pith. sign in

arxiv: 2606.06077 · v1 · pith:KLFPPLFXnew · submitted 2026-06-04 · 💻 cs.RO · cs.LG

3D Underwater Path Planning via Generative Flow Field Surrogates

Zachary Cooper-Baldock , Paulo E. Santos , Russell S.A. Brinkworth , Karl Sammut This is my paper

Pith reviewed 2026-06-28 01:26 UTC · model grok-4.3

classification 💻 cs.RO cs.LG
keywords AUV path planningcGAN surrogatespropeller wakeCFD replacement3D flow fieldsenergy-weighted planning
0
0 comments X

The pith

Conditional generative adversarial networks recover 45-60% of CFD benefits for 3D underwater path planning at edge-device speeds.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that two cGAN architectures can generate full three-dimensional flow field volumes from scalar inputs alone, replacing expensive RANS CFD data inside an energy-weighted A* planner for AUVs crossing propeller wakes. The surrogates run in tens to hundreds of microseconds per volume while retaining a large share of the energy savings and high-velocity avoidance gains that full CFD enables over uniform-current planning. A reader would care because onboard computation in complex wakes has previously been blocked by CFD run times measured in hours. The evaluation across 19,800 trajectories and 550 flow conditions quantifies how much of the CFD advantage survives the substitution.

Core claim

Both cGAN architectures recover approximately 45-60% of the CFD energy benefit and high-velocity cell avoidance benefit while operating at inference speeds compatible with edge device use.

What carries the argument

Conditional generative adversarial networks that synthesise 128^3 voxel flow field volumes from scalar operating condition inputs, used as drop-in replacements for CFD data inside the energy-weighted A* path planner.

Load-bearing premise

The flow fields produced by the cGANs are accurate enough in the regions that affect path costs and high-velocity encounters for the planner to capture most of the CFD-derived advantages.

What would settle it

A side-by-side test in which paths planned on cGAN fields achieve less than 20% of the energy reduction or high-velocity avoidance improvement that the same planner achieves when given the corresponding ground-truth CFD fields.

Figures

Figures reproduced from arXiv: 2606.06077 by Karl Sammut, Paulo E. Santos, Russell S.A. Brinkworth, Zachary Cooper-Baldock.

Figure 1
Figure 1. Figure 1: Conceptual pipeline for path planning in a predicted flow field. Sparse speed and angle data [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a - d): The ground truth flow fields from RANS CFD results for a speed of 2.5 m/s with turns of 0, 10, 20 and 40 degrees respectively. (e - h): show the generated fields from the PatchGAN GradNorm model. All velocity magnitude data has been normalised between to the range [0,1]. 5.3.1 Regularised PatchGAN GradNorm The Regularised PatchGAN GradNorm is the lighter and faster of the two 3D generators, and is… view at source ↗
Figure 3
Figure 3. Figure 3: 3D isosurface plot of the mean path energy cost (E) for the Current, CFD, PatchGAN [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: 3D isosurface plot of the mean trajectory length for the Current, CFD, PatchGAN and [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: 3D isosurface plot of the mean number of high velocity cells encountered for the Current, [PITH_FULL_IMAGE:figures/full_fig_p026_5.png] view at source ↗
read the original abstract

