A year ago this [0] table tennis robot backed by Google DeepMind was discussed on HN.
It plays much worse and the HN discussion is anchored around whether it's OK to call it "human-level" or if the authors should have clarified that they meant a human who doesn't actually play table tennis. But it was accepted as being SOTA at that time.
What happened since then? This looks like the kind of level of advance we see in, say, coding AIs, but I thought physical robotics was advancing much more slowly.
A partial answer is that the new robot cheats in ways that DeepMind didn't seem to. It has high speed cameras all over the room and can detect spin by observing the logo on the ball. But I'm not sure this explains such a big advance.
As a human player (of a not-high standard) I cannot see the spin of the ball directly. I can only infer it from the movement of my opponents bat. So I would wonder that a camera could pick it up in real time.
As a player myself, and having seen much higher level player than me, reading the spin from the ball rotation (and in fact trajectory) of the ball is a common (if advanced) skill. Sometimes the movement of the bat can be deceptive (since with the same movement, where it contact on the bat, the finger pressure can affect the spin).
For example, backspin/underspin balls will move slower after the first bounce and feel 'damper' while topspin will jump. So it's def. possible (and in fact reliable) to read the spin from the spin and trajectory of the ball.
Visually reading spin is unreliable at all levels; the ITTF passed the two-color rubber rule requiring one black and one red side to neutralize players taking advantage of their opponents being unable to read the spin from watching the ball rotation via twiddling rackets with the same color rubber on both sides, but different characteristics.
Ping pong paddles have two sides, with different characteristics for each side. Now the two sides have to have different colors so your opponent can see what you are hitting it with, where before you could use the same color on both sides and your opponent wouldn't be able to tell how the ball would react
I had a look at Google trends for France. Table tennis is slightly more common than ping pong but the latter is much more stable. Table tennis has huge peaks, the biggest one being during the OG in Paris. These parks are not reflected in there ping pong trend
Interestingly, for Youtube searches this is the other way, with a much bigger difference in favour to ping pong
Peter Ilyich Tchaikovsky was born in Votkinsk, May 7 1840.
When he was a little boy he never played out in the streets of Votkinsk like the other little children of Votkinsk, because when Tchaikovsky was one month old, his parents moved to St. Petersburg.
My biggest fear at the moment is robot armies and police forces.
Case in point : we're all expecting China needs to invade Taiwan soon, or they will run out of soldiers because of the one child policies of the 70s/80s.
Meanwhile, Ukraine is holding up against a "modern" army with quickly assembled drones.
So it all seems a bit like "they'll never put tanks through the Ardennes", sort of ?
Where and when will the first invasion of a country by a purely remote controlled, AI assisted army take place ?
Will robot battalions embed civilians to act as human shields ? Will the AI learn to mistreat the locals to maintain fear, or will they see it as a needless distraction and rush to the center of powers ?
If war is mostly played out from a disrance, will years of playing RTS give South Korea an edge ?
To some extent it already has, Ukraine had a press release a few days ago stating they had attacked and taken a position using only robots and drones for the first time
You say that now, but once we perfect AMBAC technology and accidentally release large numbers of Minovsky particles, we will need humanoid combat vehicles to fight our battles!
Autonomous suicide drone swarms are easily countered by autonomous interceptor swarms.
>Marching humanoid terminator robots
ground bots, not necessarily marching, do have their value. They can have bulletproof armor, while still be relatively lightweight and small and fast. They can easily carry even 20-25mm autocannon - very destructive weapon, sometimes can even succeed against a real tank.
And imagine when a swarm of drones lifts a ground bot, brings and drops it right into the needed point and protects it from the enemy drones while the ground bot just destructs the things around. Synergy between different weapons system has always been the super-weapon.
I don't think Russian army is very modern -- but maybe that's the reason of your quotation marks.
I kinda think that the competitions among the big dogs (US/Russia/China/etc.) would eventually green light ANY AI/Robots projects if they can justify tipping the scale somehow, and in the process completely destroys the last element of any political counter-weight. Because "fear gives men wings".
