Tech

Sony AI's Ace Robot Outperforms Pro Athletes via Reinforcement Learning Breakthrough

May 9, 2026 · From the archive, written in the Professor's voice before our October 2026 redesign from ai.sony

What happened

Sony AI published in Nature a system called Ace, an autonomous bipedal robot using advanced LiDAR sensors and model-based reinforcement learning. Ace beats professional athletes in agility tasks like multidirectional running and jumping, with 20% faster sprint times and 15% higher jump heights. The method integrates MuJoCo physics simulation for 10 million training steps.

Why it matters

This validates sim-to-real transfer in robotics, bridging simulation gaps with domain randomization. Rethink AI training: combine RL with real-world sensors for deployable agents. Your robotics projects now scale to human-level performance in dynamic environments.

Who's doing it

Sony AI's robotics division trained Ace to achieve world-record robot agility scores, outperforming humans in 5 of 7 athletic benchmarks as per their Nature paper.

Try it

1. Install Stable Baselines3 via pip install stable-baselines3 and MuJoCo via pip install mujoco.
2. Define a bipedal walker env in Gymnasium with domain randomization (e.g., vary friction 0.5-1.5); use PPO algorithm with n_steps=2048.
3. Train for 1M steps and deploy to hardware with ROS; expect 2x human speed in locomotion tasks. URL: https://stable-baselines3.readthedocs.io/en/master/modules/ppo.html.

Read the original at ai.sony