Robust Error State Sage-Husa Adaptive Kalman Filter for UWB Localization

Abstract

Given the sensors’ path and interference mitigation capabilities, ultra-wideband (UWB)-based positioning systems have demonstrated high accuracy and reliability. This work aims to improve the Sage-Husa fuzzy adaptive filter (SHFAF) proposed in previous works by modifying the motion model to a 3-D ground-based differential drive robot using IMU and wheel encoder kinematic fused control inputs. In addition to the changed motion model kinematics, this article improved the positive definite constraint on P and R during dynamic estimations, thus making the filter more robust to outliers. An improvement to the computation and derivation of the fuzzy logic system for the SHFAF based on the adaptive neuro-fuzzy inference system (ANFIS) structure was developed, and training the fuzzy system using gradient descent was applied to improve the system’s accuracy. Experimental validation was conducted using real-world data from a Clearpath Jackal robot equipped with Qorvo UWB sensors and static nodes. Regarding localization accuracy, the proposed velocity-based SHFAF (VelSHFAF) system outperformed the previous SHFAF implementation by approximately 30%–25% across two test courses, demonstrating its enhanced performance and reliability.