Will you trust Tech for practicing martial art?

[Taekiso]

Hello martial art friends,

I've been a passionate martial artist since my childhood, growing up with the Jacky Chan, Bruce Lee movies,... ;-)
I accumulated over 20 years of practice and some years of teaching as well (mainly Taekwondo).

Today with the pandemic we had limited or no-access to clubs and gyms and that made me think on how to train myself (as a lifetime athlete) and train others as well during that time. And I came up with an idea however now I'm wondering if this idea will just fulfil my own needs or it could help others that feel this need as well, so your thoughts and your feedback will be warmly appreciated.

My concept is to build an All in one measurement and workout app for martial artists, the app will track every movement and corrects the athlete's technique, pace, and performance (speed, power, reaction time,...). It could be for athletes of all levels. I’m imagining it like a new performance lab accessible for everyone. A bit like the “Fight Science” documentary from National Geographic, I was always dreaming of doing a similar performance test for myself and evaluating if my training program was effective or not.

What should I do?

Thanks

 

[Buka]

Welcome to Martial Talk, Taekiso, hope you enjoy it.

I don't really have much advice on this, other than to go for it and have fun doing it.

Probably best to focus on your most experienced area, which would be TKD.

At the very least, it gives you a good opportunity to do something helpful and explore your own basics and the things that follow those basics.

Go gettum, bro.

 

[Taekiso]

Thank you for your feedback

 

[drop bear]

Martial eye or something?

 

[Punchy]

Martial Feedback?

Great idea. Put me down as an investor.

 

[CB Jones]

Where will you get the feedback?

We use Coach's to break down mechanics in baseball. Will it be something like that with a platform to upload the video and allow people to critique and give advice.

Wonder if you could use coach's eye tech and just build the platform for sharing and advice.

 

[Taekiso]

Yes, it's kinda Coach's Eye however you won't need to upload or neither someone will have to draw over your video. Everything could potentially be done with AI/computer vision (through your smartphone or tablet or anything with a camera and a processor that could process the video), the only human involvement I see it's in the process of content creation (workout, performance test protocols,...) and teach the computer vision the right techniques...

If I get enough positive feedback I will dig deep into this.

Thanks guys

 

[CB Jones]

Yes, it's kinda Coach's Eye however you won't need to upload or neither someone will have to draw over your video. Everything could potentially be done with AI/computer vision (through your smartphone or tablet or anything with a camera and a processor that could process the video), the only human involvement I see it's in the process of content creation (workout, performance test protocols,...) and teach the computer vision the right techniques...

If I get enough positive feedback I will dig deep into this.

Thanks guys

To have a computer analyze it is gonna take a ton of information inputed into it. And how will you combine the different /opposing ideas and philosophies from the different styles?

 

[dvcochran]

Yes, it's kinda Coach's Eye however you won't need to upload or neither someone will have to draw over your video. Everything could potentially be done with AI/computer vision (through your smartphone or tablet or anything with a camera and a processor that could process the video), the only human involvement I see it's in the process of content creation (workout, performance test protocols,...) and teach the computer vision the right techniques...

If I get enough positive feedback I will dig deep into this.

Thanks guys

First, welcome to the forum.
It is a cool concept. My company does industrial control & automation. The 'learning' of a vision system (such as A.I.) can be laborious and is very dependent on comparators and fairly exact inputs images. For example, where one sets their phone to capture a movement could easily skew the image being processed as good or bad. This is compounded rather quickly when there are multiple results for a given input. So if you say there are 20 parameters to validate a movement the quagmire gets deep pretty quick; especially when you think of a comprehensive format that is trying to measure a complete martial arts system.
To give you an idea of some of the complexity in measuring systems, a vision system that can locate a part into the .x000th range takes a minimum of 3 cameras located to capture the x,y, & z axis. Measuring depth can be a real bxxxh. Color, ambient lighting and such can really mess with the measurement.
I am not trying to rain on you parade but just trying to play the 'what-if' game a little. I hope you are successful. It would be a really cool app.

 

[Taekiso]

To have a computer analyse it is gonna take a ton of information inputed into it. And how will you combine the different /opposing ideas and philosophies from the different styles?

That's right, it will take time however some of the machine learning could be automatise and done quickly once we wrote the right algorithm and also take benefit from existing AI libraries like 'pose estimation'.

For the different /opposing styles, first and foremost I will start with the basic of all martial art and gradually built on top with all the different styles in collaboration with their respective expert in that discipline.

