Improved State Propagation through AI-based Pre-processing and Down-sampling of High-Speed Inertial Data

This paper introduces a new approach to improve state propagation for unmanned aerial vehicles using AI algorithms to preprocess high-speed inertial data. Two network architectures are evaluated, an LSTM-based approach and a Transformer encoder architecture, with the former outperforming the latter. The networks are designed to directly accept input data at variable rates and provide sufficient temporal history for good performance while maintaining a high propagation rate of preprocessed IMU samples. Results show significant improvements in propagation error even for long IMU propagation times.