Autonomous underwater vehicle (AUV) launch and recovery (LAR) into the hull of an advancing host platform requires traversal of a complex, three-dimensional propeller wake whose hydrodynamic structure cannot be characterised by a uniform current model. High-fidelity Reynolds-Averaged Navier-Stokes (RANS) Computational Fluid Dynamics (CFD) simulations resolve this structure with sufficient accuracy for path planning, but their computational cost renders them impractical for onboard use. We address this gap by integrating two conditional generative adversarial network (cGAN) architectures -- a regularised PatchGAN and a 2D3DGAN with self-attention -- as drop-in replacements for RANS CFD data within a three-dimensional, energy-weighted A* path planning framework. Both generators are driven by a hierarchical pipeline that synthesises full $128^3$ voxel flow field volumes from scalar operating condition inputs alone, with end-to-end inference times of approximately 28-146 $\mu$s, compared to hours for a single RANS computation. We benchmark all four environmental knowledge levels: uniform current, ground-truth CFD, PatchGAN, and 2D3DGAN~SA across 19,800 independently generated trajectories spanning 550 distinct flow conditions. Full CFD wake knowledge reduces energy expenditure by 5.7-12.5% and high-velocity wake-core encounters by up to 77.8% relative to uniform-current planning, with both benefits scaling with operating severity. The cGAN surrogates recover approximately 45-60% of the CFD energy benefit and high-velocity cell avoidance benefit while operating at inference speeds compatible with edge device use. These results provide the first systematic quantification of the downstream path planning value of cGAN-predicted hydrodynamic fields in a three-dimensional maritime robotics application.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces two cGAN architectures (regularised PatchGAN and 2D3DGAN with self-attention) as fast surrogates for RANS CFD to generate 128^3 3D flow fields from scalar operating conditions. These surrogates are integrated into an energy-weighted 3D A* planner for AUV launch/recovery through propeller wakes. On a benchmark of 19,800 trajectories across 550 synthetic flow conditions, full CFD reduces energy use by 5.7-12.5% and high-velocity encounters by up to 77.8% versus uniform-current planning; the cGANs recover 45-60% of those benefits at 28-146 μs inference times.

Significance. If the surrogate fidelity holds in planner-critical regions, the work supplies the first quantitative demonstration that generative models can deliver usable hydrodynamic awareness for real-time maritime path planning, closing the gap between expensive CFD and edge-device deployment. The scale of the trajectory benchmark (19,800 independent cases) and direct comparison to independent ground-truth CFD are strengths that allow a clear measurement of recovered benefit.

major comments (2)
  1. [Abstract] Abstract: The headline recovery percentages (45-60% of CFD energy and avoidance benefits) rest on the assumption that the 550 synthetic RANS conditions plus 19,800 trajectories adequately sample the flow structures that matter for planning. The manuscript states that all training and benchmarking use synthetic RANS data generated from scalar inputs with no experimental PIV, LES, or at-sea measurements referenced; if real propeller wakes contain unmodeled features (hull interaction, unsteady shedding, geometry-specific effects) that alter velocity fields in ways absent from the training distribution, the reported recovery will not transfer to the claimed maritime robotics application.
  2. [Abstract] Abstract: The claim that the cGANs operate 'at inference speeds compatible with edge device use' and provide 'the first systematic quantification ... in a three-dimensional maritime robotics application' is presented without any discussion of how the synthetic condition set was constructed or validated for coverage of real wake variability, which is load-bearing for the downstream planning metrics.
minor comments (1)
  1. [Abstract] The abstract introduces two distinct cGAN architectures but does not indicate which quantitative results correspond to each; a short parenthetical or table reference would improve readability.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments on the abstract and the scope of our synthetic dataset. We address each point below and will revise the manuscript accordingly to clarify limitations while preserving the core contribution of the benchmark study.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline recovery percentages (45-60% of CFD energy and avoidance benefits) rest on the assumption that the 550 synthetic RANS conditions plus 19,800 trajectories adequately sample the flow structures that matter for planning. The manuscript states that all training and benchmarking use synthetic RANS data generated from scalar inputs with no experimental PIV, LES, or at-sea measurements referenced; if real propeller wakes contain unmodeled features (hull interaction, unsteady shedding, geometry-specific effects) that alter velocity fields in ways absent from the training distribution, the reported recovery will not transfer to the claimed maritime robotics application.

    Authors: We agree that all results are obtained on synthetic RANS data and that the reported recovery percentages cannot be claimed to transfer to real propeller wakes without experimental validation. The 550 conditions were generated by varying scalar inputs (advance speed, propeller rate, etc.) over ranges representative of AUV LAR operations to produce diverse 3D wake structures. We will revise the abstract to explicitly qualify that the 45-60% recovery is measured relative to ground-truth RANS CFD within the synthetic domain and add a limitations paragraph noting the absence of experimental PIV/LES/at-sea data. revision: yes

  2. Referee: [Abstract] Abstract: The claim that the cGANs operate 'at inference speeds compatible with edge device use' and provide 'the first systematic quantification ... in a three-dimensional maritime robotics application' is presented without any discussion of how the synthetic condition set was constructed or validated for coverage of real wake variability, which is load-bearing for the downstream planning metrics.

