Jian Zhou, Xiaoyou He, Zhihong Rao, and Dun Lan
Information-enhanced, robust variational autoencoder (RVAE),prediction
The traditional Kalman filtering algorithm for predicting UAV positions relies solely on the drone’s historical location data, which makes it challenging to accurately estimate the drone’s position over extended durations. To address this issue, this paper introduces an Information-Enhanced Kalman Filtering Algorithm based on robust variational autoencoder (RVAE) optimisation. This algorithm transforms prior flight path information into virtual measurement data, supplementing the measured information. This approach not only improves the precise estimation of UAV positions but also effectively addresses the challenges of estimation arising from UAV trajectory switching. Additionally, an RVAE module is introduced to dynamically extract noise, adaptively obtaining the measurement noise covariance matrix for the Unscented Kalman Filter. This method aims to enhance the accuracy and robustness of Kalman filter predictions. Simulation results demonstrate that the information-enhanced Kalman filtering algorithm, optimised with RVAE, significantly improves prediction accuracy. Furthermore, as the prediction period increases, the performance enhancement becomes more pronounced. Finally, real flight tests using the DJI M300 validate the effectiveness of this algorithm.
Important Links:
Go Back
