Relative State Estimation with Observer-based Intermittent Kalman Filter and Radial Basis Function Neural Network

Abstract

Relative localization is a prerequisite for aerial swarms. Vision is an important method of relative localization. However, the loss of visual information caused by the limited field of view and occlusion limits the performance of vision-based relative localization. This paper proposes a novel method to solve visual loss in relative localization. The method utilizes the intermittent Kalman filter to fuse the visual information with inertial measurement unit (IMU) data. Then, we design an IMU observer to reduce estimation error. Furthermore, to decrease abrupt oscillation of estimation, we employ the RBF neural network to fuse Visual-Inertial Odometry (VIO), Ultra Wide Band (UWB) information, and the result of the Observer-based Intermittent Kalman Filter (OIKF). Real-world experiments show that the proposed method for solving the visual loss problem reduces the mean square error by more than 30% compared with an optimization method.