    Authors: We will add a dedicated subsection describing the synthetic dataset construction, including the ranges and sampling of scalar operating conditions and the resulting coverage of wake features (velocity deficits, vortices). The inference-speed claim is supported by the reported 28-146 μs timings on standard hardware. The 'first systematic quantification' refers to the scale of the 19,800-trajectory benchmark with direct CFD comparison; we will revise the phrasing to 'first systematic quantification in simulated three-dimensional maritime robotics scenarios' to avoid overstatement. revision: yes

standing simulated objections not resolved
  • We do not have access to experimental PIV, LES, or at-sea measurements and therefore cannot demonstrate that the reported recovery percentages transfer to real propeller wakes containing unmodeled features.

Circularity Check

0 steps flagged

No circularity: benchmarks use independent CFD ground truth

full rationale

The paper trains cGAN surrogates on synthetic RANS data and reports path-planning benefits by directly comparing energy expenditure and high-velocity encounters across four knowledge levels (uniform current, ground-truth CFD, PatchGAN, 2D3DGAN-SA) on 19,800 trajectories. These metrics are computed from A* simulations using the respective flow fields as inputs; no equation reduces the 45-60% recovery figure to a fitted parameter or self-referential definition, and no self-citation chain is invoked to justify the central claims. The derivation therefore remains self-contained against external CFD benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the approach implicitly relies on standard assumptions that the training distribution matches real wakes and that voxel-grid accuracy translates to planning metrics.

pith-pipeline@v0.9.1-grok · 5861 in / 1151 out tokens · 29589 ms · 2026-06-28T01:26:39.683134+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

65 extracted references · 33 canonical work pages

  1. [1]

    Cooper-Baldock, P

    Z. Cooper-Baldock, P. E. Santos, R. S. A. Brinkworth, and K. Sammut. Hydrodynamic Analysis of Payload Bay Berthing for Underwater Vehicles.IEEE Journal of Oceanic Engineering, 49:727–748, 2024.doi:10.1109/JOE.2024.3352714

    work page doi:10.1109/joe.2024.3352714 2024

  2. [2]

    Cooper-Baldock, P

    Z. Cooper-Baldock, P. E. Santos, R. S. A. Brinkworth, and K. Sammut. Payload Bay Berthing of Underwater Vehicles with a Larger XLUUV.7th Submarine Science, Technology & Engineering Conference 2023 (SubSTEC7), 1:97–104.doi:970-0-9944841-4-7

    2023

  3. [3]

    Turnock, and Karl Sammut

    Zachary Cooper-Baldock, Stephen R. Turnock, and Karl Sammut. Wake-informed 3d path plan- ning for autonomous underwater vehicles using a and neural network approximations.Ocean Engineering, 332:121353, 2025. URL: https://www.sciencedirect.com/science/ article/pii/S0029801825010662,doi:10.1016/j.oceaneng.2025.121353. 31

    work page doi:10.1016/j.oceaneng.2025.121353 2025

  4. [4]

    Turbulence-aware uav path planning in urban environments

    Enrique Aldao Pensado, Gabriel Fontenla Carrera, Fernando Veiga López, Higinio González Jorge, and Elena Martín Ortega. Turbulence-aware uav path planning in urban environments. In 2024 International Conference on Unmanned Aircraft Systems (ICUAS), pages 280–285, 2024. doi:10.1109/ICUAS60882.2024.10556934

    work page doi:10.1109/icuas60882.2024.10556934 2024

  5. [5]

    Santos, Russell S

    Zachary Cooper-Baldock, Paulo E. Santos, Russell S. A. Brinkworth, and Karl Sammut. A generalised novel loss function for computational fluid dynamics, 2024. URL:https://arxiv. org/abs/2411.17059,arXiv:2411.17059

    arXiv 2024

  6. [6]

    PhD thesis, College of Science and Engineering, Flinders University, Australia, 2026

    Zachary Cooper-Baldock.Rapid Flow Field Approximation via Generative Machine Learning for Underwater Vehicles. PhD thesis, College of Science and Engineering, Flinders University, Australia, 2026

