CNN Based Adaptive Kalman Filter in High-Dynamic Condition for Low-Cost Navigation System on Highspeed UAV

Abstract

Kalman Filter (KF) is widely used in navigation as a data-fusion algorithm. When KF is applied in high-speed Unmanned Aerial Vehicle (UAV) mounted with low-cost integrated navigation system, its performance always deteriorates in complicated and high-dynamic conditions. Facing such scenario, we proposed a new algorithm of adaptive Kalman Filter in this paper. The new method is based on 1-dimensional Convolutional Neural Network (CNN). The key component of the algorithm is a deep neural network estimator of system noise covariance. We modeled Global Navigation Satellite System (GNSS)/Inertial Navigation System (INS) integrated navigation system and then trained the estimator using IMU and GNSS data which is sampled in real flight. Further, we tested the proposed algorithm on another real-sampled dataset of UAV and compared its performance with classical KF and Sage-Husa adaptive Filter (SHF). The results show a better adaptiveness of the proposed algorithm in high-dynamic condition and partly liberates researchers from parameter tuning.