Adaptive and Robust Kalman Filter-Based Fusion of IMU and UWB for High-Accuracy Localization

To address the accumulated drift in inertial measurement unit (IMU)-based localization and the instability of ultra-wideband (UWB) measurements caused by occlusion and multipath effects in dynamic environments, this paper proposes a fusion localization method based on an Adaptive and Robust Kalman Filter (ARKEF). The proposed method enhances the traditional Extended Kalman Filter (EKF) framework by introducing robust estimation and noise adaptation mechanisms. Specifically, the Huber loss function is employed to downweight outlier UWB measurements, while the observation noise covariance is adaptively adjusted based on residual statistics, thereby improving the system’s robustness to sensor uncertainty and anomalous data. Experimental validation begins with standalone IMU and UWB positioning to quantify their individual error characteristics. Subsequently, both standard EKF and the proposed ARKEF are implemented and evaluated under dynamic trajectories. Results show that the ARKEF achieves superior 3D localization performance, reducing the mean error to 0.151 m, compared to pure IMU (8.379 m), standalone UWB (0.488 m), and conventional EKF-based fusion (0.222 m), representing an accuracy improvement of approximately 32%. Furthermore, ARKEF outperforms other methods in terms of maximum error, root mean square error (RMSE), median error, and 95th percentile error. Axis-wise analysis demonstrates that the ARKEF significantly suppresses drift and jitter in the X, Y, and Z directions, enhancing the overall system stability and robustness, with promising potential for practical deployment.