    2026

  7. [7]

    Yousif, Linqi Yu, Sergio Hoyas, Ricardo Vinuesa, and HeeChang Lim

    Mustafa Z. Yousif, Linqi Yu, Sergio Hoyas, Ricardo Vinuesa, and HeeChang Lim. A deep- learning approach for reconstructing 3D turbulent flows from 2D observation data.Nature Scientific Reports, 13, 2023.doi:10.1038/s41598-023-29525-9

    work page doi:10.1038/s41598-023-29525-9 2023

  8. [8]

    Santos, Russell S.A

    Zachary Cooper-Baldock, Paulo E. Santos, Russell S.A. Brinkworth, and Karl Sammut. Wake- set, 2026. URL: https://huggingface.co/datasets/ZacharyCB99/WAKESET, doi: 10.57967/HF/7698

    work page doi:10.57967/hf/7698 2026

  9. [9]

    Rendezvous path planning for multiple autonomous marine vehicles.IEEE Journal of Oceanic Engineering, 43(3):640–664, 2018.doi:10.1109/JOE.2017.2723058

    Zheng Zeng, Karl Sammut, Lian Lian, Andrew Lammas, Fangpo He, and Youhong Tang. Rendezvous path planning for multiple autonomous marine vehicles.IEEE Journal of Oceanic Engineering, 43(3):640–664, 2018.doi:10.1109/JOE.2017.2723058

    work page doi:10.1109/joe.2017.2723058 2018

  10. [10]

    Three-dimensional path planning for auvs in ocean currents environment based on an improved compression factor particle swarm optimization algorithm

    Xiaohong Li and Shuanghe Yu. Three-dimensional path planning for auvs in ocean currents environment based on an improved compression factor particle swarm optimization algorithm. Ocean Engineering, 280:114610, 2023.doi:10.1016/j.oceaneng.2023.114610

    work page doi:10.1016/j.oceaneng.2023.114610 2023

  11. [11]

    Petres, Y

    C. Petres, Y . Pailhas, P. Patron, Y . Petillot, J. Evans, and D. Lane. Path planning for autonomous underwater vehicles.IEEE Transactions on Robotics, 23:331–341, 2007. doi:10.1109/TRO. 2007.895057

    work page doi:10.1109/tro 2007

  12. [12]

    J. C. Latombe.Robot Motion Planning. The Netherlands: Kluwer, 1991

    1991

  13. [13]

    J. G. Bellingham and J. S. Willcox. Optimizing auv oceanographic surveys.Proc. IEEE Symp. Autonomous Underwater Vehicles Technology, 1:391—-398, 1996

    1996

  14. [14]

    J. D. Hernández, E. Vidal, G. Vallicrosa, E. Galceran, and M. Carreras. Online path planning for autonomous underwater vehicles in unknown environments.2015 IEEE International Conference on Robotics and Automation (ICRA), 1:1152–1157, 2015. doi:10.1109/ICRA. 2015.7139336

    work page doi:10.1109/icra 2015

  15. [15]

    Omar Khattab, Keshav Santhanam, Xiang Lisa Li, David Hall, Percy Liang, Christopher Potts, and Matei Zaharia

    Peter E. Hart, Nils J. Nilsson, and Bertram Raphael. A formal basis for the heuristic deter- mination of minimum cost paths.IEEE Transactions on Systems Science and Cybernetics, 4(2):100–107, 1968.doi:10.1109/TSSC.1968.300136

    work page doi:10.1109/tssc.1968.300136 1968

  16. [16]

    Garau, A

    B. Garau, A. Alvarez, and G. Oliver. Path planning of autonomous underwater vehicles in current fields with complex spatial variability: an a* approach. InProceedings of the 2005 IEEE International Conference on Robotics and Automation, pages 194–198, 2005. doi:10.1109/ROBOT.2005.1570118

    work page doi:10.1109/robot.2005.1570118 2005

  17. [17]

    Path planning for autonomous underwater vehicles.IEEE Transactions on Robotics, 23(2):331–341, 2007.doi:10.1109/TRO.2007.895057

