Robust State and Fault Estimation in Mobile Robots under Dynamic Noise Environments using Hybrid LSTM-EKF with Adaptive Weighting

This paper proposes an adaptive hybrid estimation framework combining a Long Short-Term Memory (LSTM) network and an Extended Kalman Filter (EKF) for simultaneous robot state estimation and actuator fault detection in mobile robots operating under uncertain conditions. The LSTM network learns temporal patterns and nonlinear relationships from sensor data, while the EKF provides model-based filtering with uncertainty quantification. A novel adaptive fusion mechanism dynamically balances these complementary approaches using a weighting factor derived from the estimation uncertainties of both components. The proposed method was extensively evaluated on a differential-drive robot model subject to various noise conditions and fault profiles, including progressive drift faults and abrupt jump faults. Simulation results demonstrate that our hybrid approach significantly outperforms standalone EKF and LSTM estimators, achieving up to 74.6% improvement in fault estimation accuracy and 34.5% reduction in position error under challenging high-noise conditions. The framework maintains consistent performance across diverse fault types, showing particular effectiveness in detecting gradual fault progression while remaining responsive to sudden fault events. These findings confirm that the adaptive LSTM-EKF fusion provides enhanced accuracy, robustness, and generalization capability compared to conventional approaches.