Real-Time and Variable-Length Prediction for Drogue Motion: A Deep Learning Method

Abstract

AbstractDuring the autonomous aerial refueling (AAR) docking phase, the drogue is disturbed by multiple perturbations and will swing quickly, which increases the docking difficult for the receiver. If the drogue motion can be predicted precisely in real-time, the preview information of the drogue motion could be used in the receiver control system and the docking success rate would be improved. Most studies use the complicated hose-drogue assembly model to predict the drogue motion but the states of the hose cannot be sensed and the real-time requirement is also unsatisfied. Therefore, a deep learning based real-time and variable-length prediction method for the drogue is proposed in this paper. A gated recurrent unit model is used to extract the feature from the time series of the drogue motion. And the prediction step size is also used as an input of the model to realize the variable-length prediction. Finally, the effectiveness and correctness of the proposed drogue motion prediction method is demonstrated by simulations.KeywordsMotion predictionDeep learningAutonomous aerial refueling