Why Robots Haven't Had Their “ChatGPT Moment” Yet

Pannag Sanketi spent years at the frontier of robot learning at Google DeepMind. At AISE 2026, he unpacks the hard parts of building physical AI at scale — and where the field goes next.

AI agents have moved out of the research lab and into daily life. Robots have not — at least not yet. Everywhere you look, humanoid robots dominate the headlines, but they still aren't in our living rooms or, for the most part, on our factory floors. Why?

The session

In “From Research to Reality: The Hard Parts of Building Physical AI at Scale,” Pannag Sanketi explains why real-world robotics is so much harder than text and images. Robots must perceive objects, understand space, keep their balance, and respond to the unexpected — all while contending with mechanical precision, battery life, sensors, and safety.

The gap between an impressive demo and consistent, production-grade operation is still wide. Most humanoid demonstrations are filmed in controlled environments; the real world is far messier. Sanketi's session is a grounded look at what it actually takes to move physical AI from the lab into the field.

Why it matters now

The timing is sharp. SoftBank's Masayoshi Son says the next trillion-dollar company will come from physical AI, and Barclays projects the humanoid market growing from roughly $3B today toward $200B by 2035. Capital and talent are pouring in — but the hard technical problems remain. Understanding them is the difference between betting on hype and betting on what's real.

About the speaker

Pannag Sanketi was, until recently, a Robotics Lead at Google DeepMind, and is now Founder of Avolla Inc. His work sits at the core of what makes robots intelligent: how they learn from data. He is a co-creator of Open X-Embodiment, the largest open robotics dataset, built with 170+ researchers worldwide; developed an autonomous robot table-tennis system featured in IEEE Spectrum; and is a key author on RT-2, which integrated internet-scale vision-language data directly into robot control.

Key Takeaways

Why robots lag LLMs — the physical world is far more complex than text and images, demanding perception, balance, and real-time adaptation.
Data is the unlock — open datasets and robot-learning breakthroughs like Open X-Embodiment and RT-2 are what move the field forward.
Demo vs. deployment — a realistic view of what separates a viral robot video from production-grade physical AI.