A Hybrid Model and Learning-Based Adaptive Navigation Filter

Barak Or,Itzik Klein
DOI: https://doi.org/10.1109/TIM.2022.3197775
2022-09-02
Abstract:The fusion between an inertial navigation system and global navigation satellite systems is regularly used in many platforms such as drones, land vehicles, and marine vessels. The fusion is commonly carried out in a model-based extended Kalman filter framework. One of the critical parameters of the filter is the process noise covariance. It is responsible for the real-time solution accuracy, as it considers both vehicle dynamics uncertainty and the inertial sensors quality. In most situations, the process noise is covariance assumed to be constant. Yet, due to vehicle dynamics and sensor measurement variations throughout the trajectory, the process noise covariance is subject to change. To cope with such situations, several adaptive model-based Kalman filters were suggested in the literature. In this paper, we propose a hybrid model and learning-based adaptive navigation filter. We rely on the model-based Kalman filter and design a deep neural network model to tune the momentary system noise covariance matrix, based only on the inertial sensor readings. Once the process noise covariance is learned, it is plugged into the well-established, model-based Kalman filter. After deriving the proposed hybrid framework, field experiment results using a quadrotor are presented and a comparison to model-based adaptive approaches is given. We show that the proposed method obtained an improvement of 25% in the position error. Furthermore, the proposed hybrid learning method can be used in any navigation filter and also in any relevant estimation problem.
Systems and Control,Machine Learning,Robotics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to adaptively adjust the process noise covariance matrix in the integration of Inertial Navigation System (INS) and Global Navigation Satellite System (GNSS) to improve the performance of the navigation filter. Specifically, the paper focuses on the fact that in the case of different vehicle dynamics and sensor measurement changes, the traditional model - based methods usually assume that the process noise covariance matrix is fixed, which cannot well reflect the uncertainty changes in the actual situation, thus affecting the accuracy of real - time solutions. Therefore, the paper proposes a hybrid adaptive navigation filter that combines models and deep learning. By designing a deep neural network model to adjust the system noise covariance matrix in real - time according to the inertial sensor readings, the performance of the navigation filter is further improved. The main contributions of the paper are as follows: 1. Derive an adaptive hybrid learning algorithm that can determine the process noise covariance matrix in real - time according to the inertial sensor measurement values. This is the first time this adaptive hybrid learning algorithm has been proposed. 2. Integrate the proposed hybrid learning algorithm online into the implementation of the Extended Kalman Filter (es - EKF), so that this method can be applied to any external sensor that assists INS. 3. The principle of the proposed method can be further applied to any estimation problem that requires adaptive adjustment of the process noise covariance. Through this method, the paper aims to utilize the capabilities of deep learning to generalize the intrinsic properties from sequential data sets, so as to better cope with the changing conditions that affect the process noise values, and ultimately improve the accuracy and robustness of the navigation system.