I would really hate to live in a dystopian world worse than what is described in the books/movies.
Don't table tennis players learn to predict how the ball will act based on their opponents movements? Seems like if they aren't able to do that with a robot opponent (who doesn't look or behave like a human) then they wouldn't be able to play at their best.
I do expect this to have a "novelty edge" over human opponents - which can be closed with practice, on the human end.
And, like many AIs, it can have "jagged capability" gaps, with inhuman failure modes living in them - which humans can learn to exploit, but the robot wouldn't adapt to their exploitation because it doesn't learn continuously. Happened with various types of ML AIs designed to fight humans.
Only if you assume the AI can't improve. Otherwise, AI has a fundamental edge over humans in that they don't get old and die, and can be copied perfectly without an expensive retraining period
Chess players learned to exploit chess computers’ weaknesses in the beginning too, but they can’t any longer. This version of the robot might not learn continuously, but the next will be better.
You can predict the movement of the ball (speed, direction, spin) based on the movement of the bat relative to the ball. What the rest of the player's body is doing is irrelevant to predicting what the ball will do - but relevant to predicting where they will be when you make the return shot.
Rui Takenaka, an elite-level player who has won and lost matches against Ace, said in comments provided by Sony AI: "When it came to my serve, if I used a serve with complex spin, Ace also returned the ball with complex spin, which made it difficult for me. But when I used a simple serve - what we call a knuckle serve - Ace returned a simpler ball. That made it easier for me to attack on the third shot, and I think that was the key reason why I was able to win."
It seems like the human players might be playing in a way that tacitly overestimates their AI opponents' intelligence and underestimates their skill. AFAIK the SOTA Go AIs are still vulnerable to certain very stupid adversarial strategies that wouldn't fool an amateur (albeit they're not something you'd come up with in normal play, more like a weird cheat code). I wonder if this could get ironed out with a bit more training against humans vs. simulation.
I would love to see a video of this thing that shows the whole table. From the paper I guess they have to light the area very brightly. But it seems like a pretty serious set up.
I'm not that excited about 'x beats human at y' anymore. I am more interested in 'x beats human at made up on the spot tasks p d and q'. That is starting to happen more generically and is a bigger sign of emerging capability. We can always create something confined that will beat humans, it isn't until recently that we are starting to be able to generally beat humans at tasks.
Much like the robots beating half marathon records in China recently… who cares? Cake making robots can make cakes way faster than human bakers. Cars and motorcycles go faster than bicyclists. It is a boring given that purpose made machines perform the tasks they are built to perform better than humans.
Yeah, thinking through it a bit further, the real story here, aside from the mechanical engineering, is the application of AI/machine learning/computer vision processing. The advancements that have made it possible to reason about, simulate, and react to the complexities of a spinning ball in a fraction of a second are pretty cool. My gripe is mostly that this article isn't focusing on and detailing this.
The article's main focus is on the "vs. human" aspect and is light on technical details. I would love to hear specifics from the engineers behind this.
This is great, I remember being sorely disappointed by the hyped up Timo Boll vs Kuka robot 12 years ago. I thought it was going to be a real match and seemed like the robot would destroy him, but ended up just being a marketing stunt and felt like a fixed fight, with no real digging into the tech or why the robot "lost". Still some cool footage: https://youtu.be/tIIJME8-au8
Am I correct in my understanding that- they had specialized software that not only tracked the ball, calculated spins using the logo, and fed calculated trajectories?
Even club level players have access to tennis table 'robots'. They fire the ball at you and collect the return in a net. You can set the speed, position and spin. They are very basic compared to this robot, but useful for training.
I've always wished for something similar: autonomous car racing. No human drivers. No remote controls. Just program the cars before the race, and let them go. Maybe even load the cars with mild explosives so they go BOOM when they crash.
It plays much worse and the HN discussion is anchored around whether it's OK to call it "human-level" or if the authors should have clarified that they meant a human who doesn't actually play table tennis. But it was accepted as being SOTA at that time.
