Limited-perception swimmer navigation in flow through memory-enhanced reinforcement learning
Machine Learning: Science and Technology · 01.06.2026 · 7 (3) · 035043
DOI: 10.1088/2632-2153/ae7233 ↗Kurz gefasst
Fachartikel in der Zeitschrift Machine Learning: Science and Technology, online erschienen am 05.06.2026. Verfasst von Changxu Zhao und 2 weiteren Autor:innen. Die Arbeit nennt 65 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Shanghai Municipality.
Abstract (Originalsprache)
Navigation under conditions of limited detection resolution and insufficient information poses a major challenge in many applications. To address these limitations, we propose a reinforcement learning framework integrated with recurrent neural network (RNN) and investigate a model swimmer navigating toward a chemical source in an unsteady flow field. The algorithm utilizes fixed-length temporal windows for memory retention and is trained through iterative agent-environment interactions in dynamically generated hydrodynamic datasets. By extracting temporal features, this architecture enhances environmental comprehension without requiring additional spatial detection. Comparative analysis shows that incorporating RNN units with an optimal memory length yields significant improvements in both success rate and navigation efficiency compared with heuristic and learning-based baselines, including its direct sequential-update counterpart. Furthermore, our investigation of different observation inputs reveals that velocity-based swimmers outperform those relying on velocity gradient information. The performance of the trained neural network, validated across noisy environments and diverse flow conditions, confirms the robustness and generalization capability of the proposed framework.
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Shanghai Municipality
Zitieren
APA
Zhao, C., Zhao, D., & Li, G. (2026). Limited-perception swimmer navigation in flow through memory-enhanced reinforcement learning. Machine Learning: Science and Technology, 7(3), 035043. https://doi.org/10.1088/2632-2153/ae7233
Vancouver
Zhao C, Zhao D, Li G. Limited-perception swimmer navigation in flow through memory-enhanced reinforcement learning. Machine Learning: Science and Technology. 2026;7(3):035043. doi:10.1088/2632-2153/ae7233
BibTeX
@article{Zhao2026,
author = {Zhao, Changxu and Zhao, Dongxiao and Li, Gaojin},
title = {Limited-perception swimmer navigation in flow through memory-enhanced reinforcement learning},
journal = {Machine Learning: Science and Technology},
year = {2026},
volume = {7},
number = {3},
pages = {035043},
doi = {10.1088/2632-2153/ae7233}
}
EUVIMED