Projectile trajectory estimation: an LSTM approach
Estimation de la trajectoire d'un projectile : une approche basée sur les réseaux LSTM
Résumé
This paper presents a deep learning approach to estimate a generic mortar trajectory in a GNSS-denied environment. For this purpose, Long-Short-Term-Memories (LSTMs) are trained on projectile fire simulations. Network input data are the embedded IMU (Inertial Measurement Unit), the reference magnetic field, flight parameters specific to the considered ammunition (initial velocity, fin angle, barrel elevation) and a time vector. This paper focuses on the influence of input data normalization and navigation frame rotation during the training step, leading to rescaling a 3D-value over similar variation ranges with no information loss. LSTM estimates are compared to a classical Dead Reckoning navigation algorithm. Results clearly show the AI contribution, especially for projectile position and velocity estimation.
Origine : Fichiers produits par l'(les) auteur(s)