HydroGym: 60+ validated RL fluid-dynamics environments, zero-shot transfer to 3D wing

bravo_abad · x · 2026-08-20

Christian Lagemann and coauthors introduce HydroGym, a large standardized platform for applying reinforcement learning to fluid dynamics, with 60+ validated environments spanning laminar and turbulent flows, 2D and 3D geometries, multiple Reynolds numbers, and different solver backends.

The contribution goes beyond another RL controller: it enables a harder question — can an RL policy learn transferable physical control principles rather than overfitting to one simulation? The authors train a multi-agent controller only on a computationally cheap turbulent channel-flow surrogate, then deploy it zero-shot to the boundary layer of a 3D wing, demonstrating direct transfer from cheap simulation to a much harder physical system.

Related event: HydroGym Lands in Nature: 61+ Fluid RL Benchmarks with Zero-Shot Wing Transfer(5 posts)→

Original post →

More from Embodied

Embodied channel →