Online Recursive Gaussian Process for Target Location Based on Ultra-Short Baselines

• Published in: 2024 36th Chinese Control and Decision Conference (CCDC) pp. 4361 - 4366
• Main Authors: Lei, Zhang, Zhichu, Lei, Yanjie, Le, Kai, Hu, Jianbo, Wang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25.05.2024
• Subjects: Accuracy; Autonomous underwater vehicles; Filtering; Filtering algorithms; Gaussian process learning; Gaussian processes; Heuristic algorithms; Target tracking; ultra-short baseline
• ISSN: 1948-9447
• DOI: 10.1109/CCDC62350.2024.10587757

In complex Marine environments, accurate tracking of Autonomous Underwater Vehicles (AUV) with ultra-short baselines is one of the most challenging tasks in underwater tracking. For effective tracking performance, it is necessary to build a dynamic model that matches the actual movement of the target. However, the AUV is often highly maneuverable, and it is difficult for a single model to accurately match the motion of AUV over a long period of time. Interactive multiple model (IMM) algorithm, one of the most classical robust algorithm, requires a given model set, and its tracking performance deteriorates when the target's motion pattern differs greatly from that given by the model set. In order to overcome the problem of model uncertainty, this paper proposes a novel robust filter for AUV target tracking based on recursive Gaussian processes. Gaussian processes are used to model the AUV motion, and the AUV state and model parameters are jointly estimated by Kalman filter. Finally, we verify the effectiveness of the proposed RGPKF algorithm in typical maneuvering target tracking simulation scenarios. Moreover, the proposed algorithm shows better tracking accuracy than model-based Kalman filter.