Sony AI's Ace Robot Outpaces Pro Athletes: Reinforcement Learning Hits the Real World
Sony AI published in Nature the Ace system, an autonomous bipedal robot that outperforms professional athletes in dynamic tasks like multidirectional running and jumping. Ace uses advanced LiDAR sensors, force-plate feedback, and model-based reinforcement learning with privileged information during training. It achieves speeds up to 3 m/s and handles perturbations 20% better than prior robots.
This demonstrates how sim-to-real transfer via reinforcement learning unlocks robust real-world AI, shifting focus from simulation-only training to hybrid methods with sensor fusion. You will adapt workflows to include domain randomization, enabling robots or agents to generalize beyond labs. It changes thinking: physical AI viability now hinges on scalable RL, not just compute.
Sony AI's robotics division, under Director Peter Stone, deployed Ace prototypes that beat human baselines in 5 athletic metrics, as validated in Nature. Their framework has accelerated industry pilots, with partners reporting 30% faster task learning.
Step 1: Install Isaac Gym via NVIDIA's preview release (requires Ubuntu 20.04 and RTX GPU); this simulates physics for RL training. Step 2: Implement PPO algorithm from Stable Baselines3 (pip install stable-baselines3) with domain randomization on bipedal tasks, training for 10M steps to achieve stable locomotion. Step 3: Transfer to real robot using sim-to-real with LiDAR calibration; expect 80% policy success rate in dynamic environments. URL: https://ai.sony/news/sony-ai-announces-breakthrough-research-in-real-world-artificial-intelligence-and-robotics.