Could you suggest something else concerning the different disciplines?

 

[Taekiso]

First, welcome to the forum.
It is a cool concept. My company does industrial control & automation. The 'learning' of a vision system (such as A.I.) can be laborious and is very dependent on comparators and fairly exact inputs images. For example, where one sets their phone to capture a movement could easily skew the image being processed as good or bad. This is compounded rather quickly when there are multiple results for a given input. So if you say there are 20 parameters to validate a movement the quagmire gets deep pretty quick; especially when you think of a comprehensive format that is trying to measure a complete martial arts system.
To give you an idea of some of the complexity in measuring systems, a vision system that can locate a part into the .x000th range takes a minimum of 3 cameras located to capture the x,y, & z axis. Measuring depth can be a real bxxxh. Color, ambient lighting and such can really mess with the measurement.
I am not trying to rain on you parade but just trying to play the 'what-if' game a little. I hope you are successful. It would be a really cool app.

Thanks, I really see your point indeed the tech side is really challenging and if we crack it down this were it will interesting and valuable for all the practitioner.

I had also played the what-if game and did some tech analysis to see where the current technology limits are and realised that we could turn and fine tune existing AI libraries in the advantage of martial art.

Here are a few of existing AI libraries and you can see the computer vision is doing quite well:

Still it's not 100% perfect however with dedicated time and effort it could get better and better.

Do you see those libraries as accurate?

Thanks

 

[Deleted member 39746]

Equipment and technollogy has its place and scope, if they meet the scope and are useful then no reason not to. Like VR can be decent for sparing if the VR programme is made to be useful for it.

 

[Deleted member 39746]

apoligies for double post, wanted to elborate my point more and didnt want the edit timer restriction.


Right so how designing works is X person sets criteria for the maker to achive the maker then achives them to best of ability or makes comprises based on importance. The main issue here is the commisioneer or the maker (if one in the same) dont know much about said combat sport or MA to design something that would work for it. It is by far not the makers fault if the concept is flawed as they normally dont come up with it just are hired to make it a reality. The issue with home grown designers is they are normally not exactly the best in engineering to make something that can work.

So long as the concept is good and it is made well, i dont see why you cant. As mentioned above some VR games can be pretty good training tools if made with it in mind and are made in a set way to be useful for it.

 

[dvcochran]

Thanks, I really see your point indeed the tech side is really challenging and if we crack it down this were it will interesting and valuable for all the practitioner.

I had also played the what-if game and did some tech analysis to see where the current technology limits are and realised that we could turn and fine tune existing AI libraries in the advantage of martial art.

Here are a few of existing AI libraries and you can see the computer vision is doing quite well:

Still it's not 100% perfect however with dedicated time and effort it could get better and better.

Do you see those libraries as accurate?

Thanks

I have not done wire framing because the precision is just not there for the measurements we make. You can see the lag in the frame on every movement.
So let's say that is not an issue and is close enough that you have inputs that satisfy the requirements. How do you identify 'this is a correct down block' for example? If you have the 'how' of the input, what is the what, where, when, why computation and real world control?
If I understood the 2nd video, it takes 2D imagery and extrapolates to 3D based on known body shape data (assumption). In the dancing video it seems to do a good job of following the movement (with lag). Between the two dancers there are differences even though they are doing the same movements. How will your app identify each dancers movements as good or bad?
Just more 'what-if'.

 

[Monkey Turned Wolf]

I also have doubts about us having the technical capabilities to create an app like this, but AI is not my field so I'll trust the experts on that. If this was created successfully, and it happened to be used for an art that I train in (as what is correct in one art is incorrect in another, often), then I can see myself using it.

 

[Flying Crane]

That's right, it will take time however some of the machine learning could be automatise and done quickly once we wrote the right algorithm and also take benefit from existing AI libraries like 'pose estimation'.

For the different /opposing styles, first and foremost I will start with the basic of all martial art and gradually built on top with all the different styles in collaboration with their respective expert in that discipline.

Could you suggest something else concerning the different disciplines?

I think technology can play a role in learning and training martial arts but it cannot replace human interaction and instruction.

Your second paragraph here tells me that you are coming at this with some heavily misunderstood presumptions. First, there is no “basic of all martial art” to begin with. Second, there is no “respective expert” for most styles, meaning that (with some exceptions) there is no single authority that says exactly how something is to be done. There are various individuals who typically are seen as an authority within a certain group, which group may be small or large, but are not accepted as an authority outside of that group, even among others who practice the same style or a variant of the same style. In fact, there can be a lot of disagreement and even downright hostility between some of these people.

