Sony AI announced a significant milestone in physical artificial intelligence today with the unveiling of Project Ace, the first autonomous robotic system to achieve competitive parity with elite human table tennis players. The research, titled Outplaying Elite Table Tennis Players with an Autonomous Robot, was published on the cover of the journal Nature on April 23, 2026. This development marks the first time an AI-driven machine has reached expert-level proficiency in a high-speed, adversarial physical sport.
Project Ace utilizes a sophisticated hardware and software stack to manage the millisecond-scale demands of table tennis. The system’s perception layer consists of nine active pixel sensor (APS) cameras that triangulate the ball’s position in 3D space at a frequency of 200 Hz. To account for the complex aerodynamics of spin, the robot employs three specialized gaze control systems featuring event-based vision sensors and pan-tilt mirrors. These sensors allow the system to track the ball's logo and determine angular velocity in real time.
The robot’s physical architecture is centered on an eight-jointed robotic arm mounted on a movable track. This configuration provides the degrees of freedom necessary to execute a wide range of shots, including backspin and aggressive edge placements. The control system is powered by deep reinforcement learning policies trained in simulation via an asymmetric actor-critic architecture. These policies generate actions every 32 milliseconds, which are then processed by a model predictive control planner to ensure collision-free movement.
In performance evaluations conducted between April 2025 and March 2026, Project Ace demonstrated a progressive increase in skill. Initial tests in April 2025 saw the robot win three out of five matches against elite amateur players—defined as individuals with over a decade of experience. While it initially struggled against top-tier professionals, subsequent refinements led to a breakthrough in March 2026. During those trials, Ace defeated three professional players, including Miyuu Kihara, a top-25 world-ranked athlete, at least once. Data from the matches showed the robot maintained a return rate above 75 percent and successfully scored 16 unchallenged aces.
Peter Durr, Director of Sony AI in Zurich and project lead, stated that the achievement proves autonomous robots can match human reaction times and decision-making in complex physical spaces. Michael Spranger, President of Sony AI, noted that while previous AI successes like Gran Turismo Sophy occurred in virtual domains, Project Ace demonstrates the transferability of these technologies to the real world. Sony AI officials indicated that the underlying research into real-time sensing and adaptive control has potential applications beyond sports, specifically in manufacturing and service robotics where high-speed human-robot interaction is required.