Sony AI put a robot on the cover of Nature for beating pro table tennis players. Project Ace is a fixed base with a 6-DoF robotic arm holding a paddle, paired with event-based vision sensors that track the ball in 3D at 200 Hz. In matches under official competition rules in March 2026, it took at least one game off all three pros it played — 7 games out of 13, 3 matches out of 5. First time an autonomous system has done this in a real physical sport.
The hardware
Forget the LLM-on-a-stick demos. Ace is a real-world rig: 6-DoF robotic arm plus event-based cameras locating the ball with 3.0 mm spatial error and 10.2 ms perception latency. End-to-end motor latency is 20.2 ms. Humans clock around 230 ms. The policy is model-free reinforcement learning, trained entirely in simulation, then dropped onto the real hardware.
Why this matters
Robots beating humans at chess and Go is old news — those are software games. Ping pong is 3D, sub-second, physical. Sub-25 ms reaction time on real hardware is the new benchmark for sim-to-real RL, and Sony is quietly back in the AI conversation after years of being written off.
You Might Also Like
- Figure 03 Helix 02 Runs 67 Hours Autonomous With one Error 2026 Production Already Sold out
- Opendataloader pdf Scores 0 90 Accuracy Across 200 Real World Pdfs Highest Among Open Source Parsers
- Sentrysearch Turns Hours of Video Into a Searchable Index for 2 50 Using Googles new Multimodal Embeddings
- Alibaba Amap Embodied Quadruped Abot World Alibabas First Robot Just Beat Google and Nvidia on the Leaderboard
- Reliable Robotics Autonomous Cessna 208b 200 Orders 160m Raised First Pilotless Cargo Route Launches This Summer
Top AI Product 
Leave a comment