Martial arts is not just mimicry of movement. There is a lot of subtlety beneath the surface that can make one person’s punch (for example) dramatically different from another person’s, even if they appear to be the same from the viewpoint of an observer. These are issues of how one engages and uses their stance and root and posture to drive that punch which can differs from one system to another, and one person from another. The posture and path of movement can appear to be the same, but how power is being generated and the effectiveness of that punch, can be worlds apart. Further, the methodology of how one system goes about developing that punch and that power can be dramatically different. I can promise you, the fundamental punching practice as done by my system, Tibetan White Crane kung fu, looks nothing like the punching practice done in a wing Chun kung fu school or a Tae Kwon do school or a boxing school or a shotokan Karate school. Eventually, a seasoned practitioner in any of these systems should develop a powerful and devastating punch, and there may ultimately be similarities in the principles that are being engaged, but the training method in how they get there can vary dramatically. There is no algorithm that will be able to take this into account, there is no computer that can understand the variations from one system to another and the complexities of how the system works and then guide the student in self-discovery and learning and self-correction in a way that teaches them any of these martial systems in an accurate and effective way.

I think this is an attempt to do an end-run around the need for a competent and knowledgeable teacher. I do not believe it is possible. I do believe technology can be useful as a supplemental tool. Movement analysis can be worthwhile, but that movement then needs to be analyzed by a teacher who understands how the system works. The computer itself cannot correct the student. The student cannot self- correct without a teacher giving that guidance. The teacher needs to be present and needs to be doing the major bulk of the instruction.

If you are interested in pursuing a project like this, I suggest you look for a meaningful way to develop the technology into an effective supplemental tool, assuming the student is receiving quality instruction from a good teacher. But I think it is a mistake to believe it can become a replacement for the teacher.

 

[KenpoMaster805]

Welcome to MT Bro that sound awesome do it see how it goes.

 

[Tony Dismukes]

I'm pretty certain that the technology you are imagining does not exist and will not be possible for quite some time to come.

We are at the point where A.I. can create animations of human movements (including martial arts moves) based on wire-frame simulations and video of human models. (Although if you want a high degree of accuracy for those animations, the human models generally wear special suits with tracking dots.)

What we don't have is A.I. which can understand all the underlying body mechanics, the purpose of a movement, the adjustments necessary for an individual human body which differs from the original movement model given to the program, and how to communicate the relevant corrections in a way that makes sense to the practitioner.

There's no theoretical reason why we can't eventually develop an A.I. which can do these things, but I don't believe we're close to being able to do so.

 

[Taekiso]

apoligies for double post, wanted to elborate my point more and didnt want the edit timer restriction.


Right so how designing works is X person sets criteria for the maker to achive the maker then achives them to best of ability or makes comprises based on importance. The main issue here is the commisioneer or the maker (if one in the same) dont know much about said combat sport or MA to design something that would work for it. It is by far not the makers fault if the concept is flawed as they normally dont come up with it just are hired to make it a reality. The issue with home grown designers is they are normally not exactly the best in engineering to make something that can work.

So long as the concept is good and it is made well, i dont see why you cant. As mentioned above some VR games can be pretty good training tools if made with it in mind and are made in a set way to be useful for it.

Thank you for your feedback.

That's right VR has some interesting potential in MA and can be quite immersive as well.

 

[Taekiso]

I have not done wire framing because the precision is just not there for the measurements we make. You can see the lag in the frame on every movement.
So let's say that is not an issue and is close enough that you have inputs that satisfy the requirements. How do you identify 'this is a correct down block' for example? If you have the 'how' of the input, what is the what, where, when, why computation and real world control?
If I understood the 2nd video, it takes 2D imagery and extrapolates to 3D based on known body shape data (assumption). In the dancing video it seems to do a good job of following the movement (with lag). Between the two dancers there are differences even though they are doing the same movements. How will your app identify each dancers movements as good or bad?
Just more 'what-if'.

That's right this is still the early stage of the tech and it needs a lot of trials and error before reaching something usable and satisfying. I don't have all the answers now however I have the vision and the passion and hopefully this could be a good base to turn this into a useful reality.

The AI will compare it to the existing(on the system) movements defined as bad or good. What could you suggest?