Sony’s table tennis robot Ace smashes human-like play

Sony AI’s new table-tennis bot Ace has cleared the 2100-plus Elo bar—placing it in the expert tier among human players. According to the company, Ace is the first robot to reach this level in any sport, a milestone announced on The Decoder. The machine’s camera-laden eyes clock incoming spin at 1000 fps, faster than the human eye can register, while its carbon-fiber arm adjusts paddle angle within milliseconds, matching or beating professional reaction times.

That speed hides the hardware wall: under bright LED panels Ace can maintain play, but dim club lighting triggers frame drops that turn its 1 ms “expert” latency into 20 ms unforced errors. Early signals suggest the demo runs on proprietary neural nets trained on 50,000 rallies, yet Sony hasn’t published baseline specs for torque, paddle pressure, or power draw—critical numbers for gyms weighing the $50k price tag.

Practically, Ace’s target isn’t tournaments; it’s filling club courts during off-peak hours. Operators would need air-conditioned enclosures, ceiling-mounted cameras, and custom power rails to stop the robot from overheating mid-rally. Community chatter on Reddit already questions whether club owners will foot the bill for a machine that still shuts down after 90 minutes of continuous use.

The real signal here is latency, not Elo. For robotics to leap from demo to deployment the next step is adaptive error correction, not faster loops. Until the hardware can handle dust, glare, and vibration the same way a tired amateur can, the gap between “expert level” and “expert useful” remains wide.