Publication
Online Model Adaptation of Autonomous Underwater Vehicles with LSTM Networks
Miguel Bande Firvida; Bilal Wehbe
In: Miguel Bande Firvida; Bilal Wehbe (Hrsg.). Online Model Adaptation of Autonomous Underwater Vehicles with LSTM Networks. OCEANS MTS/IEEE Conference (OCEANS-2021), OCEANS 2021: San Diego – Porto, September 20-12, Porto, Portugal, Pages 1-6, ISBN 978-0-692-93559-0, IEEE, 2021.
Abstract
This work addresses the online learning and adaptation of Autonomous Underwater vehicle (AUV) models. A framework is presented that employs long-short term memory (LSTM) networks which can be used to model the temporal dependencies in the AUV data. To reduce the effect of catastrophic forgetting, a memory efficient rehearsal method is developed with including and forgetting strategies to manage the butter of training samples. The proposed method is validated using experimental data proving its capability to adapt to real changes in the vehicles dynamics.
Projects
- EurEx-LUNa - EurEx - Persistent under-ice navigation
- TRIPLE-nanoAUV 1 - Localisation and perception of a miniaturized autonomous underwater vehicle for the exploration of subglacial lakes