Augmented Kalman Filter Design in a Localization System Using Onboard Sensors With Intrinsic Delays

Abstract

This paper presents a novel approach to address the limitations of the conventional Augmented Kalman Filter (AKF) when only on-board incremental sensors are used, which is a more realistic strategy for diverse robot configurations. The proposed AKF focuses on an online calibration of odometric parameters based on velocity measurements. Moreover, it shows the unsuitability of the AKF when data is not perfectly synchronized and introduces a real-time synchronization scheme to enhance the accuracy of parameter adjustments. The effectiveness of the method is demonstrated using an autonomous wheelchair equipped with encoders, a LIDAR and an IMU. A pre-module is added to synchronize LIDAR and IMU signals with respect to the encoders. The lags are included in the state vector and updated based on their reliability to synchronize the data in real time, which addresses the problem of real sensors where the delay period can change over time. The system also incorporates a fuzzy controller that designs a variable filter covariance inversely proportional to measurement reliability, enabling a dynamic parameter adjustment. The experiments show that the proposed localization outperforms the Extended Kalman Filter and the new AKF results provide better parameter estimation compared to when the implementation improvements are not taken into account.