flâneur

How Can We Make Robotics More like Generative Modeling? | Eric Jang

evjang.com · 4,949 words · saved by 3 readers

I recently gave a talk (YouTube) at the RSS’22 L-DOD workshop. Here’s a lightly edited transcript and slides of the talk in blog form.

I recently gave a talk (YouTube) at the RSS'22 L-DOD workshop . Here's a lightly edited transcript and slides of the talk in blog form. This talk is not so much about how we take gradient steps to train robots, but rather how we as researchers and engineers can iterate effectively on these systems as we scale them up. Since this is a workshop about large offline datasets for robotic learning, I don't need to convince the audience that what matters most for making capable robots is having high quality data. Diverse data can help your neural networks handle situations not seen at training time.

saved by

related reading