An international team including researchers from the University of Washington, University of Michigan, RWTH Aachen University, and the Technical University of Munich has developed HydroGym, a platform for training machine learning models to actively control fluid dynamics. The system, detailed in a study published in Nature, demonstrated a 38% reduction in surface friction and 11% drag reduction on a simulated airplane wing, using reinforcement learning to optimize control strategies.
International Collaboration Behind HydroGym
The HydroGym project brings together researchers from the University of Washington (UW), University of Michigan Engineering, RWTH Aachen University, and the Technical University of Munich (TUM), with funding from the U.S. National Science Foundation, Boeing Co., and other institutions. The platform aims to address the complexities of fluid dynamics, which are critical to industries like energy, transportation, and defense. Fluid flows are central to several trillion-dollar industries,
said Steven Brunton, a senior co-corresponding author of the study and Boeing Professor in AI & Data-Driven Engineering at UW. An improved ability to understand and control these flows could have an immense economic and ecological impact.
The collaboration focuses on reinforcement learning, a machine learning technique that trains AI agents through environmental interactions. This approach has previously transformed fields such as protein folding and nuclear fusion. By integrating physics knowledge into AI training, HydroGym reduces the trial-and-error required to optimize control strategies by up to 65%. I hope this helps move the field from individual demonstrations towards a more systematic and collaborative approach to discovering general principles for controlling complex flows,
said Christian Lagemann, first author of the study and former postdoctoral researcher at UW under Brunton.
Reinforcement Learning in Fluid Control
HydroGym tests active methods for controlling fluid flows, such as shape morphing, tiny flaps, or spinning elements, to reduce drag, improve lift, and manage heat. The platform’s design allows researchers to train models in inexpensive surrogate environments and test them in realistic scenarios. One demonstration involved a channel with holes, similar to an air hockey table, where a machine learning model controlled air flow to minimize friction. The same model was then applied to a simulated airplane wing, reducing surface friction by 38% and overall drag by 11%.
Instead of developing controllers for isolated flow problems with no common framework for comparison, we can now study how control strategies transfer across different geometric shapes and types of flow,
Lagemann said. The team’s work highlights the potential for AI to generalize across complex systems, a challenge in traditional fluid dynamics modeling.
Zero-Shot Transfer and Real-World Applications
A key breakthrough in HydroGym’s development is zero-shot transfer,
where models trained on simple geometries achieve high performance on complex scenarios without additional training. This capability was demonstrated by applying a controller developed for a basic channel to a simulated airplane wing. One of the key findings is zero-shot transfer: in other words, learning in simple geometries to distill the key physics, and deploying the models in very complex geometries with very high control performance,
said Ricardo Vinuesa, co-corresponding author of the study and U-M associate professor of aerospace engineering.
Funding and Future Directions
HydroGym’s development was primarily funded by the U.S. National Science Foundation and Boeing Co., with additional support from the University of Michigan and other institutions. The project builds on years of research into fluid mechanics and machine learning, with collaborators including Wolfgang Schröder, professor of fluid mechanics at RWTH Aachen, and Nikolaus Adams, professor of aerodynamics and fluid mechanics at TUM.
The team plans to expand HydroGym’s applications, focusing on areas like noise reduction in jet engines and cooling systems for supercomputers. While the platform’s immediate goals are industrial, the underlying research could influence broader scientific fields, from climate modeling to biomedical engineering.
The success of HydroGym underscores the growing synergy between AI and traditional scientific disciplines. As researchers refine these tools, the potential to revolutionize fluid dynamics control—spanning everything from aircraft efficiency to environmental sustainability—becomes increasingly tangible.