    Clment Petres, Yan Pailhas, Pedro Patron, Yvan Petillot, Jonathan Evans, and David Lane. Path planning for autonomous underwater vehicles.IEEE Transactions on Robotics, 23(2):331–341, 2007.doi:10.1109/TRO.2007.895057

    work page doi:10.1109/tro.2007.895057 2007

  18. [18]

    Optimal auv path planning for extended missions in complex, fast-flowing estuarine environments

    Dov Kruger, Rustam Stolkin, Aaron Blum, and Joseph Briganti. Optimal auv path planning for extended missions in complex, fast-flowing estuarine environments. InProceedings 2007 IEEE International Conference on Robotics and Automation, pages 4265–4270, 2007. doi: 10.1109/ROBOT.2007.364135

    work page doi:10.1109/robot.2007.364135 2007

  19. [19]

    Kulkarni and Pierre F.J

    Chinmay S. Kulkarni and Pierre F.J. Lermusiaux. Three-dimensional time-optimal path planning in the ocean.Ocean Modelling, 152:101644, 2020. doi:10.1016/j.ocemod.2020.101644. 32

    work page doi:10.1016/j.ocemod.2020.101644 2020

  20. [20]

    Mahmoud Zadeh, D.W.M

    S. Mahmoud Zadeh, D.W.M. Powers, A.M. Yazdani, K. Sammut, and A. Atyabi. Effi- cient auv path planning in time-variant underwater environment using differential evolu- tion algorithm.Journal of Marine Science and Application, 17:585–591, 2018. doi: 10.1007/s11804-018-0034-4

    work page doi:10.1007/s11804-018-0034-4 2018

  21. [21]

    S. M. LaValle.Planning Algorithms. Cambridge University Press, 2006

    2006

  22. [22]

    Multi-goal path planning using multiple random trees.IEEE Robotics and Automation Letters, 6(2):4201–4208, 2021

    Jaroslav Janoš, V ojtˇech V onásek, and Robert Pˇeniˇcka. Multi-goal path planning using multiple random trees.IEEE Robotics and Automation Letters, 6(2):4201–4208, 2021. doi:10.1109/ LRA.2021.3068679

    arXiv 2021

  23. [23]

    Mobile robot path planning using an improved ant colony optimization.International Journal of Advanced Robotic Systems, 15(3):1729881418774673, 2018.doi:10.1177/1729881418774673

    Khaled Akka and Farid Khaber. Mobile robot path planning using an improved ant colony optimization.International Journal of Advanced Robotic Systems, 15(3):1729881418774673, 2018.doi:10.1177/1729881418774673

    work page doi:10.1177/1729881418774673 2018

  24. [24]

    Alvarez, A

    A. Alvarez, A. Caiti, and R. Onken. Evolutionary path planning for autonomous underwater vehicles in a variable ocean.IEEE Journal of Oceanic Engineering, 29(2):418–429, 2004. doi:10.1109/JOE.2004.827837

    work page doi:10.1109/joe.2004.827837 2004

  25. [25]

    Perception as policy: Learning perceptual representations for safe decision-making

    Mark Pfeiffer, Sudeep Shukla, Cesar Cadena, and Davide Scaramuzza. Perception as policy: Learning perceptual representations for safe decision-making. In2017 IEEE International Conference on Robotics and Automation (ICRA), pages 6236–6243. IEEE, 2017

    2017

  26. [26]

    A deep-network solution towards model-less obstacle avoidance

    Lei Tai, Jiang Li, and Ming Liu. A deep-network solution towards model-less obstacle avoidance. In2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 2759–2764. IEEE, 2017

    2017

  27. [27]

    Speed, power and cost implications for GPU acceleration of Computational Fluid Dynamics on HPC systems, 2024

    Zachary Cooper-Baldock, Brenda Vara Almirall, and Kiao Inthavong. Speed, power and cost implications for GPU acceleration of Computational Fluid Dynamics on HPC systems, 2024. arXiv:2404.02482

    arXiv 2024

  28. [28]

    Raissi, P

    M. Raissi, P. Perdikaris, and G.E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics, 378:686–707, 2019. doi:10.1016/ j.jcp.2018.10.045

    2019

  29. [29]

