Dr. Kwon’s DG 3D motion study

[semisensei]

I was stoked to see this week that Dr. Kwon himself is embarking upon a motion study of the disc golf swings. Looks like he has many of the top MPO athletes in his lab at TWU getting recorded with 3D motion capture. I don't know much more than the fact that Dr. Kwon is running the study, but I'm looking forward to seeing some results!
If anyone else has any info on the study, please post it here!

https://www.instagram.com/tv/CpfuMh9KhvS/?igshid=YmMyMTA2M2Y=

 

[Mocheez]

https://www.instagram.com/reel/Cpd75jNJZ-r/?igshid=YmMyMTA2M2Y=

 

[Sheep]

This is good that they are doing pro players finally.

Cause the previous data was from people with really really bad form.

 

[Brychanus]

big wigg 360 incoming:
https://www.instagram.com/p/CpgOg3KJmI8/

 

[drk_evns]

I wonder if eventually we'll be able to ask AI to generate the perfect disc golf swing based off of this data (and future data).

 

[sidewinder22]

I wonder if eventually we'll be able to ask AI to generate the perfect disc golf swing based off of this data (and future data).

While all this data is fascinating and should be useful in some capacity, I'm a little leery of what kind of form extremes Skynet would recommend purely based on the mathematics. Everybody is built different, so it would have to somehow take that into consideration. Even the Swing Catalyst guys talk about how messy the data is from ball golfers and how changing the same thing for two different players can yield opposite results.

IMO there is an art to the swing.

 

[Brychanus]

Can you imagine the mess it would make if you tried to make Eagle throw exactly like GG or vice versa?

 

[Sard0nyx]

Can you imagine the mess it would make if you tried to make Eagle throw exactly like GG or vice versa?

I'll bet Eagle could throw like GG. But not the other way around.

 

[Brychanus]

I'll bet Eagle could throw like GG. But not the other way around.

This could actually be a fun learn-by-speculation-and-debate thread. I can't throw so I'll vent my energy by stirring it up!

Observations:
GG:
- Is taking advantage of his mass being more centered in his thiccc body around his core. So that vertical hop+drop with gravity is especially useful for his body type.
- Accordingly, his form emphases the increase in momentum he gets like a ballerina revolving around the more vertical axis when he plants.
- This form also helps him get the most out of his good but not absolute top-tier levers.
- He's got a very late leverage boost on the disc due to his posture & arm mechanics, especially with the late pronation at the wrist. Very large angle at the shoulder entering the hit.

Eagle:
- Has longer levers (I think in the relative and absolute sense).
- Stretching out horizontally helps him maximize the move against the doorframe lagging in that dimension. His form also takes advantage of his very long legs leveraging him into the plant. Notice how long his feet are too. Wow.
- Incredible mobility/flexibility within and between his upper and lower body (GG's got plenty too, but Eagle always reminds me of Gumby. I mean holy ****).

Speculating/questions:
- I'd think GG would probably not get nearly as much out of a very horizontal form like Eagle's. He lacks the levers and would lose the boost he gets from his beefy mass dropping into the plant more vertically.

-I suspect Eagle might be able to pull off a vertical form like young Simon (or could have if he started working on it young).

-However, Eagle still might get more out of his horizontal form than a vertical form for power due to his extreme leverage advantages moving in that direction (?). Perhaps even more than Simon (who has said Eagle throws at least 50' farther than him head-to-head) (?)

-Eagle's flexibility possibly allows him to get even more lag/separation from the lower to upper body than GG moving horizontally, which might be another reason GG's form invests in the vertical component (?).

-There are a lot of other differences between these two guys. Lever sequences, joint anatomy, etc all looks different to me when looking closely. Eagle even uses that weird "suitcase" grip that you rarely see elsewhere - just form, or also anatomy?

These are some of the things that I suspect contribute to the "art" that SW is referring to.

 

[drk_evns]

I think that's a good perspective, and probably true in the short term, but after using a few different current AI's for work I'm not ready to underestimate what it will be able to do in the (relatively near) future.

With more and more data like this, I bet it could make specific recommendations BASED on your body type. Imagine a time when it's able to take a look at video of your form and compare it to the 3d data it has. It could build a form for you based on your body and current form. What if it creates an invisialine-style series of small adjustments that quickly transitions your form to the idealized version? I don't think this is far off. It will probably start (or probably already has started) in more mainstream sports.

_

I also think the idea that GG couldn't throw like Eagle (and vice versa) is really interesting. My intuition tells me that if they had the same amount of time to develop the other form as they did their own, they would both still have success. Hard to know if even the top throwers are all throwing the ideal way for their bodies.

Would GG, Eagle, or Tamm throw further if they did what David Wiggins Jr. does?

