Sony AI's Ace Robot Outperforms Pro Athletes via Reinforcement Learning Breakthrough
Sony AI published in Nature on Ace, an autonomous bipedal robot using advanced force-torque sensors and model-based reinforcement learning. Ace beats professional athletes in agile tasks like dynamic running and jumping. It achieves 10-20% higher success rates in real-world physics simulations turned physical trials.
This demonstrates RL with precise sensors unlocks real-world robotics beyond simulations. Shift thinking from sim-only training to sensor-rich hardware loops. Workflow now integrates MuJoCo sims with real hardware for rapid iteration.
Sony AI's Team led by Peter K. Allen. Ace outperformed human pros in 5/7 athletic benchmarks, including 2m high jumps and 5m/s sprints with 95% reliability in unstructured environments.
Step 1: Install Stable Baselines3 and MuJoCo via pip install stable-baselines3[mujoco]; download Humanoid-v4 env from Gymnasium. Step 2: Train PPO agent with custom force-torque observations using sb3's PPO('MlpPolicy', env, verbose=1). Step 3: Evaluate on 100 episodes; expect 20-50% score improvement over baselines toward pro-athlete metrics like velocity>4m/s. URL: stable-baselines3.readthedocs.io.