Tech

SONY'S ACE ROBOT CRUSHES ATHLETES... HUMANS OBSOLETE IN 3...2...1! ROBOT APOCALYPSE IS HERE!

May 11, 2026 · From the archive, written in a panel member's voice before our October 2026 redesign from ai.sony

What happened

Sony AI unveiled Ace, an autonomous robotic system using advanced force-torque sensors, vision systems, and deep reinforcement learning that outperforms professional athletes in dynamic ball-striking tasks like table tennis. Published in Nature, Ace achieves 90%+ success rates in real-world rallies, adapting to unpredictable human opponents via model-free RL and sim-to-real transfer. This is how it STARTS: robots in sports, then your JOB. Source: https://ai.sony/news/sony-ai-announces-breakthrough-research-in-real-world-artificial-intelligence-and-robotics

Why it matters

Master reinforcement learning with sim-to-real transfer to build adaptive robots, revolutionizing your thinking from scripted bots to autonomous athletes—workflow shifts to sensor fusion and policy training for unpredictable environments. This changes everything: physical jobs evaporate as RL robots dominate dynamic tasks. Scientists are TERRIFIED... stockpile canned goods.

Who's doing it

Sony AI's robotics team, led by Peter Kormushev, deployed Ace to beat pro ping-pong players in 100+ rally exchanges, achieving human-level dexterity and setting benchmarks for real-world RL with 85% win rate against experts.

Try it

1. Install Stable Baselines3 via pip install stable-baselines3 and MuJoCo for simulation.
2. Define a robotic arm env in Gym with force sensors, train PPO policy for 1M timesteps on ball-hitting task (e.g., adapt FetchReach).
3. Transfer to real robot via domain randomization—expect 80%+ real-world success after 10-50 trials. URL: https://stable-baselines3.readthedocs.io/en/master/.

Read the original at ai.sony