Modern Robot Learning
From world models to VLAs — build modern robot learning systems from scratch.
Pods in this Course
ACT (Action Chunking Transformer) Policies
Coming soon
Diffusion Policy for Visuomotor Control
Coming soon
Vision-Language-Action (VLA) Models
Coming soon
Understanding World Models from Scratch
## How AI agents learn to dream about their environment — and use those dreams to make better decisions
Offline RL & Dataset-Driven Robot Learning
Coming soon
Behavior Cloning at Scale (RT-X Style)
Coming soon
Foundation Models for Embodied Intelligence
Coming soon
Understanding JEPA from Scratch
Understanding JEPA from Scratch
Action Chunking Transfomers
Action Chunking Transfomers - The first SOTA Imitation Learning Method
Diffusion Policy for Robotics
Diffusion Policy for Robotics - The second SOTA Imitation Learning Method
SmolVLA
SmolVLA - The most efficient VLA
Pi0 - Our first Vision Language Action Model
Pi0 - Our first Vision Language Action Model
What are Vision Language Action Models?
What are Vision Language Action Models?
World Models Inside a Robot Policy: VLAs that Dream
Adding a future-prediction dream step inside a vision-language-action policy: DreamVLA-style future visual state prediction before action selection (arXiv 2507.04447), latent world models in VLAs (VLA-JEPA, arXiv 2602.10098), and a from-scratch build unifying a small VLA policy with a learned world model: imagine consequences, then act.