 

[Brychanus]

In terms of form variants I'd guess each top player could all take each one pretty far if they had taken similar time to develop, but there if we're talking about maximizing/peak potential the fit-to-body is probably very important (e.g., the details of how GG vs. Eagle get their 600'+ probably matter and may differ as a function of where the forces & leverage come from). I also think it would matter if you stringently mean the "exact same" form, and if that's even a coherent concept in two different players' bodies.

I would totally believe that even top throwers in disc golf might not have settled into a theoretical "optimal" form. E.g., baseball has had more than 100 years of development, and even with a lot of money invested in sports science and expert mechanics coaches there are still innovations. Fitting techniques to individuals can earn you good money* if you're very good at coaching it. Whether we're mostly just seeing the "floor" rise there and we're at the "ceiling" at this point for MLB pitchers is interesting. To think that disc golf is really not far from the beginning of that process is incredible. Top form can drift and backslide too in MLB pitching and in disc golf.

I'm a little skeptical that AI models trained on the skin-surface contact electrodes would be enough to get at some of the important differences of the musculoskeletal system etc. If there are strong ordered relationships between motion patterns and underlying musculoskeletal anatomy an AI could probably pick it out in principle. I'm not sure how detailed the data would need to be to solve the optimizing problem for individuals (please continue speculating!).


*It was hard to find reputable sources, but circa 2018: "Typical salaries for pitching coaches, one of the higher-paid staff coaching jobs, reportedly range from $200,000 to $360,000."

 

[Sheep]

In terms of form variants I'd guess each top player could all take each one pretty far if they had taken similar time to develop, but there if we're talking about maximizing/peak potential the fit-to-body is probably very important (e.g., the details of how GG vs. Eagle get their 600'+ probably matter and may differ as a function of where the forces & leverage come from). I also think it would matter if you stringently mean the "exact same" form, and if that's even a coherent concept in two different players' bodies.

I would totally believe that even top throwers in disc golf might not have settled into a theoretical "optimal" form. E.g., baseball has had more than 100 years of development, and even with a lot of money invested in sports science and expert mechanics coaches there are still innovations. Fitting techniques to individuals can earn you good money* if you're very good at coaching it. Whether we're mostly just seeing the "floor" rise there and we're at the "ceiling" at this point for MLB pitchers is interesting. To think that disc golf is really not far from the beginning of that process is incredible. Top form can drift and backslide too in MLB pitching and in disc golf.

I'm a little skeptical that AI models trained on the skin-surface contact electrodes would be enough to get at some of the important differences of the musculoskeletal system etc. If there are strong ordered relationships between motion patterns and underlying musculoskeletal anatomy an AI could probably pick it out in principle. I'm not sure how detailed the data would need to be to solve the optimizing problem for individuals (please continue speculating!).


*It was hard to find reputable sources, but circa 2018: "Typical salaries for pitching coaches, one of the higher-paid staff coaching jobs, reportedly range from $200,000 to $360,000."

A fun fact to point out.
GG used to be about 80-100 lbs heavier.

So his form developed from a lot bulkier frame than he currently has.

Though per the discussion in person yesterday, I'd not say his form is "atrocious" as the person I spoke with said. Because form is about building a kinetic chain with your body structure, not necessarily "certain special motions."

Were mainly looking for "key points" and "key leverage" area's and how to efficiently and safely get to those points with the available flexibility, fitness and other factors.

https://youtu.be/AnaASTBn_K4
https://youtu.be/3HABgm-UUi0

There is unfortunately limited material out there to reference, but this stuff is great here, because they are using the same system to do what the disc golf study is, but the main difference is the people doing it in these video's are far smarter than the people doing the disc golf stuff. I know that's a bold statement, but It's what I believe.
Not saying the people doing the disc golf thing are dummies by any means, but to each their own.

There is tons of data we can get from it, but it also depends on how they share, document and ask the questions then use that data.

And as Brychanus here also mentions, were still in younger stages of understanding drawing from every parallel we can.

An AI model can eventually give us "some" data, but that still also depends on to many factors that an AI model cannot necessarily represent in a data to practice situation.

What we can use this data to figure out is optimal body shapes for instance. Such as using the data to figure out where the best place to pocket is. Where the best trajectory point is. Where certain parts of the power need to be added. But this all also depends on if the AI can create that data or not.

Plus another huge factor in how we throw is the friction of throwing a disc with the hand, everyones different grip and flexation of the wrist.

 

[drk_evns]

I don't know when, but I bet we will see AI/machine learning used to identify individual form inefficiencies and give recommendations based on ideal form for disc golf. It's already happening in other places.

AI doesn't even need physical trackers to analyze video, and if we imagine a model trained on every disc golf video in history, it will already have more insight than any human alive.

Convolutional Neural Networks (CNNs): CNNs are commonly used for image recognition and object detection, making them ideal for analyzing video footage of athletes. They work by extracting features from each frame of the video and using these features to identify patterns and objects.

