​Imitation Learning for Robotics 101

​Hey robot friends,

​This time we are taking a break from the latest research and going back to basics. Anjana and I will walk through our dead simple implementation of a transformer-based flow-matching DiT policy.

​With a bit of luck, we will also do something interactive and train a model live together.

The idea is to connect the high-level concepts we have been discussing with what they actually look like in code. So next time we dive into large cutting-edge models, the components should feel less abstract and easier to connect to things we already know.

Original paper: https://dit-policy.github.io