Yunfu Deng

I am a robotics PhD candidate at the University of Wisconsin–Madison, advised by Prof. Josiah P. Hanna. My research focuses on building the predictive tools that let robots act reliably in the physical world—before deployment, not after failure. I work on learning how the world moves and behaves from data, and on turning that knowledge into robots that know what they are about to do, before they do it.

In particular, much of my work asks how a robot can anticipate what will happen before it acts: how to learn world models that predict how the physical world responds to action, and how to perform policy evaluation that estimates how a policy will behave before it is deployed. That is, how can the experience a robot collects—whether in simulation or in deployment—be translated into a more faithful understanding of how it, and the world, will behave?

I believe capable, general-purpose robots will come from machines that understand the physical world and can anticipate what they are about to do in it—not from machines that just talk about it. I am trying to turn Gundam (EX-S / ALICE) into reality.