Game, set, match … Robot!.

Adam Spencer
19 hours ago
4 min read

Sony’s table tennis robot Ace took sets off elite humans at one of the fastest sports on Earth. BTW it sees the ball eleven times faster than you do.

https://www.youtube.com/watch?v=vGuumAtXmsw

(First published on my Substack where you can get #NerdNews, marvellous maths and general geekery.)

Last week, a humanoid robot smashed the human half-marathon world record.

NerdNews said at the time “that’s cool, but dexterity and spontaneous reactions are even more important”.

Robots: “Hold my beer”.

Ace in the hole.

This month, Sony AI landed the cover of Nature with a paper that should make every elite table tennis player in the world spit their Gatorade. Their robot, Ace, has been quietly taking sets off the pros.

Back in April 2025, in a custom-built arena at Sony's Tokyo headquarters, Ace went head-to-head with five elite university players and two Japanese pros.

We’re talking full International Table Tennis Federation rules, with licensed umpires and a room full of serious geeks watching on.

The numbers are extraordinary.

Against the elite players (the ‘training 20 hours a week for over a decade’ type), Ace played 13 games and won 7. It took three of five matches outright. Against the two pros, Ace lost both matches but scrapped its way to one game win.

Then, while the paper was in peer review, Sony kept tinkering.

By December 2025, an upgraded Ace returned for a second tournament and beat three of four high-level opponents.

That sound you hear is forty years of robotics PhD students giving it a serious Lleyton Hewitt “Come On”.

Eyes on the prize.

Let’s get our optics-geek on here. Ace doesn't just have eyes. It has twelve sensors watching the table.

Nine are conventional high-speed cameras locating the ball in 3D space at 200 times a second with 3 millimetre accuracy. The other three are "event-based vision sensors". These chips don't capture frames, but fire when individual pixels change brightness, with sub-millisecond precision.

This means Ace can read the logo on a ball clocking 700 revolutions per second, and use it to measure how much spin the player has put on the shot. Spin is the thing that has been wrecking robot table tennis players for decades.

And that reading of spin sits atop incredible end-to-end latency, or speed of sight.

From the ball leaving the bat to Ace’s arm responding takes 20.2 milliseconds.

The same number for an elite human? Around 230 milliseconds.

Ace sees and reacts to the world about eleven times faster than the human standing across the net.

"This is bigger than table tennis. It's the very first time there's been a human expert-level demonstration of competitive play in the real world across any sport — not just table tennis." — Peter Stone, Chief Scientist, Sony AI

Making a racket.

So how do you teach a robot table tennis?

You don't. You let it teach itself.

Ace's "brain" is a deep reinforcement learning system.

Keen NerdNews readers will remember when DeepMind’s AlphaZero played 44 million games of chess against itself over a few hours, to then destroy humanity’s best chess computer.

Or Sony's earlier Gran Turismo Sophy, which beat the world's best video game racing drivers back in 2022.

Similarly, Ace played roughly 3,000 hours of table tennis against itself in simulation before it ever touched a real bat. No coaching. No hand-coded "if topspin, then chop". Just trial and error, millions of times over.

"There's no way to program a robot by hand to play table tennis. You have to learn how to play from experience." — Peter Dürr, Project Lead, Sony AI.

The hardware that executes all that simulated wisdom is the icing on the cake. Ace’s eight-jointed robotic arm with two sliding rails and six rotating joints, returns balls at nearly 20 metres per second.

It even has a built-in cup to hold the ball so it can serve one-handed.

Isn’t that cute … and terrifying!

Does this ping come with a pong?

For those of you getting Skynet vibes, relax. A bit.

Ace's professional opponents found ways to crack it. Low-bounce serves and tricky "knuckle" balls (no spin, unpredictable wobble) exposed gaps. The pros adapted between points. Ace, for now, doesn't.

It's a phenomenal player. It is not yet a phenomenal opponent.

But the moment that should make us sit up came after Ace pulled off one particular return.

Watching from the side was Japanese gun Kinjiro Nakamura. He told the researchers no one else would have been able to make that shot. He hadn't thought it was possible.

"But the fact that it was possible … means that there is a possibility that a human could do it too." — Kinjiro Nakamura, 1992 Olympian.

That's the kicker. Robots aren't just catching us in physical skill sports. In some of them, they may end up teaching us.

Don't take your bat and ball and go home … yet.

Sure it looks pretty cool when a robot can win at ping pong.

But what is really exciting is the techniques behind Ace.

Fast vision. Real-time decision-making. A body that can act on those decisions in milliseconds.

Exactly what’s needed for warehouses, surgery, disaster response, and a thousand other places where milliseconds matter and the world doesn’t sit still.