Skytrak+ Face angle and club path

[Lombo]

I've been really interested in the new Skytrak+ mainly because of the club data offered. In this pricerange, you're not really going to get that information which in my mind is a gamechanger. However, I have not read or heard with everything that I've viewed on how EXACTLY the measurements related to club data are generated. I read a lot about AI generated. The way the term AI gets thrown around today is all over the place. I keep seeing how Golftec has imputed their expertise to the algorithm (and I assume this being related primarily to club data) but again, nothing specific. If it was directly measured, we would know that as Skytrak would be not hiding that, but they aren't saying that at all. On the Skytrak website they say that the unit DELIVERS this data - and it then also says machine learning algorithms are added to that data. What does that EXACTLY mean? So it's not measured for sure - I've also heard that its accurate to about 1.5 degrees and hopefully in the future it can be around 1 degree? What does that mean and how does machine learning come into play? I've also read that even though they could offer impact location, they are not at this time - why not? Again, I am really interested in the product and would definitely consider this unit when it's time for me to upgrade from my present Skytrak, but a little more transparency here would be appreciated -

[Bubba22]

Cheers Lombo. You’re asking some very good questions which we may never get a direct answer. My understanding is that the club data is calculated. I think we need to understand that everything in golf simulation is calculated. The difference is that units that calculate with more direct measurements will tend to be more accurate than indirect measured or non measured data. Club data can be accurately measured if from ball data but certain assumptions need to be made and gear effect etc will affect the data. I think the Skytrak plus will likely measure swing speed and maybe angle of attack. The AI model will likely provide feedback only if the true club data is known to allow for proper learning. So I am not sure how AI works here but they obviously have algorithms that probably are close but who knows.

[Morini]

My guess is that it uses 3d reconstruction method as outlined in this Rapsodo patent. This is patent is for a down the line device, but there is a good chance that they method is similar.

I don't think it is jumping on the AI bandwagon, and is clearly a very clever way of getting near high end LM results using cheap hardware.

I will let you read the patent, but as an analogy, let's assume you wanted to measure your table tennis stroke data. You hook up a decent webcam of some sort and swing away. You look at the video, and it is not ideal kind of blurry, no stickers, ... so can't make x,y,z measurements in consecutive frames to "measure" the racket movement.

But you do notice that similar swings look very similar on video, blurry or not. So instead of measuring, create a reference library and compare. Creating that reference library would be a pain, as you'd need markers and high speed video (in addition to the cheap camera) to create a database of properly measured swings to compare against. So you model the racket in your favority 3d software, and use Unreal Engine to execute a variety of virtual swings. Now either simulate those swings captured with a crappy camera (using a filter of some sort), or even simpler, just film your screen with the actual crap camera, and save the virtual 3d motion data from the simulated swings. There is your reference library, thousands of tagged swing. Now you can film your regular table tennis shots, and the AI will predict "racket data" from the reference set.

Maybe not exactly like that, but along those lines.