Tech

Sony AI's Ace Robot Outpaces Pro Athletes in Real-World Tasks via Reinforcement Learning

May 12, 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 breakthrough with Ace, an autonomous bipedal robot using advanced LiDAR sensors, proprioceptive feedback, and model-based reinforcement learning. Ace outperforms professional athletes in agile tasks like hurdle racing and backflip sequences, achieving 95% success rates in dynamic environments. The system integrates sim-to-real transfer to bridge simulation gaps.

Why it matters

This validates end-to-end reinforcement learning with sim-to-real adaptation for robotics, proving AI can master physics-rich, unpredictable settings. Shift your thinking: virtual training accelerates real-world deployment 1000x faster than manual programming. Workflow now prioritizes RL frameworks over hardcoded rules for adaptive agents.

Who's doing it

OpenAI's robotics team deployed similar RL-trained humanoid prototypes, achieving 30% faster manipulation speeds than humans in warehouse tasks, as shown in their 2024 dexterous hand demos.

Try it

1. Install Isaac Gym with 'pip install isaacgym'.
2. Set up a bipedal robot env and train PPO policy: from rl_games import train; train('legged_robot.yaml', num_envs=4096) for 10M steps.
3. Deploy to real robot via sim-to-real domain randomization; expect 80% policy transfer success. URL: https://github.com/isaac-sim/IsaacGymEnvs.

Read the original at ai.sony