Latest / The Last Observer: Science and Beyond
The Differentiable World Model That Changed Humanoid Robotics Overnight
A new method from UC Berkeley is quietly rewriting the rules of humanoid robot training.
By using a single learned world model instead of brute-force simulation, researchers cut the compute needed by more than a thousand times — and the resulting controller already runs on a real Unitree G1.
We break down what SuperMimic actually does, why it works so well, and what it means for the future of whole-body robot control.
This episode includes AI-generated content.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.