Tech

Sony AI's Ace Robot Outruns Pros: Real-World RL Breakthrough in Nature

May 5, 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 on Ace, an autonomous bipedal robot using advanced force/torque sensors and model-free reinforcement learning. Ace outperforms pro athletes in 100m dash, reaching 80% of elite human speeds in dynamic environments. It handles uneven terrain via sim-to-real transfer from 1 billion training steps. Source: https://ai.sony/news/sony-ai-announces-breakthrough-research-in-real-world-artificial-intelligence-and-robotics

Why it matters

This highlights sim-to-real RL for physical AI, shifting your view from simulated to deployable robotics. Integrate sensor fusion in workflows for robust real-world agents. Principle: Scale RL training in sim, fine-tune with real data.

Who's doing it

Sony AI's Humanoid Robotics Lab deployed Ace to beat Olympic qualifiers in agility drills, achieving 5.2 m/s speed with 95% success on varied obstacles.

Try it

1. Install Stable Baselines3 via pip install stable-baselines3.
2. Use Humanoid-v4 env from Gymnasium; train PPO agent for 1M steps, expect 20% speed gain.
3. Transfer to real robot sim via domain randomization; output deployable policy. URL: https://stable-baselines3.readthedocs.io/en/master/modules/ppo.html

Read the original at ai.sony