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

The table tennis robot Ace, developed by Sony AI, defeated professional players under official rules with licensed referees, and the research results have been published in the journal Nature; in the same week, the Honor humanoid robot "Lightning" completed the Beijing Humanoid Robot Half Marathon in 50 minutes and 26 seconds, breaking the human world record. (Previous coverage: Musk asserts: Pure AI and robotics companies will dominate the future, and humans will become a burden to enterprises) (Background supplement: Imagining RobotFi: What new gameplay will emerge when robots go on-chain?) 50 minutes and 26 seconds. This time breaks the human half-marathon world record of 57 minutes and 20 seconds set by Ugandan athlete Jacob Kiplimo in Lisbon, but the record-breaker is not a human, but a humanoid robot. In the same week, Sony AI's table tennis robot Ace defeated professional players in an official match held under the rules recognized by the International Table Tennis Federation and presided over by licensed referees. The research paper was published in the journal Nature. The occurrence of these two events within the same timeframe highlights the concentrated explosion of Physical AI (AI driving physical machines to operate in real-world environments) moving from laboratories into real competitive arenas. When the team led by Peter Dürr at Sony AI designed Ace, they faced a specific engineering problem: the speed, spin variations, and flight trajectory of a table tennis ball require perception and movement to be coordinated within milliseconds. Ace's hardware architecture consists of: 9 synchronized cameras and 3 vision systems responsible for tracking the ball's movement and spin; 8 joints to control the paddle: 3 for positioning, 2 for direction, and 3 for hitting force and speed. Describing the visual processing speed, Dürr said: "Fast enough to capture movements that the human eye can only see as a blur." The training method is the key differentiator. Ace did not learn by observing human movements, but trained entirely in a simulated environment. This allowed it to develop hitting strategies different from humans, making it difficult for opponents to predict its moves using habitual ball-reading methods. Performance records show: In tests in April 2025, Ace had a record of 3 wins and 2 losses against elite players; from December 2025 to early 2026, records of defeating professional players began to appear. Mayuka Hirata, a player who lost to Ace, described a dilemma never encountered in human duels: "Because I couldn't understand its reactions, there was no way to perceive what kind of balls it didn't like or wasn't good at." Without emotional signals or body language, the opponent lost the psychological information long relied upon in competitive sports. Dürr stated that the original purpose of Ace's design was to study how robots can react quickly and accurately in dynamic environments, and the same perception and control technologies can be applied to manufacturing and service robot fields. On April 19, 2026, the Beijing Yizhuang Humanoid Robot Half Marathon was held in Daxing District, with a 21-kilometer course extending from Tongming Lake Park to Nanhaizi Park. Over 12,000 human runners and more than 100 robots started simultaneously on separated tracks. "Lightning," developed by Honor, finished in 50 minutes and 26 seconds, with an average speed of about 25 kilometers per hour. For comparison: the world record for the half marathon by a top human athlete is 57 minutes and 20 seconds, a difference of 6 minutes and 54 seconds. In the same event last year, the fastest robot finished in 2 hours, 40 minutes, and 42 seconds. In one year, the record was shortened by 110 minutes. The competition rules prioritized autonomous navigation. Another Honor robot finished the course in 48 minutes under remote control, but it was not counted in the official rankings. Honor engineers stated that the structural reliability and liquid cooling system verified during the development of Lightning are already ready for deployment in industrial scenarios. These two breakthroughs share a common structure: the integrated improvement of perception speed, physical control precision, and autonomous decision-making capabilities. Sony Ace's 9-lens perception system corresponds to Lightning's autonomous navigation; Ace's simulated self-training corresponds to Lightning's 110-minute improvement curve; the underlying capabilities are converging. The next battlefield for Physical AI is not competition, but manufacturing, logistics, and services: scenarios that also require rapid perception and precise execution in unstructured environments. What Ace and Lightning provide is the first batch of quantifiable external verification that this set of capabilities is mature enough to be put into action.