AI and 3D technology to assist refs?

Discussion in 'Fencing Discussion' started by lovefoil, Dec 11, 2018.

  1. lovefoil

    lovefoil Made the Cut

    Joined:
    Jul 23, 2010
    Messages:
    340
    Likes Received:
    17
    I saw that gymnastics will be using AI and 3D technology sensors to analyze movements and assist the judges; it will be tested at next year's World Championships. I was wondering if anything like this was considered for fencing and could help with the right-away weapons... seems like they have a massive database with elements that the athletes movements are compared to. The replay can be seen from multiple views and angles, so the judges can make better decisions. As the techniques and speed improves, they believe the naked eye is not sufficient to make those judgments alone and this technology would help.

    There would be a challenge to come up with a massive database of right-away actions that the majority agrees to, I guess. But I wonder how receptive the community would be to something like this...
     
  2. jjefferies

    jjefferies Podium

    Joined:
    Dec 16, 2005
    Messages:
    3,929
    Likes Received:
    198
    It would obviously add in a new element to the complexities of the game. Whether that element is a trained grace or just random luck could only be know after extensive use. Personally I would opine in favor of dumb luck. At which time the ROW practitioners might finally realize the futility of ROW and come over to the dark side and become epeeists. ;)
    Take that Inq!!
     
  3. bbower

    bbower DE Bracket

    Joined:
    Feb 20, 2003
    Messages:
    535
    Likes Received:
    153
    Given the number of high ranking referees who don't think right of way can be described using the written word, I'm pretty sure this would be nearly impossible.

    As you pointed out, first come up with a massive database of actions that the majority agree to. That information has the benefit of aiding referee education as well, and would be a huge improvement over the completely asinine way referees are taught now (at least in the US). Once you've addressed that problem, you can maybe start thinking about using software as an aid to refereeing.
     
    InFerrumVeritas, jdude97 and lovefoil like this.
  4. lovefoil

    lovefoil Made the Cut

    Joined:
    Jul 23, 2010
    Messages:
    340
    Likes Received:
    17
    hmmm.. epee is quite slow... booooring ;) but, maybe we can reduce the target to foil target instead of eliminating the white lights in foil and enforce passivity rules and give double red and black cards... :)
     
  5. catwood1

    catwood1 is a Verified Fencing Expertcatwood1 Podium

    Joined:
    Apr 10, 2007
    Messages:
    2,436
    Likes Received:
    166
    It would cost approximately a zillion dollars to build this. Maybe 2 zillion. After doing that, MAYBE it would work.
     
    lovefoil likes this.
  6. Mihail

    Mihail Podium

    Joined:
    Apr 28, 2004
    Messages:
    657
    Likes Received:
    104
    With all due respect to high ranking referees, I think there's a disconnect between the ability to intuitively understand fencing actions and the ability to translate intuitions to algorithms and data sets. Just like referees/fencers get good at understanding fencing actions by watching and performing them, assigning data values to human actions takes practice and experience with technology.

    Speaking of sabre only (I'm not experienced with foil) I know we love to have an old school approach to conventions: "I can't tell you exactly what an attack is but I know it when I see it." Well, I think when the right software engineer/athlete looks at the actions, they'll discover many different but repeated patterns. A swing-through here, an acceleration of the body/feet/arm there, change in arm angle, the relative speed of both fencers, landing of the foot -- these are all things that are easily detectable with fairly cheap and available tech (e.g. https://brekel.com/brekel-pro-body-v2/)

    This is stuff that is not being currently experimented with, but absolutely should be.
     
  7. Inquartata

    Inquartata Podium

    Joined:
    Jul 12, 2001
    Messages:
    37,114
    Likes Received:
    1,414
    Maybe someone could design an AI to do that. Then it could teach the reffing AI.

    Then it would decide our fate in a microsecond: extermination. Or worse, force us all to fence epee!
     
  8. Alex_Paul

    Alex_Paul Podium

    Joined:
    Jul 26, 2004
    Messages:
    1,099
    Likes Received:
    142
    You have two possible approaches:

    You change the game by setting a new set of codes as to what constitutes ROW these could be closely linked to what we think of as ROW now or not. Essentially if fencer A moves their foot in direction B at speed C they have row until XYZ. This could be done and could be fairly simple but would change the game. Fencers would hate it initially but they would get used to it and learn how to fence the new system.

    OR

    You feed it real life data and use machine learning like companies are doing with self driving. You could just log a ton of fencing and the way the referee decided (ruling out any ahem... contentious decisions) and use that data to allow the software to write the rules to what it sees.
     
  9. Gav

    Gav is a Verified Fencing ExpertGav Moderator!!

