zero2robot: a free from-scratch course whose PPO/SAC chapters are a guided reading of CleanRL

[kaushikb11]

Sharing this here because CleanRL is the "read the real thing" reference for two chapters of the course, and the single-file philosophy is a big influence.

zero2robot is a free, from-scratch course on embodied AI: you build a robot brain one readable file per chapter, from a bare MuJoCo loop up to a vision-language-action policy you train and drive in the browser.

The RL chapters build PPO (ch2.1) and SAC (ch2.2) from scratch, one file each, and then send you to read the corresponding CleanRL single-file implementation as the ground truth to compare against. The idea is the same as CleanRL's: no framework to excavate, the code is the lesson. Everything runs on a free Colab T4 or a CPU laptop, --seed everywhere, and it is honest about CPU-vs-GPU reproducibility.

Textbook: https://www.zero2robot.com/
Source: https://github.com/kaushikb11/zero2robot

Thanks for CleanRL. It made the "one clean readable file per algorithm" bar feel achievable. Feedback on the PPO/SAC chapters very welcome.