Active cloaking in Stokes flows via reinforcement learning

Hydrodynamic signatures at the Stokes regime, pertinent to motility of micro-swimmers, have a long-range nature. This implies that movements of an object in such a viscosity-dominated regime can be felt tens of body-lengths away and significantly alter dynamics of the surrounding environment. Here,...

Full description

Saved in:
Bibliographic Details
Published in:Journal of fluid mechanics 2020-11, Vol.903, Article A34
Main Authors: Mirzakhanloo, Mehdi, Esmaeilzadeh, Soheil, Alam, Mohammad-Reza
Format: Article
Language:English
Subjects:
Agents (artificial intelligence)
Algorithms
Animal learning
Artificial intelligence
Decision making
Fluid dynamics
Hydrodynamics
JFM Papers
Machine learning
Optimization
Reinforcement
Stealth technology
Stokes flow
Swimming
Viscosity
Visibility
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Hydrodynamic signatures at the Stokes regime, pertinent to motility of micro-swimmers, have a long-range nature. This implies that movements of an object in such a viscosity-dominated regime can be felt tens of body-lengths away and significantly alter dynamics of the surrounding environment. Here, we devise a systematic methodology to actively cloak swimming objects within any arbitrarily crowded suspension of micro-swimmers. Specifically, our approach is to conceal the target swimmer throughout its motion using cooperative flocks of swimming agents equipped with adaptive decision-making intelligence. Through a reinforcement learning algorithm, the cloaking agents experientially learn an optimal adaptive behaviour policy in the presence of flow-mediated interactions. This artificial intelligence enables them to dynamically adjust their swimming actions, so as to optimally form and robustly retain any desired arrangement around the moving object without disturbing it from its original path. Therefore, the presented active cloaking approach not only is robust against disturbances, but also is non-invasive to motion of the cloaked object. We then further generalize the proposed approach and demonstrate how our cloaking agents can be readily used, in any region of interest, to realize hydrodynamic invisibility cloaks around any number of arbitrary intruders.
ISSN:0022-1120
1469-7645
DOI:10.1017/jfm.2020.665