    Hanrahan, M

    S. Hanrahan, M. Kozul, and R.D. Sandberg. Studying turbulent flows with physics-informed neural networks and sparse data.International Journal of Heat and Fluid Flow, 104:109232, 2023.doi:10.1016/j.ijheatfluidflow.2023.109232

    work page doi:10.1016/j.ijheatfluidflow.2023.109232 2023

  30. [30]

    Understanding physics-informed neural networks: Techniques, applications, trends, and challenges.AI, 5(3):1534–1557, 2024

    Amer Farea, Olli Yli-Harja, and Frank Emmert-Streib. Understanding physics-informed neural networks: Techniques, applications, trends, and challenges.AI, 5(3):1534–1557, 2024. URL: https://www.mdpi.com/2673-2688/5/3/74,doi:10.3390/ai5030074

    work page doi:10.3390/ai5030074 2024

  31. [31]

    Characterizing possible failure modes in physics-informed neural networks (characterizing-pinns-failure-modes) v0

    Aditi Krishnapriyan. Characterizing possible failure modes in physics-informed neural networks (characterizing-pinns-failure-modes) v0. 1.0. Technical report, Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States), 2021

    2021

  32. [32]

    Reconstruction of three-dimensional turbulent flow structures using surface measurements for free-surface flows based on a convolutional neural network

    Anqing Xuan and Lian Shen. Reconstruction of three-dimensional turbulent flow structures using surface measurements for free-surface flows based on a convolutional neural network. Journal of Fluid Mechanics, 959:A34, 2023.doi:10.1017/jfm.2023.154

    work page doi:10.1017/jfm.2023.154 2023

  33. [33]

    Controllable generative adversarial network.IEEE Access, 7:28158–28169, 2019.doi:10.1109/ACCESS.2019.2899108

    Minhyeok Lee and Junhee Seok. Controllable generative adversarial network.IEEE Access, 7:28158–28169, 2019.doi:10.1109/ACCESS.2019.2899108

    work page doi:10.1109/access.2019.2899108 2019

  34. [34]

    Conditional generative adversarial nets, 2014

    Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets, 2014. URL: https://arxiv.org/abs/1411.1784,arXiv:1411.1784

    Pith/arXiv arXiv 2014

  35. [35]

    Physics-informed generative adver- sarial networks for stochastic differential equations.SIAM Journal on Scientific Computing, 42(1):A292–A317, 2020.doi:10.1137/18M1225409

    Liu Yang, Dongkun Zhang, and George Em Karniadakis. Physics-informed generative adver- sarial networks for stochastic differential equations.SIAM Journal on Scientific Computing, 42(1):A292–A317, 2020.doi:10.1137/18M1225409

    work page doi:10.1137/18m1225409 2020

  36. [36]

    AIAA, 2020.doi:10.2514/6.2020-3185

    Emre Yilmaz and Brian German.Conditional Generative Adversarial Network Framework for Airfoil Inverse Design. AIAA, 2020.doi:10.2514/6.2020-3185. 33

    work page doi:10.2514/6.2020-3185 2020

  37. [37]

    Navier–stokes generative adversarial network: a physics-informed deep learning model for fluid flow generation.Neural Computing and Applications, 34:11539–11552, 2022

    Pin Wu, Kaikai Pan, Lulu Ji, Siquan Gong, Weibing Feng, Wenyan Yuan, and Christopher Pain. Navier–stokes generative adversarial network: a physics-informed deep learning model for fluid flow generation.Neural Computing and Applications, 34:11539–11552, 2022

    2022

  38. [38]

    Self-attention generative adversarial networks

    Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial networks. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors,Proceedings of the 36th International Conference on Machine Learning, volume 97 ofProceedings of Machine Learning Research, pages 7354–7363. PMLR, 6 2019

    2019

  39. [39]

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A. Efros. Image-to-image translation with conditional adversarial networks. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5967–5976, 2017.doi:10.1109/CVPR.2017.632

    work page doi:10.1109/cvpr.2017.632 2017

  40. [40]

