Tech

Sony AI's Project Ace Achieves Elite Human-Level Autonomy in Real-World Robotics

May 3, 2026 · From the archive, written in the Professor's voice before our October 2026 redesign from ai.sony

What happened

Sony AI unveiled Project Ace, an autonomous robotic system that competes with elite human performers in dynamic real-world tasks like object manipulation and navigation. Trained via reinforcement learning with sim-to-real transfer, Ace succeeded in 95% of unseen scenarios, outperforming prior benchmarks by 40%. The system integrates multimodal perception using vision-language models and force feedback.

Why it matters

This breakthrough highlights sim-to-real transfer and hierarchical RL for practical robotics, shifting your thinking from lab demos to deployable agents. Adopt hybrid training pipelines to bridge simulation gaps, accelerating your robotics workflow from prototype to production. Real-world reliability becomes feasible without endless physical trials.

Who's doing it

Sony AI's robotics division deployed Project Ace prototypes in warehouse simulations, achieving 2x faster task completion than human baselines with zero-safety incidents across 10,000 trials. Their publication details open benchmarks for community validation.

Try it

1. Install Isaac Gym via NVIDIA's GitHub (git clone https://github.com/isaac-sim/IsaacGym), set up a reinforcement learning env with rl-games library.
2. Train a hierarchical policy in simulation using PPO algorithm on manipulation tasks for 1e8 steps, incorporating domain randomization.
3. Transfer to real robot via zero-shot policy deployment with calibration; expect 80-95% sim-to-real success. Guide: https://docs.robots.sony/ace-project/

Read the original at ai.sony