by Laurent Acharian, Sylvain Duranton & Olivier Kahn.
With 2 Olympic titles and 10 world titles under his belt at just 29 years old, the judo champion Teddy Riner is already one of the top French athletes of all time in French sports. Not only have exceptional physical characteristics driven his success, but also rigorous methodology and unmatched analytic know-how in both competition and in training. In an perplexing article published in Usbek & Rica website , Laurent Acharian, Sylvain Duranton and Olivier Kahn, who work in the consulting firm BCG Gamma and who passionately follow judo, explain why Teddy Riner could be likened to a robot programmed to win.
Teddy Riner isn’t really all human. Sure he has his prizes, the greatest story in judo, with 2 Olympic titles, 10 world titles and 5 European crowns. There’s also his 144 consecutive wins… and still counting.
Seven years without losing a single fight in one of the most demanding sports
He has gone more than seven years without losing a single fight in one of the most demanding athletic disciplines where it only takes one looking in the wrong direction, one misstep, to get knocked flat on your back. Slowly but surely, Teddy is inching closer to Yasuhiro Yamashita, the greatest Japanese champion in the heavyweight category with 203 consecutive victories, though he has won just 1 Olympic title and 4 world championships.
There is also his physique. Standing at 2.04 meters and weighing 138 kg (6.5 feet and 304 lbs) when he’s in fighting shape, Riner moves his Titan body at the speed of a middleweight. He can deadlift 170 kg (375 lbs)… with just one hand. That’s the weight of a good quality wood stove!
Teddy Riner is a bulldozer on the mat
Teddy Riner is a real bulldozer on the mat who, from the moment he puts his hands to the ground, squeezes, crushes, picks up and pulverizes all his opponents. The unluckiest of them is the Brazilian Rafael Silva. Nine matches in the biggest competitions… and nine losses, including the Olympic Games, where he was humiliated in front of a rude and rowdy crowd. What a nightmare!
“Teddy is an alien,” to quote Armelle O’Brien, his former physiotherapist on the French team who for a long time needed an assistant to help handle Riner.
So that’s what he is, an alien… But this man whose parents named him Teddy in honor of their first teddy bear, is he really a robot? According to the artificial intelligence researcher Laurence Devilliers, who just wrote a book on the subject, “a robot is a machine composed of at least three elements: sensors to understand its environment, processors that analyze the gathered information in order to make a decision, and mechanisms that allow it to act in the real world.”
Every second the judo fighter must choose the action that will take him to the next moment and maximize the longterm probability that he will reach the moment of victory.
Everything is a question of algorithm for the first two dimensions, and to a certain extent, all the great judo fighters could be considered robots. For the training algorithm (the way the judo fighters learn to fight, memorize movements and scenarios, learn how to adapt to all types of variables), they use what is best understood as machine learning, or reinforcement learning.
A judo match can be modeled as a series of states (the initial state is the salute at the beginning of the fight, the final state is victory or defeat), the transition from one state to another is instigated by a fighter’s action, a movement. Every second the judo fighter must choose the action that will take him to the next moment and maximize the longterm probability that he will reach the moment of victory.
There is, however, a considerable size difference between Robot Teddy and other judo fighter robots: to learn how actions allow to move from one state to another, and which will take him to victory, Teddy has stored a lot more data filed under “victory” than all his opponents. Data on external factors (for example, his opponents have very different builds) but also internal (for every series of movements that he practices during training, he notes his success rate and the proportion of times he reaches the state of “victory.”
Victory is contagious, and Teddy is, in many ways, the AlphaGo of his sport, and certainly less the Google DeepMind.
To make the best decision during a match, Teddy chooses a model that’s very familiar for data scientists: the prescriptive model. On top of that, the quality of his “data lake” gives him a certain advantage over his opponents. His enormous calculating power enables him to chain together Bayesian optimization models. Every one of his actions generates a new state, and another and so on. This action sequence is optimized to maximize the probability of winning. It’s an art!
If AlphaGo became the champion of the game Go, successively defeating Fan Hui, Lee Sedol and Ke Jie, it’s because it learned from each victory and that it could not win without learning from victory. In short, victory is contagious, and Teddy, in many ways, is the AlphaGo of his sport, and certainly less the Google DeepMind.
An invincible program?
Concerning the quality of automation — the famous “mechanisms” of Laurence Devilliers — we have to imagine a startup that invents a judo robot. Surely he would look a lot like Teddy. There are two unknown factors for Teddy’s definitive switch to the world of robots: Has he successfully gotten rid of all forms of emotion? And can his program recognize bugs? In other words, will he know defeat again one day?
We have to remember images of Teddy crying like a little boy on the winners podium. Sore loser. And human, terribly human. Never again seen since.
On the first point, we should remember the champion’s last defeat. September 13, 2010 in Tokyo, Teddy Riner was judged defeated by the Japanese fighter Daiki Kamikawa. The arbitration panel made him feel two things: defeat, of course, and injustice. Evidently, this data was not part of his programming. Remember footage of Teddy crying thief, questions about next steps in his career, and crying like a little boy on the winners podium. What a sore loser. And human, terribly human. Never again seen since.
So, is the program now invincible? Will it take him to the top to win a fourth gold medal at the 2024 Olympic Games in Paris (2020 already seems certain). We would very much like to answer yes. His competition still doesn’t seem to have found the magic key to take down Teddy the robot. During a world semi-final in September 2017, the Georgian fighter Tushivilli believed he could win, but a few months later at the world final, he experienced “the ippon (knockout) and a half” in a 180-degree turn from his over-confident claim. Humiliating.
Riner’s real flaw is wearing bolts
Honestly, to this day, no one seems to be able to challenge the Colossus, including Japanese judo fighters who still dominate the sport and are preparing to carry out an unprecedented raid in Tokyo in 2020. On their home turf, they will certainly take the premier category once again.
We when we see Riner jump into the pool with more than 20 kilos to lose since his Olympic victory, we could say it will be harder and harder for him to blast the weight and to find his punch and mobility again. And the groin and shoulder injuries that await… Riner the Machine is not infallible.
How exciting it would be to see him reach the highest point of his career in Paris, at 35 years old, and collapse in tears making his big comeback. We’ll be watching…
Laurent Acharian, Sylvain Duranton & Olivier Kahn
Source: Deep Learning on Medium