Recurrent Neural Networks (RNNs): RNNs are useful for analyzing sequences of data, such as a series of video frames. They can be used to identify patterns in an athlete's movements over time and can help predict future movements.

Generative Adversarial Networks (GANs): GANs are a type of neural network that can generate realistic synthetic data. They can be used to create synthetic video footage of an athlete's movements, which can be used to augment the training data and improve the accuracy of the AI model.

Pose Estimation Models: Pose estimation models use computer vision techniques to estimate the 2D or 3D pose of an athlete's body from video footage. These models can help identify the position of each joint and limb, allowing coaches and trainers to analyze an athlete's movement mechanics and identify areas for improvement.

That's just what exists now.

I think a human understanding of the swing is an art in the same way religion is a human understanding of life.

 

[Brychanus]

I don't know when, but I bet we will see AI/machine learning used to identify individual form inefficiencies and give recommendations based on ideal form for disc golf. It's already happening in other places.

AI doesn't even need physical trackers to analyze video, and if we imagine a model trained on every disc golf video in history, it will already have more insight than any human alive.



That's just what exists now.

I think a human understanding of the swing is an art in the same way religion is a human understanding of life.

Totally. I note that each one of those available techniques and their relatives usually still need experiments, external validation, & outcome criteria to evaluate their performance (often mediated through expert human brains). Unless there's something fundamentally different happening in sports science & kinematics than neuroscience & cognition.

 

[DannyS]

Plus another huge factor in how we throw is the friction of throwing a disc with the hand, everyones different grip and flexation of the wrist.[/QUOTE]


I believe that this is the most undertalked about component of disc golf mechanics.

 

[Lumberjack504]

I don't know when, but I bet we will see AI/machine learning used to identify individual form inefficiencies and give recommendations based on ideal form for disc golf. It's already happening in other places.

AI doesn't even need physical trackers to analyze video, and if we imagine a model trained on every disc golf video in history, it will already have more insight than any human alive.



That's just what exists now.

I think a human understanding of the swing is an art in the same way religion is a human understanding of life.

I'm only really familiar with CCNs and image recognition, so grain of salt and all that, but how would this work in only a 2 dimensional pixel space? In terms of the neural network analyzing pixels, that is.

 

[Polecat132]

There are some pose estimation models like mediapipe that estimate 3D coordinates. It's not super precise yet, but you can make some pretty cool stuff with it like this:

 

[Sheep]

Plus another huge factor in how we throw is the friction of throwing a disc with the hand, everyones different grip and flexation of the wrist.

I believe that this is the most undertalked about component of disc golf mechanics.

The grip is something that isn't a standard.
Were learning that the power grip might not be the best thing.

Now were talking about front and back loads, were talking about different hand positions. I've experimented a lot with grip in my struggle for nose angle due to wrist problems. I can throw "sorta" more nose down with a traditional grip, but not like the pro's, so my shots go out 350-400 and stall vs drive. That's a nose up throw.

These are things that the computer cannot really see/understand and such.

We can determine a best angle of attack with modeling and wind tunnels. (yes I'll get it built at some point, to many projects stacking up, paying bills is first)
so we can figure out optimum angle to get it out of our hand to angle of attack.
But how much wrist curl, etc.

It's a complicated topic thats really simple. It's what makes it interesting.
The details are really complex, but to perform them, its all really simple and we gotta make sure to not over complicate things to much. Because our body will do whats important, we gotta remeber to guide our body to the key points, not force our body to those points.

 

[RowingBoats]

I think that's a good perspective, and probably true in the short term, but after using a few different current AI's for work I'm not ready to underestimate what it will be able to do in the (relatively near) future.

With more and more data like this, I bet it could make specific recommendations BASED on your body type. Imagine a time when it's able to take a look at video of your form and compare it to the 3d data it has. It could build a form for you based on your body and current form. What if it creates an invisialine-style series of small adjustments that quickly transitions your form to the idealized version? I don't think this is far off. It will probably start (or probably already has started) in more mainstream sports.

_

I also think the idea that GG couldn't throw like Eagle (and vice versa) is really interesting. My intuition tells me that if they had the same amount of time to develop the other form as they did their own, they would both still have success. Hard to know if even the top throwers are all throwing the ideal way for their bodies.

Would GG, Eagle, or Tamm throw further if they did what David Wiggins Jr. does?

It will be like invisalign, but with pain lasers. Deviate from the expected physical parameters in the conformity room, and the penalty gradually extends from hair singing to complete amputation of the offending limb.

 

[Brychanus]

It will be like invisalign, but with pain lasers. Deviate from the expected physical parameters in the conformity room, and the penalty gradually extends from hair singing to complete amputation of the offending limb.

For some reason you reminded me of this: