An adaptive INS/GPS/VPS federal Kalman filter for UAV based on SVM
Xuan Xiao, Chao Shi, Yi Min Yang, Yuan Liang, Xiang Guo · 2017
In this paper, an adaptive modified federal Kalman filter is applied to the autonomous navigation with multi-sensors in condition that Unmanned Aerial Vehicle (UAV) is dynamically tracking Unmanned Ground Vehicle (UGV). To satisfy the autonomous navigation demands such as good accuracy, real time and high reliability for UAV, a new integrated navigation mode, in which Inertial Navigation System (INS) is aided by Global Position System (GPS)/Visual Positioning System (VPS), is proposed. Subsequently, a novel method is introduced which determined the information-sharing factors dynamically based on Singular Value Decomposition (SVD), and it not only solves the blindness of the information distribution factor of the conventional federated filter but also reduces the amount of calculation. According to the analysis of the theory about Support Vector Machine (SVM), an optimal objective kernel function is designed to select the effective source of information, thus it isolates the fault sensor. Simulation results show that the proposed integrated navigation system can provide abundant navigation information with sub-level navigation accuracy and good fault-tolerant performance. UAV can obtain reliable navigation information by this modified federal Kalman filter algorithm even when GPS or VPS is continuously interrupted for a period.