Surprisingly, there are only two physical wind tunnels capable of speeds that can cater to cyclists located in the United States. And given their scarcity, it’s no surprise that renting one is both costly and time consuming.
But what if all you needed was a short video— no travel, no expensive facility, just software handling the rest? That’s exactly what WindTunnel.AI is bringing to the cycling world. By leveraging radiance fields, specifically NeRFs, they can simulate a wind tunnel effect on a cyclist, providing precision aero data without the need for a physical facility.
Aerodynamics plays a crucial role in cycling performance. Even marginal gains, small reductions in drag, can translate into significant improvements in speed and efficiency. Cyclists obsess over these details, fueling a multi-billion-dollar industry where riders spend thousands on optimized gear, bike fittings, and wind tunnel sessions. But despite the demand, access to real wind tunnel testing remains extremely limited, with long wait times and high costs.
WindTunnel.AI was founded by a team of competitive cyclists who knew firsthand that wind is the biggest obstacle in the sport. An invisible force that accounts for the majority of resistance a rider faces. What started as a curiosity about scanning their own bike positions and running computational fluid dynamics (CFD) tests soon turned into a high-fidelity, scalable solution.
The company evolved from an early concept that involved mapping entire mountain bike courses using NeRFs, but the team soon realized that hosting and processing such vast datasets wasn’t viable. Pivoting, they focused on refining a video to NeRF to CFD pipeline that could reconstruct accurate aerodynamic models from simple video footage.
NeRFs have proven incredibly powerful for dense 3D reconstruction, often outpacing traditional multi-view stereo (MVS) techniques in speed and detail. But creating a 3D model isn’t enough. Aerodynamic testing requires a watertight, 3D-printable mesh, not just a point cloud.
“To go from NeRF to CFD, we needed a high-fidelity, manifold mesh that traditional meshing algorithms struggled to produce,” explains Eric Semianczuk, CEO of Wind Tunnel, “After countless iterations, experimenting with everything from Gaussian Splatting to OpenVDB to neural networks, I developed a workflow that extracts a CFD compatible mesh from NeRF-based reconstructions.”
By automating much of this process, the team created a pipeline capable of running large volumes of tests efficiently, making this technology accessible to individual cyclists, not just elite teams or manufacturers.
WindTunnel.AI is changing that by offering a Cycling Tuned Aero Test for just $30, a fraction of the cost of traditional wind tunnel sessions. Wind Tunnel AI pairs NeRFs with OpenFOAM.
What You Get for $30
Aerodynamic Drag (CdA) Results: 11 aero drag values in watts/CdA, measured across 11 different wind yaw angles.
Two Data Visualizations:
A graph showing raw CdA values across 11 yaw angles.
3 different wind speeds (50 km/h, 36 km/h, and 27.4 km/h).
A graph displaying power loss (watts) based on drag data.
As AI-powered Physical Simulation becomes more prevalent, something NVIDIA has been pushing in its latest messaging. The boundary between the digital and physical worlds is blurring. WindTunnel.AI is at the forefront of this shift, bringing aerodynamic insights to everyday cyclists through the power of Radiance Fields and AI-driven simulation.
Right now, radiance fields and CFD operate as separate technologies, linked only through the process of generating accurate 3D meshes. But with rapid advancements in research, that could change.
“Future models could be designed specifically for CFD compatibility, or CFD systems could evolve to handle NeRF-based 3D representations directly,” speculates Semianczuk. “Given the importance of aerodynamic modeling, I can only see this technology continuing to evolve and accelerate.”
More information can be found on Wind Tunnel’s website.
New to Gaussian Splatting? Start here
