Limitations

Last Updated: 4/3/2024

At our core, we are a measurement company. No measurement tool is perfect, but you can expect from the toolmaker to be clear of what it can and can not do.

Below we keep a regularly updated list of our known limitations. As we find additional information or  improve approaches, we will  edit this list. If you experience or are aware of issues not listed here, please reach out to support@airo.app. We have grouped them by category.

• Digital Aero Twin

  • Limited support for para athletes: AiROs sister company, Inspire Gold is proud to consult with multiple paralympic champions and athletes. The way our Digital Aero Twin works, we currently do not natively support the creation of Digital Twins with missing limbs, significant limb length discrepancies, or atypical ranges of motion.

  • Body shapes not represented in our training data: AiROs Digital Twin is trained on a dataset of thousands of 3D scans of people from the general population. The more an athlete’s morphology diverges from the scans represented in the data set, the harder it is for our technology to build a representative model. We have found challenges building twins of champion body builders and cyclists with a very pronounced difference in lower to upper body musculature. Unfortunately, our dataset also contained limited scans of highly trained competitive female cyclists. We are planning to collect a more representative training data set in the future.

  • Stomach flex and breathing: Most people contract their stomach muscle when someone takes a picture of them. When on the bike in cycling pose, some riders  also flex their stomach muscles, while some riders relax their stomach  muscles. Chest and waist diameter changes substantially during a breathing cycle. While those settings are fine-tuneable in  Digital Aero Twin Settings, we made the choice not to add a breathing slider to the Position Editor.

  • Limited foot range of motion: To not complicate position adjustments and allow us and our customers to focus on upper body changes our foot currently does not angulate as part of the pedal stroke. However, this means for some bike positions and for riders with strong foot angulations we don’t fit foot poses well. We also don’t allow for cleat adjustment on the foot (also done to retain simplicity). We will look to find a better solution that retains ease of use, while fitting to more foot parameters.

  • No wrist degrees of freedom: We are aware triathletes and time trialists care about their hand positions and extension choice. We decided to keep the number of sliders to less than 20 for the initial version. We plan to roll out a different menu with “advanced settings” so we can serve customers that want a simple UI and those that want full functionality.

  • Digital Aero Twin does not wear shoes: We do not expect a shoed version of to lead to different fit or helmet recommendations, so we have not prioritized this feature. Let us know if this is an issue somehow.

  • No clear definition of saddle setback: Since we only simulate the rider, there is no saddle. There are different saddles, and people sit different on different saddles, so our saddle setback is an approximate of a typical saddle.

• Helmets

  • No through flow: As we are focused on drag right now and aim to provide fast and affordable calculations, we decided not to mesh the internal vent structures for now and not simulated through flow through the helmet. Results are most representative for people with curly or significant hair that blocks the internal vent structures.

  • Only one helmet size:

    Of the helmets we have that come in different sizes we currently only have one size in the app. We are considering expanding to different helmet sizes in the future, once our user interface allows to keep that all well managed.

  • No helmet straps or fit system:

    Our 3D scans do not include helmet straps or fit systems.

• Other Equipment

  • No bike: We chose to simulate the poses without a bike and report the CdA and power values with a fixed offset accounting for bike CdA. You can set bike CdA offset in settings. We made that decision since adding a bike would have tripled simulation time and cost, while not having significant effect on pose recommendations. We confirmed this through spot tests of position recommendations generated from simulations with and without bikes.

  • No skinsuit / aerosocks: Our Digital Aero Twin is as a uniform smooth texture. We do not account for clothing wrinkles, seams or textured fabric. The results are most representative of an athlete wearing a smooth skinsuit or traveling at pre-transition speeds of the skinsuit. High performance skinsuits can reduce the drag of the upper arms so for athletes wearing those skinsuits we estimate our simulations overpredict the drag and of the arms. The same applies to aero socks.

• CFD

  • Straight head wind only (for now): All simulations on AiRO happen with a straight head wind. In the future you will be able to choose from zero degree only, or a yaw sweep, where the yaw sweep will have to cost additionally, since it increases the amount of simulations five or seven-fold. From our experience, differences between helmets or positions at zero-degree yaw are highly corelated to the differences at five degrees. At ten and fifteen degrees the differences are starting to diverge more, but those yaw angles are a small fraction of the total yaw distribution, outside of explicitly windy races.

  • One wind speed only: To accurately model the effects of different wind speeds on our positions we would need a transition model, a way to user specify surface textures and a wind speed model. While we plan to add transition models in the future, we decided to prioritize easy of use and minimal settings, and using a turbulence modelling approach that has significant academic validation.

  • No transition model: We are using the k-omega SST turbulence model since it is the defacto standard for sports aerodynamics and has shown to accurately simulate differences in body position in multiple peer reviewed academic papers. There are now approaches to add a transition model, and two papers have shown promising increases in accuracy. Our current validation work was done without a transition model, and we are content with the results, so we will wait until us and academia can validate the accuracy and robustness of the transition models further before switching. Our goal was to be able to offer simulations for less than $10 and around five minutes, and we felt the additional cost and time of a transitional model was not offset by the small increase in accuracy.

  • Not pedaling: in our default simulation mode our avatar is not pedaling. When running simulations, AiRO automatically places the legs at a crank angle of 30 degrees, which in multiple published studies has shown to be the static angle that most represents pedaling. We can approximate pedaling closer by averaging four crank position, but this is currently not a released feature, since it would increase the cost per simulation by a factor of four. If there is interest among our customers, we can release this model.