AI Roadmaps
← All roadmaps
AdvancedRoadmap 13 · 10 steps · 7h 28m of video

Robotics & Embodied AI

Teach robots with data: robot learning basics, imitation learning, ACT, diffusion policies, vision-language-action models, and sim-to-real.

Before this: Reinforcement Learning. Deep Learning. Hardware: step 9 wants a recent NVIDIA GPU or a cloud GPU; step 10 needs a physical SO-101 arm, so treat it as optional and use the ACT and Diffusion Policy steps with LeRobot simulation datasets instead.

  1. 1h 05m
    Step 1

    Introduction to robot learning

    What makes learning for robots different from other ML.

    Robot Learning 2026 – Lecture 1: Introduction to Robot Learning | ETH Zürich

    Oier Mees

  2. 51 min
    Step 2

    Imitation learning

    Behavior cloning, its failure modes, and common fixes.

    Robot Learning 2026 – Lecture 3: Imitation Learning | ETH Zürich

    Oier Mees

  3. 56 min
    Step 3

    ALOHA and Action Chunking Transformers

    Predict chunks of actions for smooth, precise manipulation.

    ALOHA and ACT: LeRobot Research Presentation #1 by Alexander Soare

    Hugging Face

    huggingface/lerobot★ 29k
  4. 13 min
    Step 4

    Diffusion policy

    Generate robot actions with diffusion models.

    Why Diffusion Policy Is Changing Robot Learning

    Robraintics

  5. 35 min
    Step 5

    Vision-language-action models

    How LLM-style models become robot controllers.

    LLMs Meet Robotics: What Are Vision-Language-Action Models? (VLA Series Ep.1)

    Ilia

  6. 8 min
    Step 6

    VLA models compared

    OpenVLA, pi0, and RT-2 side by side.

    Vision Language Action Models - OpenVLA, π0, RT-2, Gemini Robotics

    Architecture Bytes - AI

    openvla/openvla★ 7.3k
  7. 1h 16m
    Step 7

    Pi0: a generalist robot policy

    A flow-matching VLA trained across many robots.

    Pi0 - generalist Vision Language Action policy for robots (VLA Series Ep.2)

    Ilia

    Physical-Intelligence/openpi★ 14.7k
  8. 1h 01m
    Step 8

    Robot foundation models

    Sergey Levine on where generalist robot models are headed.

    Sergey Levine (UC Berkeley): Robot Foundation Models

    EMBER Center @ UC Berkeley

  9. 9 min
    Step 9

    Reinforcement learning for robots in simulation

    Train locomotion policies in Isaac Lab.

    Reinforcement Learning For Robots in Python: Isaac Lab Tutorial

    NeuralNine

    isaac-sim/IsaacLab★ 8.7k
  10. 1h 15m
    Step 10

    Optional: train a real robot arm with LeRobot

    Hardware-dependent: collect demos and train a policy on a low-cost arm.

    How to Train an AI Robotic Arm With No Code (Hugging Face LeRobot + SO-101)

    Brogan M. Pratt

    huggingface/lerobot★ 29k

Build it

Watching isn't learning. Pick one and put it on GitHub.

Next roadmap