Hybrid extended Kalman filter long short-term memory framework for robust state and fault estimation in mobile robots under unknown disturbances

Abstract

Introduction. Reliable and accurate state estimation plays a central role in mobile robotics, ensuring effective localization, navigation, and control in uncertain and dynamic environments. Traditional estimation methods such as the extended Kalman filter (EKF) and the unscented Kalman filter (UKF) are widely used for nonlinear systems; however, their performance degrades when facing unknown disturbances or modeling inaccuracies. Problem. In real-world mobile robots, unexpected motor faults and unmeasured disturbances significantly reduce the estimation accuracy and may lead to mission failure. Classical EKF and UKF approaches rely on static models and Gaussian noise assumptions, which make them unsuitable for systems affected by unknown or time-varying uncertainties. The goal of this work is to design and validate a hybrid extended Kalman filter long short-term memory (EKF–LSTM) framework capable of achieving joint state and fault estimation for mobile robots operating under unknown disturbances. Methodology. The proposed approach combines a model-based EKF with an offline-trained LSTM neural network. The EKF performs nonlinear state estimation using physical robot dynamics and noisy Global Positioning System (GPS) measurements, while the LSTM predicts additive motor faults based on temporal data. The LSTM outputs are incorporated into the EKF as pseudo-measurements with adaptive covariance tuning, ensuring stability and robustness. Results. Simulation results demonstrate that the hybrid EKF–LSTM reduces trajectory root mean square error (RMSE) by 4.6 % and fault RMSE by 68 % compared to a standalone EKF, and by more than 50 % compared to the UKF. The framework effectively tracks abrupt fault variations and remains resilient to unknown inputs and sensor noise. Scientific novelty. Unlike existing hybrid filters, the proposed method introduces adaptive covariance fusion between EKF and LSTM estimators, enabling reliable operation under directional dynamics and unmodeled disturbances. Practical value. The proposed hybrid EKF–LSTM framework enhances fault-tolerant localization for autonomous robots, providing a scalable solution for real-time applications such as search-and-rescue operations, industrial automation, and autonomous navigation in noisy or GPS-denied environments. References 33, tables 2, figures 5.