    Yousif, Linqi Yu, and Hee-Chang Lim

    Mustafa Z. Yousif, Linqi Yu, and Hee-Chang Lim. High-fidelity reconstruction of turbulent flow from spatially limited data using enhanced super-resolution generative adversarial network. Physics of Fluids, 33(12):125119, 12 2021. arXiv:https://pubs.aip.org/aip/pof/ article-pdf/doi/10.1063/5.0066077/15873357/125119\_1\_online.pdf, doi:10. 1063/5.0066077

    work page doi:10.1063/5.0066077/15873357/125119 2021

  41. [41]

    Film: visual reasoning with a general conditioning layer

    Ethan Perez, Florian Strub, Harm de Vries, Vincent Dumoulin, and Aaron Courville. Film: visual reasoning with a general conditioning layer. InProceedings of the Thirty-Second AAAI Conference on Artificial Intelligence and Thirtieth Innovative Applications of Artificial Intelligence Conference and Eighth AAAI Symposium on Educational Advances in Artificial...

    2018

  42. [42]

    The use of GANs and transfer learning in model-order reduction of turbulent wake of an isolated high-rise building.Building and Environment, 246:110948, 2023

    Shahin Masoumi-Verki, Fariborz Haghighat, Nizar Bouguila, and Ursula Eicker. The use of GANs and transfer learning in model-order reduction of turbulent wake of an isolated high-rise building.Building and Environment, 246:110948, 2023. doi:10.1016/j.buildenv.2023. 110948

    work page doi:10.1016/j.buildenv.2023 2023

  43. [43]

    Data-driven prediction of unsteady flow over a circular cylinder using deep learning.Journal of Fluid Mechanics, 879:217–254, 2019

    Sangseung Lee and Donghyun You. Data-driven prediction of unsteady flow over a circular cylinder using deep learning.Journal of Fluid Mechanics, 879:217–254, 2019. doi:10.1017/ jfm.2019.700

    2019

  44. [44]

    Reconstruction of 3d porous media from 2d slices, 2021

    Denis V olkhonskiy, Ekaterina Muravleva, Oleg Sudakov, Denis Orlov, Boris Belozerov, Evgeny Burnaev, and Dmitry Koroteev. Reconstruction of 3d porous media from 2d slices, 2021. URL: https://arxiv.org/abs/1901.10233,arXiv:1901.10233

    arXiv 2021

  45. [45]

    Autonomous systems – echo voyager

    Boeing. Autonomous systems – echo voyager. Technical report, Boeing Defense,

  46. [46]

    URL: https://www.boeing.com/defense/autonomous-systems/echovoyager/ index.page

  47. [47]

    Echo voyager system specification and overview

    Boeing. Echo voyager system specification and overview. Technical report, Boe- ing, 2023. URL: https://www.boeing.com/resources/boeingdotcom/defense/ autonomoussystems/echo-voyager/echo_voyager_product_sheet.pdf

    2023

  48. [48]

    G. Alfonsi. Reynolds-averaged navier–stokes equations for turbulence modelling.Appl. Mech. Rev, 62:040802, 2009.doi:10.1115/1.3124648

    work page doi:10.1115/1.3124648 2009

  49. [49]

    T. H. Shih, W. W. Liou, A. Shabbir, Z. Yang, and J. Zhu. A new k-e eddy viscosity model for high reynolds number turbulent flows - model development and validation.Comput. Fluids, 24:227–238, 1995.doi:10.1016/0045-7930(94)00032-T

    work page doi:10.1016/0045-7930(94)00032-t 1995

  50. [50]

    Y . C. Pan, H. X. Zhang, and Q. D. Zhou. Numerical prediction of submarine hydrody- namic coefficients using cfd simulation.J. Hydrodyn., 24:840–847, 2012. doi:10.1016/ S1001-6058(11)60311-9

    2012

  51. [51]

    A. B. Phillips, S. R. Turnock, and M. Furlong. Influence of turbulence closure models on the vortical flow field around a submarine body undergoing steady drift.J. Mar. Sci. Technol., 15:201–217, 2010.doi:10.1007/s00773-010-0090-1

    work page doi:10.1007/s00773-010-0090-1 2010

  52. [52]

