Modern Robot Learning from Scratch - Lecture 4 - The Transformer Architecture for Robotics | Modern Robot Learning from Scratch

Modern Robot Learning hinges on intelligent policies which can be embedded into hardware. These policies are almost always built upon the "Transformer" Architecture. Have a look at the most common AI policies which are implemented in the Modern Robot Learning (by Google, Nvidia, Physical Intelligence Labs, HuggingFace): (1) Action Chunking with Transformers (ACT) (2) Groot - Vision Language Action Model (3) Pi0 - Vision Language Action Model (4) SmolVLA - Vision Language Action Model The Transformer architecture is at the heart of vision-language-action models. If anyone wants to truly understand Modern Robot Policies, understanding Transformers becomes imperative. This is exactly what we cover in detail in this lecture