😺 Sony's new robot can beat professional ping pong players
Summary
Sony AI's new robot, Ace, has achieved a significant milestone by becoming the first physical, real-time, adversarial AI to consistently beat elite human ping-pong players under official ITTF rules. The robot utilizes nine cameras to triangulate ball position at 200 Hz, three event-based vision cameras to read ball spin, and boasts a latency of 10.2 milliseconds. Its deep reinforcement learning (RL) policy was trained entirely in simulation with zero fine-tuning on a real court. Ace wins by nearly never missing, returning 75% of shots with up to 450 rad/s of spin, rather than by powerful winning shots. This achievement, detailed in a Nature cover paper, marks a departure from previous AI victories in digital or static games, highlighting advancements in real-world robotics and AI.
Key takeaway
For CTOs and VP of Engineering evaluating AI's physical capabilities, Sony's Ace robot demonstrates that highly specialized AI systems can now outperform humans in complex, real-time physical adversarial environments. You should consider how these advancements in sensor fusion, low-latency control, and simulation-trained reinforcement learning could translate to industrial automation, precision manufacturing, or other dynamic robotic applications within your organization.
Key insights
Sony's Ace robot is the first AI to beat elite humans in a real-time, adversarial physical sport.
Principles
- Simulation-trained RL can transfer to complex physical tasks.
- High-speed sensing and low latency are critical for real-time robotics.
Method
Ace triangulates ball position with nine cameras at 200 Hz, reads spin with event-based vision cameras, and uses a deep reinforcement learning policy trained solely in simulation to drive its arm.
In practice
- Explore event-based vision for high-speed object tracking.
- Investigate sim-to-real transfer for robotic control systems.
Topics
- Sony Ace Robot
- Table Tennis AI
- Deep Reinforcement Learning
- Computer Vision Systems
- Anthropic Pricing Strategy
Code references
Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Product Manager, Tech Journalist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by The Neuron.