    Modelling turbulence: Standard, rng, and realizable ke models theory

    ANSYS. Modelling turbulence: Standard, rng, and realizable ke models theory. Technical report, Washington Univ., Seattle WA, USA, 2006. URL: https://courses.washington. edu/mengr544/handouts-10/Fluent-k-epsilon.pdf. 34

    2006

  53. [53]

    M. C. Özden, A. Y . Gürkan, Y . A. Özden, T. G. Canyurt, and E. Korkut. Underwater radiated noise prediction for a submarine propeller in different flow conditions.Ocean Eng., 126:488– 500, 2016.doi:10.1016/j.oceaneng.2016.06.012

    work page doi:10.1016/j.oceaneng.2016.06.012 2016

  54. [54]

    Andersen, J

    P. Andersen, J. J. Kappel, and E. F. Spangenberg. Aspects of propeller developments for a submarine.Proc. 1st Int. Symp. Mar. Propulsors, pages 1–8, 2009

    2009

  55. [55]

    M. A. Chamberlain, P. R. Oke, G. B. Brassington, P. Sandery, P. Divakaran, and R. A. S. Fiedler. Multiscale data assimilation in the bluelink ocean reanalysis (bran).Ocean Model., 166, 2021. doi:10.1016/j.ocemod.2021.101849

    work page doi:10.1016/j.ocemod.2021.101849 2021

  56. [56]

    M. E. Q. Pilson.An Introduction to the Chemistry of the Sea, 2nd Ed.Cambridge Univ. Press, 2012

    2012

  57. [57]

    M. H. Sharqawy, J. H. V . Lienhard, and S. M. Zubair. Thermophysical properties of seawater: A review of existing correlations and data.Desalination Water Treat., 16:354–380, 2010. doi:10.5004/dwt.2010.1079

    work page doi:10.5004/dwt.2010.1079 2010

  58. [58]

    F. J. Millero, G. Perron, and J. E. Desnoyers. Heat capacity of seawater solutions from 5c to 35c and 0.5 to 22% chlorinity.J. Geophys. Res.,, 78:4499–4507, 1973. doi:10.1029/ JC078i021p04499

    1973

  59. [59]

    V . J. Castelli, E. M. Stanley, and E. C. Fischer. The thermal conductivity of seawater as a function of pressure and temperature.Deep Sea Res. Oceanogr. Abstr, 21:311–319, 1974. doi:10.1016/0011-7471(74)90102-8

    work page doi:10.1016/0011-7471(74)90102-8 1974

  60. [60]

    J. D. Isdale.Viscosity of simple liquids including measurement and prediction at elevated pressure. PhD thesis, Dept. Chem. Process Eng., Univ. of Strathclyde, Glasgow, U.K., 1976

    1976

  61. [61]

    H. U. Sverdrup, M. W. Johnson, and R. H. Fleming.The Oceans, Their Physics, Chemistry and General Biology. Prentice-Hall, 1942

    1942

  62. [62]

    Which training methods for gans do actually converge? InProceedings of the 35th International Conference on Machine Learning, volume 80

    Lars Mescheder, Andreas Geiger, and Sebastian Nowozin. Which training methods for gans do actually converge? InProceedings of the 35th International Conference on Machine Learning, volume 80. PMLR, 2018

    2018

  63. [63]

    Deconvolution and checkerboard artifacts

    Augustus Odena, Vincent Dumoulin, and Chris Olah. Deconvolution and checkerboard artifacts. Distill, 2016. URL: http://distill.pub/2016/deconv-checkerboard, doi:10.23915/ distill.00003

    2016

  64. [64]

    Spectral normalization for generative adversarial networks.CoRR, abs/1802.05957, 2018

    Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks.CoRR, abs/1802.05957, 2018. URL: http://arxiv.org/ abs/1802.05957,arXiv:1802.05957

    Pith/arXiv arXiv 2018

  65. [65]

    Strided convolution instead of max pooling for memory efficiency of convolutional neural networks

    Riadh Ayachi, Mouna Afif, Yahia Said, and Mohamed Atri. Strided convolution instead of max pooling for memory efficiency of convolutional neural networks. In Med Salim Bouhlel and Stefano Rovetta, editors,Proceedings of the 8th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT’18), Vol.1, pages ...

    2020