News + Trends
Guide dog 2.0: This robot talks and guides you to your destination
by Kim Muntinga
Sony AI: Table tennis robot Ace beats human professionals
24/4/2026
Translation: machine translated
A robot plays table tennis according to official rules and beats experienced opponents. With Ace, Sony AI shows how far AI has already come in real physical competition.
A serve with a strong topspin, a lightning-fast counter-attack, a long rally. What sounds like a classic table tennis match has one crucial difference: on one side of the table is not a human, but a robot. His name is Ace. He was developed by Sony AI - and he plays at a level that was previously only reserved for very good players.
Sony AI officially unveiled Ace in April 2026, accompanied by a scientific publication in the journal Nature. According to Sony AI, Ace competes under the official rules of the International Table Tennis Federation (ITTF)
As Reuters reports, Ace won three out of five matches against elite amateur players in the matches described in the technical article in April 2025. However, the robot still lost twice against professionals at that time. According to Sony AI, Ace then also defeated professional players in December 2025 and March 2026.
Why table tennis of all sports?
Table tennis has been regarded as a particularly tough discipline in robotics for decades. The ball is small, fast and rotates strongly. A system has to detect its trajectory practically in real time, categorise the spin, plan its own shot and implement the movement immediately. Even slight delays decide whether the ball lands on the table, gets stuck in the net or misses. Sony AI describes precisely this combination of perception, planning and rapid execution as one of the great open frontiers of physical AI.
Ace builds on previous AI milestones, but goes one step further. Sony AI itself refers to the development line from Deep Blue to AlphaGo to GT Sophy, the racing AI for Gran Turismo. The key difference: these systems primarily operated in digital or highly simulated spaces. Ace now brings similar learning principles to the physical world, where blurring, material properties, friction, trajectories and human unpredictability play a much greater role.
How Ace works
Technically, Ace combines several components. The system uses nine synchronised cameras and several image processing units, supplemented by an innovative perception system with event-based sensors. The robot is controlled using a control approach based on deep reinforcement learning.
The sensors detect the trajectory and rotation of the ball, from which the control system calculates the appropriate movement in fractions of a second. This is how Ace realises what is crucial in table tennis: precise positioning, clean timing and controlled rebounds.
Source: Sony AI
In addition, there is a specially developed robot platform with eight joints. According to project manager Peter Dürr, this design is no coincidence, but the minimum required to execute shots at a competitive level. The robot not only has to bring the racket into the correct position, but also align it precisely and generate the necessary acceleration. Only the combination of sensor technology, control and mechanics makes the whole thing competitive.
The training approach is also exciting. Ace first trained completely in the simulation and then transferred what it had learnt to the real machine. It is precisely this step that is considered particularly difficult in robotics. What works in the simulation often fails in practice due to the smallest deviations in material, timing or movement. The fact that Ace makes this leap is what makes the project so remarkable.
Not an unfair trick, but a fairness test
Sony emphasises that Ace is not intended to win through raw, mechanical advantages. The team made a conscious effort to create comparable conditions. The robot should therefore not just hit harder or faster than any human, but actually play table tennis under fair conditions.
Ace is still purely a research system that relies on a controlled environment. However, as a feasibility study, it shows how far physical AI has already come.
Header image: Sony AI
My interests are varied, I just like to enjoy life. Always on the lookout for news about darts, gaming, films and series.
News + Trends
From the latest iPhone to the return of 80s fashion. The editorial team will help you make sense of it all.
Show all