    Joined:
    Aug 19, 2000
    Messages:
    3,522
    Likes Received:
    554
    "AI" the new "3DTV"
     
  10. jkormann

    jkormann Podium

    Joined:
    Nov 8, 2002
    Messages:
    3,229
    Likes Received:
    248
    And it would be fascinating to see it's interpretation of the rules.
     
    lovefoil likes this.
  11. Mac A. Bee

    Mac A. Bee is a Verified Fencing ExpertMac A. Bee Podium

    Joined:
    Feb 5, 2008
    Messages:
    1,959
    Likes Received:
    211
    USA Fencing has been "working" on just a *video* ref compilation for 10 years.:rolleyes:
     
    Inquartata and InFerrumVeritas like this.
  12. Inquartata

    Inquartata Podium

    Joined:
    Jul 12, 2001
    Messages:
    37,114
    Likes Received:
    1,414
    In both cases, how would it handle atypical fencing? For instance, in the first case there are vet fencers or youth fencers who might not be able physically to "move their foot in direction B at speed C" but perhaps only C-x or whatever. What about fencers with weird styles and odd idiosyncratic "moves"? Would we use an AI taught this way only for high-level senior fencing? How to define the parameters of "high-level"? Would all senior fencers have to force themselves into the mould of what the AI expects? Will it be able to handle the differences between, say, Szilagy and Gu? Or will it just force everyone to fence alike in the end?

    In the second case, is this "ton of fencing" and referee rulings going to extend to vet and youth fencing ( not to mention a wide, ah spectrum of refereeing ability ) as well, and if so how will the machine differentiate between them?

    What happens when you have a vet fencer competing against seniors, or a 13-y.o. against a 40-y.o.? Does the AI explode trying to use multiple algorithms? :D

    All in all I think it's a bad idea. It might eventually yield greater consistency in refereeing, but it would also drive fencing to increased homogeneity.
     
  13. lovefoil

    lovefoil Made the Cut

    Joined:
    Jul 23, 2010
    Messages:
    340
    Likes Received:
    17
    Well, the AI would use data points. See demo below... fencing is more complex with 2 fencers who have to be analyzed simultaneously... the rules are the same for the weird fencers, small fencers, young fencers, vet fencers... I think it is very neat..

    http://www.asahi.com/ajw/articles/AJ201812030005.html
     
  14. Inquartata

    Inquartata Podium

    Joined:
    Jul 12, 2001
    Messages:
    37,114
    Likes Received:
    1,414
    The rules are the same, but the abilities and characteristics of the fencers are not.

    And then there are the rules which aren't rules, eg the "conventions". The attaque compose in sabre for instance is not in the rules. Would the AI judge by the convention, or the rule? Or, the rules define the attack without reference to speed. Given two fencers who start simultaneously and do not prepare but one of whom is faster or "more committed" than the other, does the AI go by the convention or the rule?
     
  15. jkormann

    jkormann Podium

    Joined:
    Nov 8, 2002
    Messages:
    3,229
    Likes Received:
    248
    I want to see all the fencers dressed like crash-test dummies, in some high-viz material with the black/white circles to allow the camera to capture angles and speed.
     
    Grey Sabreur likes this.
  16. Alex_Paul

    Alex_Paul Podium

    Joined:
    Jul 26, 2004
    Messages:
    1,099
    Likes Received:
    142
    In the first case yes, fencers would have to conform to the new rules as is always the case with new rules. You would have to get those rules right and fencers would hate it but it is certainly an option.

    In terms of using machine learning the same applies, the more data the better and then you apply a best fit. You obviously wouldn't want to feed it with data from bad referees!

    In say 20 years time with radio connected sensors on the hand, foot and the body there is no reason at all that a computer shouldn't be able to referee better and more consistently than a human without messing up the sport.

    There isn't any data that a referee should be using that isn't based on the 3d positioning of various parts of the two fencers. At the point where you have that data in real time I believe that developing a rule base will be challenging but entirely possible.
     
    JS Hart and lovefoil like this.
  17. Gav

    Gav is a Verified Fencing ExpertGav Moderator!!

    Joined:
    Aug 19, 2000
    Messages:
    3,522
    Likes Received:
    554
    I think the question is whether we actually want a robot to referee.

    You all might be surprised to hear this, considering my background, but I'm actually against it.
     
  18. Allen Evans

    Allen Evans Podium

    Joined:
    Oct 25, 2003
    Messages:
    4,899
    Likes Received:
    970
    One of the reasons fencing in foil and saber is where it is today is because of the tension between the athletes/coaches and the officiating and how that relationship has changed over time Would automating refereeing end that tension and result in stagnation? Perhaps (what can be programmed can also be un-programmed) but I think that the process would be less dynamic, less forgiving (see the new rules about epee non-combativity as an example) and....more boring.
     
  19. Quinn

    Quinn Made the Cut

    Joined:
    Feb 28, 2013
    Messages:
    67
    Likes Received:
    15
    On the other hand, it's the closest we could concievably come to consistent and (arguably) fair refereeing. You couldn't stumble upon a referee who happens to disagree with your idea of priority, forcing you to fence in a style you're less comfortable in.
     
  20. Inquartata

    Inquartata Podium

    Joined:
    Jul 12, 2001
    Messages:
    37,114
    Likes Received:
    1,414
    Until there were "technical difficulties", power surges, or Russian hacking...

    Or until the AI decided it was Italian.
     

Share This Page