Pete Florence: Dense Visual Representations, NeRFs, and LLMs for Robotics
The Gradient: Perspectives on AI
00:00
Can We Just Generically Make Policy Learning Work Better?
The project was led by Danny Dries, who's a student to you, Berlin. The idea of having a latent condition NERF had been explored initially in some follow-up works to the original NERF paper. And then other works has started looking at, okay, well, I can have a latent code, but I want to be able to infer what the latent code is like in the moment when I look at a scene. In this case, since it can benefit from all the beautiful things about 3D learning, but also just regular RGB pixel learning too. Yeah. So the, in practice, what you do is you learn this latent conditions NERF and then
Transcript
Play full episode
