Skytrak Plus Club Data

[Bubba22]

So the new Skytrak Plus offers improved camera measurement, larger hitting area, shot optimizer and shot score as well as calculated club data. Hopefully, the Skytrak folks can come on here and enlighten us on the measured and calculated data. It is both dual radar and 2 cameras. What are your thoughts on the calculated club data?

[OttawaP]

Yes I agree. Some sort of clarity would be really helpful. I’m not against calculated numbers if they are actually accurate. I imagine we will start seeing some skytrak+ vs eye xo type videos pretty soon which should end the discussion pretty quickly.

[Bubba22]

I agree OttawaP. Club data comparison is hard between units as it will depend on alignment, interference etc to be confident on differences. However you are right that comparisons will be nice. With ball and club data everything is eventually calculated. Even ball speed is calculated with distance and time. Smash factor is ball speed/club speed. Etc etc Obviously the more data that is measured, assuming calculations are right, then the more confident one is on the calculated data.

[l40niners1]

Pretty disappointing but not unexpected. In a98cr125's latest video he said that the margin of error in club data calculation is 1 degree for irons/drivers. Which in golf can be a pretty big difference.

I don't see the Skytrak+ being a unit most people would upgrade to. BLP now costs about $2400 brand new with the 20% discount code so paying the difference in yearly membership fee ($250-$100=$150 yearly difference) and having a year free with BLP it would take 5 years of ownership for the BLP to cost more. In those 5 years I'd bet there would be a better option to upgrade to at a lower price point.

[Bubba22]

We have discussed this before as ST was going to add calculated club data to its original unit not that long ago.

https://golfsimulatorforum.com/forum/skytrak/148407-poll-skytrak-calculated-club-data-inclusion

[Bubba22]

The big issue with fully calculated club data is getting accurate data with respect to gear effect, especially with the woods. If you hit it dead center all the time then estimated club data can be good. Without knowing the face contact, I think it is very difficult to do.

[Lombo]

I referenced my concern that calculating club data based on ballflight was a tough estimate because of having to assume ballstrike quality - and this is the answer I received below:
It leads me to think that the "machine learning" is going to play a very large role in the ST+ calculations for clubhead data.



"ST+ does not assume a centered strike on any driver shot. The training set and techniques used could provide impact location, but at this time we've chosen not to. Hope that helps with your concern".

[Morini]

The big issue with fully calculated club data is getting accurate data with respect to gear effect, especially with the woods.

Not sure how big of an issue gear effect is in terms of the + getting club data. It is enormous for any attempt at using only ball data. It is also an issue for Doppler indoor because those units can't see enough ball flight to indirectly measure spin axis. So they back it out from path and HLA. But that won't work without accounting for gear effect. As mentioned, that is for dopplers. A photometric unit can measure spin axis directly. Foresight does that, which of course requires stickers and multiple high frame rate cameras ($$$).

But the cheaper and less capable photometric units could also try to back face out from path and spin axis. And that is where gear effect can complicate things. The question comes down to exactly how Skytrak + measures/calculates face.

As Bubba alludes to, everything is "calculated" to some degree. Skytrak OG CHS was an extreme case, where there NO measurement of clubhead, only indirectly measured via ball data. But calculated does not necessarily mean inaccurate. It depends on how many indirect measurements you have, and the constraints.

It is not confirmed, but there are enough clues out there right now to conclude that the + does indeed use the "3d reconstruction" method of obtaining club data that is described in the Rapsodo patents. A poster named Nicholas Bufano posted those here about 2 or 3 years ago, and he ended up being bang on. I felt that since the patents described a DTL device, they were intended for Rapsodo's own devices, not a Skytrak follow up. I think we will both be proven correct as they will use the tech for both.

Essentially what it does is to capture the clubhead using stereo 240 fps cameras. The clubhead will be a blur ... but it will be a consistent blur. Swing the exact same club at the exact same speed, AoA, path and face, and the blurry image will be very similar if not identical. Now using simulation (ie 3d rendering) create thousands of simulated captured swings. Then when your cameras capture a blurry image, let AI loose on it, and it will determine club data based on how well it matches the reference images. Maybe not exactly like that, but along those lines. Quite brilliant to be honest.

Back to the question about face. So is it measured or calculated? If calculated, how is it backed out, and what assumptions are made? It sounds as if the "3d reconstruction" method can on its own come up with a number for face ... but given how sensitive the number is to the timing of moment of impact, I don't think that is sufficient. It is very likely that they also use other indirect measures as well. It is likely that 3d reconstruction gives a decent measurement of impact location. With that and path and spin axis, face can be calculated. It can also calculate face with reasonable accuracy indirectly from from HLA and path (would need to know or at least estimate club static loft, d-plane etc).

I am guessing wildly, but I think I am on the right track. The million dollar question: How accurate is it? And is the error normally distributed or does it work better for certain swings than others? Nick Clearwater, Golftec VP of player develoopment let it be known that he is hoping for accuracy to improve to half a degree.

The best case scenario as I see it is that Skytrak+ will shrink the error to around that, and will also at some point report AoA and impact location. I think we can also safely assume that the ball data is very accurate. I would be shocked if it isn't. If this best case scenario plays out ... it will be interesting, as stickerless accurate complete club data will be very appealing to serious game improvement golfers. If I am correct about the hardware, Skytrak + does not have expensive high frame rate video, and instead uses two cheap(er) 240 fps cameras. As such, it would easily win any price war against Foresight and Uneekor.

[Bubba22]

My assumption Morini was that the 3D patent was for only DTL and would be used with the MLM2 to provide club data. That’s why I predict that I’m a future update the MLM2 will have a subscription upgrade for those that want club data. I suspect this can be used for the Skytrak plus as well but just from the side. The great thing with the MLM2 is the down the line camera view is similar and better than the Mevo plus and shows face contact. So now you have spin axis, face contact and definitely with AI all other club data can be somewhat concluded. Perhaps Skytrak plus is going to do it the same way but just from the side. Funny it doesn’t get angle of attack?

As well, I do suspect other units like the upcoming ProTee units will use overhead images and AI with 3d rendering to measure/calculate club data.

[Morini]

Perhaps Skytrak plus is going to do it the same way but just from the side. Funny it doesn’t get angle of attack?

Yes, it clearly sounds like it. The + is designed by Rapsodo of course, but there has been multiple clues posted by insiders that make it very likely to be a side on version of this. I will post some of those comments when I have a minute. I believe it has AoA but does not report externally ... yet. But who knows.

[Galf55]

Have there been any testing of club data against gcquad, hmt, or tm?