Sony AI's Ace Robot Outperforms Pro Athletes via Reinforcement Learning Milestone
Sony AI published in Nature the Ace system: an autonomous bipedal robot using advanced force-torque sensors and model-based reinforcement learning. Ace beats professional athletes in dynamic ball-striking tasks, achieving 80% success rate in unpredictable environments. This integrates sim-to-real transfer with impedance control for real-world robustness. Source: https://ai.sony/news/sony-ai-announces-breakthrough-research-in-real-world-artificial-intelligence-and-robotics
This teaches sim-to-real RL transfer: train policies in simulation, then deploy with domain randomization to handle real physics. Shift your robotics workflow from teleoperation to end-to-end learning for agility tasks. Real-world AI now prioritizes sensor fusion over pure vision, enabling superhuman performance in unstructured settings.
Sony AI's robotics team, led by Josh Peters, deployed Ace prototypes that strike balls at 1.5x human pro speeds with 95% precision in lab tests. Their open-source RL baselines improved industry benchmarks by 40% on locomotion tasks.
Step 1: Install Isaac Gym via NVIDIA Hub: nvidia-isaacgym. Step 2: Train a bipedal policy with PPO RL, adding domain randomization (e.g., random friction 0.5-1.5). Step 3: Transfer to real robot via system ID; expect 70% sim-to-real success on first deployment. URL: https://gymnasium.farama.org/environments/muJoCo/mujoco/