% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@INPROCEEDINGS{Lagemann:1041823,
      author       = {Lagemann, Christian and Rüttgers, Mario and Gondrum, Miro
                      and Meinke, Matthias and Schröder, Wolfgang and Lintermann,
                      Andreas and Brunton, Steven L.},
      title        = {{H}ydro {G}ym-{GPU}: {F}rom 2{D} to 3{D} {B}enchmark
                      {E}nvironments for {R}einforcement {L}earning in {F}luid
                      {F}lows},
      volume       = {69},
      address      = {Jülich},
      publisher    = {Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag},
      reportid     = {FZJ-2025-02455},
      series       = {Schriften des Forschungszentrums Jülich IAS Series},
      pages        = {57-63},
      year         = {2025},
      comment      = {Proceedings of the 35th Parallel CFD International
                      Conference 2024},
      booktitle     = {Proceedings of the 35th Parallel CFD
                       International Conference 2024},
      abstract     = {Fluid flow modeling and control is a significant modern
                      challenge with potential impacts across science, technology,
                      and industry. Improved flow control could enhance drag
                      reduction, mixing, and noise reduction in areas like
                      transportation, energy, and medicine. However, progress in
                      flow control is currently hindered by the lack of
                      systematically standardized benchmarks and the high
                      computational cost of fluid simulations. While
                      two-dimensional problems have been extensivelystudied,
                      three-dimensional simulations with larger meshes are rarely
                      considered due to the need forhighly parallelized and
                      specialized solvers. As a result, the engineering burden of
                      encapsulating thesesimulations in benchmark environments has
                      proven to be a significant barrier. In this paper, a
                      GPU-based extension of the HydroGym platform coupling the
                      multiphysics solver framework m-AIA witha state-of-the-art
                      reinforcement learning platform is presented for fluid flow
                      control problems. Basedon the highly-parallelized lattice
                      Boltzmann solver, which is part of m-AIA, a new set of
                      three-dimensional, non-differentiable fluid flow
                      environments is added that extend existing flow
                      controlchallenges to a new level of physical and
                      computational complexity.},
      month         = {Sep},
      date          = {2024-09-02},
      organization  = {35th Parallel CFD International
                       Conference 2024, Bonn (Germany), 2 Sep
                       2024 - 4 Sep 2024},
      cin          = {JSC},
      cid          = {I:(DE-Juel1)JSC-20090406},
      pnm          = {5111 - Domain-Specific Simulation $\&$ Data Life Cycle Labs
                      (SDLs) and Research Groups (POF4-511) / HANAMI - Hpc
                      AlliaNce for Applications and supercoMputing Innovation: the
                      Europe - Japan collaboration (101136269)},
      pid          = {G:(DE-HGF)POF4-5111 / G:(EU-Grant)101136269},
      typ          = {PUB:(DE-HGF)8 / PUB:(DE-HGF)7},
      doi          = {10.34734/FZJ-2025-02455},
      url          = {https://juser.fz-juelich.de/record/1041823},
}