Sony AI table tennis robot defeats professional players, Honor humanoid robot breaks half-marathon world record

Sony AI-developed table tennis robot Ace defeated professional players in official matches with referees, and the research results have been published in Nature; in the same week, Honor’s humanoid robot “Lightning” finished the Beijing Humanoid Robot Half Marathon in 50 minutes and 26 seconds, breaking the human world record.
(Background recap: Elon Musk asserts that pure AI and robot companies will dominate the future, and humans will become corporate burdens)
(Additional context: Imagining RobotFi: What new gameplay could arise when robots are also on the blockchain?)

Table of Contents

Toggle

• How Ace defeated humans
• Why Honor Lightning can run under 51 minutes
• Where the boundaries of Physical AI are moving

50 minutes and 26 seconds. This time broke the human half-marathon world record of 57 minutes and 20 seconds set by Ugandan athlete Jacob Kiplimo in Lisbon, but it was not a human that broke the record—it was a humanoid robot.

In the same week, Sony AI’s table tennis robot Ace defeated a professional player in an official match under the rules recognized by the International Table Tennis Federation, with licensed referees. The research paper was published in Nature.

These two events occurred within the same timeframe, highlighting the rapid emergence of Physical AI (artificial intelligence-driven physical entities operating in real-world environments) from labs into actual competitive arenas.

How did Ace defeat humans

Peter Dürr’s team at Sony AI faced a unique engineering challenge when designing Ace: the changing speed, spin, and flight trajectory of table tennis balls require perception and movement to be coordinated within milliseconds.

Ace’s hardware architecture includes: nine synchronized cameras plus three visual systems responsible for tracking the ball’s movement and spin; eight joint controllers for the paddle: three for positioning, two for orientation, three for striking force and speed. Dürr describes the visual processing speed as “fast enough to capture motion that the human eye can only see as residual images.”

The training method is a key difference. Ace did not learn by observing human actions but was entirely self-trained in a simulated environment. This enabled it to develop different hitting strategies from humans, making it difficult for opponents to predict using habitual ball-reading techniques.

Performance records show that in a test in April 2025, Ace won three out of five matches against elite players; from December 2025 to early 2026, records of defeating professional players began to appear.

Player Mayuta Hirata, who lost to Ace, described a dilemma never encountered in human competition: “Because I couldn’t understand its reactions, I had no way to perceive what kinds of balls it disliked or was good at.” Without emotional signals or body language, opponents lose the psychological cues long relied upon in sports.

Dürr said that the original purpose of Ace’s design was to study how robots can respond quickly and precisely in dynamic environments; the same perception and control technologies can be applied in manufacturing and service robots.

Why can Honor Lightning run under 51 minutes

On April 19, 2026, the Beijing Etron humanoid robot half marathon was held in Daxing District, with a 21-kilometer course from Tongming Lake Park to Nanhai Park. Over 12,000 human runners and more than 100 robots started simultaneously on separate tracks.

Honor’s “Lightning” finished in 50 minutes and 26 seconds, with an average speed of about 25 km/h. For comparison: the world record for top human half-marathoners is 57 minutes and 20 seconds, a gap of 6 minutes and 54 seconds.

At the same event last year, the fastest robot took 2 hours, 40 minutes, and 42 seconds to complete. In one year, the record was shortened by 110 minutes.

The race prioritized autonomous navigation. Another Honor robot, remotely controlled, completed the course in 48 minutes but was not officially ranked. Honor engineers stated that the structural reliability and liquid cooling system verified during Lightning’s development are now ready for deployment in industrial scenarios.

Where are the boundaries of Physical AI moving

These two breakthroughs share a common structure: integrated improvements in perception speed, physical control precision, and autonomous decision-making ability. Sony Ace’s nine-camera perception system corresponds to Lightning’s autonomous navigation; Ace’s simulated self-training aligns with Lightning’s 110-minute record improvement, indicating underlying capabilities are converging.

The next battleground for Physical AI is not competition but manufacturing, logistics, and service: scenarios that also require rapid perception and precise execution in unstructured environments. Ace and Lightning demonstrate that this set of capabilities has matured enough to be quantifiably validated in real-world applications.