Concepts

These pages explain how wbc-mjlab is put together — the mental model before you dive into CLI commands, MDP field reference, or extension how-tos.

Suggested reading order

Topic	Page
Layer stack	Modularity — shared MDP, robot entity, preset, task
Config assembly	Architecture overview — pipeline from template → --task
Paper recipes	Presets and tasks — apply_* vs registered task ids
Curriculum sampling	Adaptive sampling (RSI) — RSI bins and failure-driven resets
Reference obs layout	Observations — modular ref_* terms, paper layouts, per-term noise
New hardware	Robots and extensions — why robots must register

Where to go next

Need	Section
Field-level MDP reference	Shared MDP
Train, resume, multi-GPU	Training
Task catalog + apply_zest code	Tasks and presets
Build an extension package	Extensions
Viser RSI / reference overlays	Visualization (Viser)