The next Olympic champion - First AI Robot to beat a Professional Table Tennis Player

[Takkyu_wa_inochi]

VISION IA (@vision_ia) on X

Un robot vient de battre des joueurs de tennis de table professionnels. Publié aujourd'hui en couverture de Nature. Ace, développé par Sony AI, est le premier système autonome de l'histoire à atteindre un niveau expert dans un sport physique compétitif. Les résultats : → 3 https://t.co/ymdcdOHMLN

x.com

 

[NetProphet]

VISION IA (@vision_ia) on X

Un robot vient de battre des joueurs de tennis de table professionnels. Publié aujourd'hui en couverture de Nature. Ace, développé par Sony AI, est le premier système autonome de l'histoire à atteindre un niveau expert dans un sport physique compétitif. Les résultats : → 3 https://t.co/ymdcdOHMLN

x.com

Yeah but what equipment does it use? 😉

 

[latej]

Regular ITTF equipment.

Impressive feat indeed...

But I was thinking, part of the game is watching the opponent, and predicting what is he/she doing. It's hard or impossible if it is not human form, but that is coming too, I guess, no excuses ;-)

 

[Takkyu_wa_inochi]

Yeah but what equipment does it use?

It’s written in the Nature article
À Victas blade and Dignics05 max

Outplaying elite table tennis players with an autonomous robot - Nature

An autonomous robot system, Ace, combines event-based vision and reinforcement learning to compete with elite human table tennis players, highlighting the potential of physical AI agents to perform complex, real-time interactive tasks.

www.nature.com

 

[Takkyu_wa_inochi]

Ace Research Project | Sony AI

Sony AI's robot masters pro table tennis using superhuman perception and real-time control, bringing reinforcement learning from virtual to physical sport.

ace.ai.sony

 

[NextLevel]

Got some related links from friends and family so I found the source video.

AI-powered robot beats elite table tennis players

In feat hailed as milestone in robotics, Sony AI’s Ace wins three out of five matches played under official rules

www.theguardian.com

 

[TampaBayTableTennis]

it's really incredible to see

 

[NextLevel]

it's really incredible to see

What is also interesting is that it isn't even fully adopting our technique at all if it ever will. It is trapping the ball at a higher height than humans would close to the table rather than backing up to play. So there is potentially still room for improvement with a different form of reinforced development.

 

[OvrChkn]

The AI will eventually realize that it needs to go down the EJ process!

 

[golden_son]

The AI will eventually realize that it needs to go down the EJ process!

If they feed it enough TTD data, yes

"My racket is insufficient. I require Golden Viscaria and Zyre 03."
...
"Golden Viscaria and Zyre 03 is insufficient. I require a 'player version' blade that is not commercially produced and booster for my Zyre 03."
...
"My Robot Edition Golden Viscaria and boosted Zyre 03 is insufficient. Here's a formula for a new rubber composition and engineering documents to craft a new blade for it."

 

[pong99]

It would be interesting to see how the learning model works if you put LP on one side.

How does an AI figure out which side to use and how long would it take?

 

[NextLevel]

It would be interesting to see how the learning model works if you put LP on one side.

How does an AI figure out which side to use and how long would it take?

That is an amazing idea.

 

[zeio]

Regular ITTF equipment.

Impressive feat indeed...

But I was thinking, part of the game is watching the opponent, and predicting what is he/she doing. It's hard or impossible if it is not human form, but that is coming too, I guess, no excuses ;-)

But one thing it does in common with human (even more extreme) is wrapping the ball (1, 2)!

https://www.reddit.com/r/baduk/comments/1rsggtz/comment/oa8aoid/

• Fan Hui, on the psychological shock to Go professionals
  
  That is why he keeps returning to mentality rather than pure technique. He even says that if a human such as Aja Huang had been physically sitting across from him in place of the machine, he would have felt better. A person still grants you the feeling of being seen. A machine does not care how long you think, whether you are humiliated, or whether the game becomes a permanent shadow in your life. It simply continues. “What constrains us most may be our knowledge itself. I call it the prison of knowledge.”

https://en.wikipedia.org/wiki/AlphaGo_versus_Fan_Hui

 

[vossi39]

I think the biggest discrepancy of this robot vs human shootout, is the vision part. Robot has 9 separate cameras only to track the ball in 3d space and additional cameras to track spin + additional cameras to watch the opponent and his movement. The human has to do all this with his 2 eyes.

Still an incredible engineering exercise, both IT and mechanics.

 

[Takkyu_wa_inochi]

Now I want this robot for Christmas

 

[zeio]

Kihara got the hang of it after G1 and could've exploited the "loophole" more.

Hirano gave up after G1 as usual.

March 2026 Matches (video 1 of 3) | Sony AI
https://www.youtube.com/watch?v=xmUkuVaWyD4
March 2026 Matches (video 2 of 3) | Sony AI
https://www.youtube.com/watch?v=fIAom-4CJqI

March 2026 Matches (video 3 of 3) | Sony AI
https://www.youtube.com/watch?v=7G3c5Z8LCl0

Ace vs. Hirano | Pro Match Highlights | Sony AI Table Tennis Robot
https://www.youtube.com/watch?v=8AGAnb4-HbM

 

[NextLevel]

Kihara got the hang of it after G1 and could've exploited the "loophole" more.

Hirano gave up after G1 as usual.

March 2026 Matches (video 1 of 3) | Sony AI
https://www.youtube.com/watch?v=xmUkuVaWyD4
March 2026 Matches (video 2 of 3) | Sony AI
https://www.youtube.com/watch?v=fIAom-4CJqI

March 2026 Matches (video 3 of 3) | Sony AI
https://www.youtube.com/watch?v=7G3c5Z8LCl0

Ace vs. Hirano | Pro Match Highlights | Sony AI Table Tennis Robot
https://www.youtube.com/watch?v=8AGAnb4-HbM

They both lost even though Kihara took a game. I wonder how Mizutani would do....

 

[zeio]

Ono, who had the lowest WR regardless of gender, turned out to be the only one to win all 3 bo3 matches (though the score in the match highlights doesn't match the ones in the table). Other than Okano, Ace had a much lower success rate against the other men.

What Happened After the Nature Paper: Ace vs. Professional Players
https://ai.sony/blog/what-happened-after-the-nature-paper-ace-vs.-professional-players

Ace vs. Ono | Pro Match Highlights | Sony AI Table Tennis Robot
https://www.youtube.com/watch?v=SXp3PzGrVRE