Concurrent shape and controller optimization of autonomous underwater vehicles for path following tasks using evolutionary algorithm-driven reinforcement learning

•Concurrent design of the shape and control system of an AUV for path following.•3D path following in environments with unsteady non-uniform flow model.•An evolutionary algorithm is used for morphological optimization.•Reinforcement learning is applied for controller training. Autonomous underwater...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Ocean engineering Jg. 336; S. 121841
Hauptverfasser: Sariman, Cagatay, Hallawa, Ahmed, Schmeink, Anke
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.09.2025
Schlagworte:
Autonomous underwater vehicles
Evolutionary algorithms
Path following
Reinforcement learning
Autonomous underwater vehicles
Evolutionary algorithms
Reinforcement learning
Path following
ISSN:0029-8018
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•Concurrent design of the shape and control system of an AUV for path following.•3D path following in environments with unsteady non-uniform flow model.•An evolutionary algorithm is used for morphological optimization.•Reinforcement learning is applied for controller training. Autonomous underwater vehicles (AUVs) are advanced robotic systems adept at efficiently tackling diverse marine tasks such as resource exploration, seabed mapping, pipe inspections, and military operations. However, achieving these tasks demands precise navigation, which hinges on a dependable controller. Yet, crafting such a controller proves challenging due to the intricate hydrodynamics of AUVs and the uncertainties of underwater environments. On the other hand, the shape of the AUV greatly influences its performance. In our study, we introduce a novel approach utilizing evolutionary algorithms and reinforcement learning within the UR-EARL framework to optimize both the form and behaviour of AUVs for path following tasks. We employ a Lindenmayer-Systems-based evolutionary algorithm to refine the AUV shape, while training the controller using the twin delayed deep deterministic policy gradient algorithm. Additionally, we propose a step-by-step optimization strategy to break down the complex control challenge into more manageable segments, solving them sequentially. Simulation results demonstrate the effectiveness of our methodology in consistently designing AUVs capable of accurately reaching predefined targets and following given 3D paths.
ISSN:0029-8018
DOI:10.1016/j.oceaneng.2025.121841