Tech

Sony AI's Ace Robot Outpaces Pro Athletes: Real-World RL Milestone in Nature

May 3, 2026 · From the archive, written in the Professor's voice before our October 2026 redesign from ai.sony

What happened

Sony AI introduced Ace, an autonomous robotic system using advanced force-torque sensors and model-based reinforcement learning. Ace outperformed professional athletes in a dynamic ball-catching task, achieving 80% success rate in unpredictable throws versus humans' 50%. Published in Nature, it marks a breakthrough for AI in unstructured environments.

Why it matters

This demonstrates the power of sim-to-real transfer in reinforcement learning: train in simulation, fine-tune with real sensors for robust physical AI. Change your thinking: dynamic environments demand hybrid model-based RL over pure data-driven methods. Workflow update: always pair simulators with sensor fusion for robotics prototyping.

Who's doing it

OpenAI's robotics team, with researcher Sergey Levine, applied similar RL to dexterous manipulation bots, enabling a Shadow Hand to solve Rubik's Cube in under 60 seconds blindfolded.

Try it

1. Install Isaac Gym simulator from NVIDIA at https://developer.nvidia.com/isaac-gym, run 'pip install isaacgym'.
2. Load a ball-catching RL env, train with PPO algorithm specifying force-torque observations for 1M steps.
3. Transfer to real robot via domain randomization; expect 70%+ success in initial real-world trials after 10 episodes.

Read the original at ai.sony