Overview
- Sony AI detailed Ace in a peer‑reviewed Nature paper published Wednesday, reporting wins against elite players under official rules with accredited referees.
- After submission, the team staged December 2025 tests against two professionals and two elite amateurs, with Ace beating both elites and one pro as shot speed and edge placement improved.
- The system uses nine high‑speed cameras with active‑pixel sensors and three event‑based gaze units to track the 40 mm ball, estimate spin, and compute flight paths in about 10.2 milliseconds.
- An eight‑joint arm executes the learned control policy and returned more than 75% of spinning shots within tested limits, winning points through serve placement and varied strokes.
- Built with deep reinforcement learning and lessons from Sony’s Gran Turismo Sophy, the platform is pitched for industrial and assistive uses, and Sony AI’s president said similar fast‑sensing hardware could see military use.