What happened since then? This looks like the kind of level of advance we see in, say, coding AIs, but I thought physical robotics was advancing much more slowly.
A partial answer is that the new robot cheats in ways that DeepMind didn't seem to. It has high speed cameras all over the room and can detect spin by observing the logo on the ball. But I'm not sure this explains such a big advance.
[0] https://news.ycombinator.com/item?id=43861207
Also IT'S TABLE TENNIS, NOT PING PONG!
Alas HN has finally found its next religious war!
I’ve been feeling a little bored after that whole tabs vs spaces one was settled.
For example, backspin/underspin balls will move slower after the first bounce and feel 'damper' while topspin will jump. So it's def. possible (and in fact reliable) to read the spin from the spin and trajectory of the ball.
https://www.youtube.com/watch?v=EH8kZDc7OLk
Is it also MOVING STAIRCASE, NOT ESCALATOR?
Interestingly, for Youtube searches this is the other way, with a much bigger difference in favour to ping pong
When he was a little boy he never played out in the streets of Votkinsk like the other little children of Votkinsk, because when Tchaikovsky was one month old, his parents moved to St. Petersburg.
— Victor Borge
where it is easy to see
the cryptic admonishment
T.T.T
When you feel how depressinglyslowly you climb
it's well to remember that
Things Take Time
-- Piet Hein
Case in point : we're all expecting China needs to invade Taiwan soon, or they will run out of soldiers because of the one child policies of the 70s/80s.
Meanwhile, Ukraine is holding up against a "modern" army with quickly assembled drones.
So it all seems a bit like "they'll never put tanks through the Ardennes", sort of ?
Where and when will the first invasion of a country by a purely remote controlled, AI assisted army take place ?
Will robot battalions embed civilians to act as human shields ? Will the AI learn to mistreat the locals to maintain fear, or will they see it as a needless distraction and rush to the center of powers ?
If war is mostly played out from a disrance, will years of playing RTS give South Korea an edge ?
https://www.businessinsider.com/ukraine-russia-position-take...
"Citizen, congratulations on reaching your age of majority. Report for your Patriotic Assurance Implant at surgical bay 43B."
>Marching humanoid terminator robots
ground bots, not necessarily marching, do have their value. They can have bulletproof armor, while still be relatively lightweight and small and fast. They can easily carry even 20-25mm autocannon - very destructive weapon, sometimes can even succeed against a real tank.
And imagine when a swarm of drones lifts a ground bot, brings and drops it right into the needed point and protects it from the enemy drones while the ground bot just destructs the things around. Synergy between different weapons system has always been the super-weapon.
I kinda think that the competitions among the big dogs (US/Russia/China/etc.) would eventually green light ANY AI/Robots projects if they can justify tipping the scale somehow, and in the process completely destroys the last element of any political counter-weight. Because "fear gives men wings".
I would really hate to live in a dystopian world worse than what is described in the books/movies.
And, like many AIs, it can have "jagged capability" gaps, with inhuman failure modes living in them - which humans can learn to exploit, but the robot wouldn't adapt to their exploitation because it doesn't learn continuously. Happened with various types of ML AIs designed to fight humans.
https://www.nature.com/articles/s41586-026-10338-5
I would love to see a video of this thing that shows the whole table. From the paper I guess they have to light the area very brightly. But it seems like a pretty serious set up.
Reminds me of this old The Onion story: https://theonion.com/ping-pong-somehow-elicits-macho-posturi...
I was expecting/hoping for a humanoid robot.
How easy is it to introduce artifacts that reduce accuracy and performance?
https://sonyresearch.github.io/ace_public/
> Exactly! He was a machine designed to hit blerns. I mean come on, Wireless Joe was nothing but programmable bat on wheels.
> Oh? And I suppose Pitch-o-mat 5000 was just a modifier howitzer?
> Yep!
Now build a robot that can catch a bullet.