Tech

Sony AI's Ace Robot Outperforms Pro Athletes in Dynamic Tasks Via Reinforcement Learning

May 6, 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 force-torque sensors and model-based reinforcement learning. Ace beats professional athletes in multidexterous tasks like ball kicking with 20% higher success rates in unstructured environments. The method combines sim-to-real transfer with 1 million hours of simulated training. Source: https://ai.sony/news/sony-ai-announces-breakthrough-research-in-real-world-artificial-intelligence-and-robotics

Why it matters

Reinforcement learning with rich sensory data unlocks real-world robotics; forget pure vision-only approaches. This shifts your thinking to multimodal sensing for robustness in dynamic settings. Apply sim-to-real pipelines to accelerate prototyping without endless physical trials.

Who's doing it

Sony AI's team, led by Peter K. Allen, deployed Ace to achieve 95% success in soccer dribbling tasks, surpassing human pros by 15% in speed and precision.

Try it

1. Install Stable Baselines3 via pip install stable-baselines3.
2. Set up a MuJoCo environment for bipedal locomotion: from stable_baselines3 import PPO; env = gym.make('Humanoid-v4').
3. Train with PPO for 1M timesteps; expect 2x improvement in forward velocity over random policy. URL: https://stable-baselines3.readthedocs.io/en/master/modules/ppo.html

Read the original at ai.sony