Max Robotics — Global Service Robot Directory + US Certification

About

Julian Ibarz is a leading researcher at the intersection of deep reinforcement learning (RL) and real-world robotics, whose work has fundamentally advanced how robots learn complex manipulation skills. His primary research areas include vision-based robotic manipulation, safe reinforcement learning, and grounding large language models for robotic control. Ibarz’s most influential contribution is QT-Opt (575 citations), a scalable RL framework that enabled robots to learn vision-based grasping directly from camera inputs, bypassing traditional hand-coded perception pipelines. He also co-authored the widely-cited "How to train your robot with deep reinforcement learning" (536 citations), which distilled practical lessons from years of real-world RL deployment. In the landmark paper "Do As I Can, Not As I Say" (516 citations), Ibarz helped pioneer the SayCan framework, which grounds large language models in robotic affordances—a breakthrough that allows robots to follow high-level natural language commands while respecting physical constraints. His work on RT-1 (512 citations) introduced a robotics transformer architecture capable of scaling real-world control across diverse tasks. Ibarz has also advanced safe RL through Recovery RL (193 citations), enabling robots to learn autonomously while avoiding dangerous states. With over 2,700 total citations, his research continues to shape how robots learn, adapt, and safely interact with the physical world.