Motion-Aware Adaptive UWB and IMU Fusion for Infrastructure Free Following Robot Localization

Abstract

Reliable relative localization is essential for following robots in indoor environments where satellite navigation is unavailable and external positioning infrastructure is impractical. UWB ranging provides accurate distance information, while IMUs capture continuous target motion; however, UWB measurements suffer from multipath effects, body blockage, and motion-induced fluctuations, and inertial estimates drift over time. Existing methods often rely on fixed environmental anchors or filtering parameters, limiting their robustness in dynamic following scenarios. We present a motion-aware adaptive UWB and IMU fusion method for infrastructure-free following robot localization. Two UWB anchors are mounted on the robot, while the target carries a UWB tag and an IMU. IMU-derived motion intensity adaptively adjusts the wavelet-based UWB denoising threshold, and the denoised ranges are converted into relative positions through a dual-anchor geometric model. An adaptive Kalman filter then fuses UWB observations and IMU motion by adjusting measurement covariance according to short-term ranging stability. Experiments under different motion trajectories show that the proposed method suppresses ranging fluctuations, smooths trajectories, and reduces error peaks. It achieves an overall RMSE of 0.182 m, compared with 0.621 m, 0.494 m, and 0.418 m for Raw UWB, Fixed Wavelet UWB, and Fixed KF, respectively. These results demonstrate the effectiveness of the proposed adaptive denoising and covariance-based fusion strategy for infrastructure-free following robot localization.