Sony AI's Project Ace: Finally, Real-World Robotics That Rivals Elite Humans
Sony AI published Project Ace, a fully autonomous robotic system trained via reinforcement learning in simulated and real-world environments. It excels in table tennis, achieving rally durations competitive with professional players: over 100 strokes per match. The system uses a vision-language-action model integrated with high-fidelity physics simulation for zero-shot transfer to reality.
This demonstrates end-to-end reinforcement learning with sim-to-real transfer, bridging the gap between simulation and physical deployment. Change your thinking: simulation-first training accelerates robotics development by orders of magnitude. In your workflow, adopt MuJoCo or Isaac Gym for scalable RL training before hardware tests.
Sony AI's robotics division deployed Project Ace, enabling a robot arm to sustain table tennis rallies averaging 137 strokes against elite human players, surpassing prior benchmarks by 30%.
Step 1: Install Sony's open-source tiered learning framework from GitHub: git clone https://github.com/SonyAI-Ace/project-ace.git; pip install -r requirements.txt. Step 2: Train in simulation using their RL pipeline: python train.py --task table_tennis --sim mujoco, targeting 100+ episode rewards. Step 3: Transfer to real robot via domain randomization; expect 80% sim performance on hardware like Franka Emika Panda arm.