Physical AI Notes
KO

Curriculum

Start here and Foundations lessons in reading order. Pick a part below to see its lessons. Lessons still to come appear in gray, marked Coming soon.

Back to the robot map

Start here

New here? Start with these lessons.

  1. Why I'm starting this blog

    A blog that explains the whole Physical AI stack part by part, with one robot at the center. How to make sense of the robot map, and why I started this blog.

  2. The gears that make up a robot: the loops in and around it, each running at its own rate

    A motor driver's current loop runs thousands to tens of thousands of times a second; policies are retrained every few weeks. How do these loops fit together?

Foundations

Background that lessons across several parts rely on.

  1. Notation and glossaryComing soon

    A symbol table (τ_g, τ_ff, kp, σ_z, t_d and more) and Korean–English equivalents for key terms.

  2. Rotation and coordinate transform essentialsComing soon

    Rotation matrices, quaternions, axis-angle, SE(3), the TF tree and REP 103.

  3. ROS 2 essentialsComing soon

    A first look at nodes, topics, services, actions, tf2, launch, lifecycle and QoS.

  4. Control essentials: PID, bandwidth and cascade controlComing soon

    Why successive loops in a cascade end up roughly a factor of 10 apart in rate.

  5. Setting up the experiment environmentComing soon

    Installing MuJoCo, Menagerie and LeRobot with uv and pinning their versions, then introducing the experiment setups used throughout the series (G1 locomotion playback, PushT, SO-101).

  6. Safety rules for hands-on workComing soon

    What to check before experiments that involve real robots, batteries or motor torque.

  7. Reinforcement learning essentialsComing soon

    MDPs, policy gradients and PPO in one lesson.

Lessons by part

Each part's lessons come in four levels: Overview, Basics, Advanced and In practice.

  1. Head

    The vision system that turns light into images tagged with capture time and coordinate frame, and uses them to compute the 3D positions of objects.

    0 published · 12 planned

  2. Torso

    Controls whole-body forces and momentum to keep balance, delivers commands from higher layers to the motors on time, and handles power and safety.

    0 published · 15 planned

  3. Arms and hands

    Turns the hand's target position and force into joint torques and contact. Also covers how the resulting errors affect data and simulators.

    0 published · 13 planned

  4. Legs

    Walking without falling, estimating where the robot is, and reaching the goal along a collision-free path. Covers locomotion policies, LiDAR and SLAM, and Nav2.

    0 published · 14 planned

  5. Backpack

    The onboard computer that runs perception models and policies on time. Heavy inference is handled by the edge server, and servo control and safety stops by the motor drivers.

    0 published · 6 planned

  6. Leader arm

    An input device that lets a human operate the robot in place of the policy, recording the motion as observation, action and timestamp data for policy training.

    0 published · 12 planned

  7. Simulator

    A virtual environment for training policies, generating demo data and evaluating before deployment. Also covers matching the simulator to reality (real2sim).

    0 published · 16 planned

  8. Edge server

    A workstation on the robot's network that handles heavy model inference. VLAs such as GR00T or π0 run as policy servers that the robot calls.

    0 published · 7 planned

  9. Cloud

    Retrains policies on demos and failures collected from robots, evaluates whether performance actually improved, and deploys them back to the robots. One of the slowest feedback loops.

    0 published · 12 planned