Ace, an autonomous AI-powered table-tennis robot, can compete with and beat elite human players under official competition rules.

In April 2025 tests, Ace won three of five matches against top players but fell to seasoned professionals who regularly compete in professional leagues.

By late 2025 and early 2026, Ace improved to beat some elite and several professional players, including Miyuu Kihara, thanks to faster reaction, sharper precision, and aggressive shot placement near the table edge.

The coverage includes a DOI and direct links to the main research article, a News and Views piece, and the accompanying video for readers to explore further.

The research emphasizes real-time physical AI challenges—perception, noisy-sensor state estimation, and adversarial human interaction—over simulated environments.

Beyond sport, the piece highlights broader security and military implications of ultra-fast autonomous systems and the accompanying ethical considerations.

Ace relies on nine active pixel-sensor cameras to track the ball in 3D, plus additional sensors for velocity and spin, all supported by advanced reinforcement learning and precision hardware.

The control pipeline converts noisy observations to actions every 32 ms using a Soft Actor-Critic–trained policy, with multiple skills and a policy sampler; actions are refined into real-time trajectories via convex optimization and an MPC-based reset trajectory for safe play.

Experts see potential for real-time, high-speed control AI in other human-interactive domains like manufacturing and service robotics, beyond table tennis.

Ace’s learning was largely simulated, with real-world play applying learned policies directly to the robot without retraining on new real data.

A genetic algorithm created a library of serves selected by human-informed criteria based on training performance.

The episode highlights key timings and supplementary content, including a segment focused on the table-tennis robot and a gravity measurement discussion.