zero2robot: a free, from-scratch robotics course built entirely on MuJoCo (with MuJoCo compiled to WASM for a browser playground)

[kaushikb11]

Sharing something I built entirely on MuJoCo, in case it is useful to people here.

zero2robot is a free, from-scratch course on embodied AI. You build a robot brain one readable file per chapter, starting from a bare MuJoCo simulation loop and working up to a vision-language-action policy you train, evaluate, export, and drive yourself. The idea is that the code is the product: every chapter is one file you read like a textbook and run, no framework to excavate.

MuJoCo does all the heavy lifting:

• Every environment and every chapter runs on MuJoCo. CPU env resets are bitwise-reproducible, which the course leans on to teach determinism and honest evaluation.
• I compiled MuJoCo to WebAssembly so learners can run the sim and drive trained policies in the browser with no install: https://play.zero2robot.com/
• It all runs on a free Colab T4 or a CPU laptop. No robot or GPU required.

Textbook: https://www.zero2robot.com/
Playground (MuJoCo in WASM): https://play.zero2robot.com/
Source: https://github.com/kaushikb11/zero2robot

Would love feedback from this community, especially on the MuJoCo modeling and determinism side, where I tried to be honest about what reproduces bitwise (CPU) and what does not (GPU). Thanks to the MuJoCo team; this course would not exist without it.