, without the need for a global navigation satellite system or external corrections. Two filtering techniques are evaluated: (1) the chi-square increment robust Kalman filter (CI-RKF), which rejects inconsistent measurements, and (2) the multi-kernel maximum correntropy extended Kalman filter (MKMC-EKF), which mitigates non-Gaussian noise using adaptive kernel processing. Both methods were tested on real-world vehicular data sets under dynamic conditions. The MKMC-EKF outperformed the CI-RKF in accuracy and stability, achieving root-mean-square-error improvements from 37% to over 99% in velocity and up to 99% in acceleration and angular rates. These results demonstrate the effectiveness of correntropy-based filtering with atomic sensors for accurate, GNSS-free navigation in challenging environments.