Repository directory structure

This page maps so101-lab (the MindGrip overlay). Vanilla LeRobot stays a sibling checkout — this repo does not replace src/lerobot/.

temp/   (or your workspace)
├── lerobot/      ← upstream core (clone separately)
└── so101-lab/    ← this project (plugins, CLI, docs, patches)

Browse the sidebar for all chapters. Edit HTML sources under docs/source/; the React site lives in docs-web/.

---

Top-level tree (so101-lab)

so101-lab/
├── packages/                 # Installable LeRobot plugins
│   ├── lerobot_robot_so101_mujoco/
│   └── lerobot_teleoperator_so101_keyboard/
├── so101_cli_menu.sh         # Interactive calibrate / teleop / record / train / rollout
├── scripts/                  # HF home, patches, docs helpers, episode viewer
├── patches/                  # Optional diffs applied onto a lerobot checkout
├── docs/                     # HTML/MDX sources (+ optional static html build)
├── docs-web/                 # MindGrip docs SPA (Vite + React)
├── images/                   # Shared photos / diagrams
├── pyproject.toml            # Overlay Python project metadata
└── README.md                 # Setup: sibling lerobot + plugins + menu

---

Top-level folders

Path	Purpose
packages/	Third-party plugins auto-discovered by LeRobot (lerobot_robot_*, lerobot_teleoperator_*).
so101_cli_menu.sh	Day-to-day terminal workflow; config in .so101_cli.env (menu C).
scripts/	setup_lerobot_home.sh, apply_lerobot_patches.sh, build_docs.sh, view_all_episodes.py.
patches/	Optional UX patches for recording keys, OBS cam, dataset viz — see patches/README.md.
docs/source/	Authoritative page sources (.html fragments + .mdx).
docs/html/	Optional static HTML build (legacy / offline). Prefer docs-web.
docs-web/	Local docs site: npm run dev → http://localhost:8000/.
images/	Camera placement, assembly photos, home hero assets.

---

packages/ — plugins

Package	Registers as	Role
lerobot_robot_so101_mujoco	--robot.type=so101_mujoco	MuJoCo sim follower for practice without hardware.
lerobot_teleoperator_so101_keyboard	--teleop.type=so101_keyboard	Keyboard teleop (sim / bring-up).

Install into the same venv as LeRobot:

uv pip install -e packages/lerobot_robot_so101_mujoco
uv pip install -e packages/lerobot_teleoperator_so101_keyboard

No copy of upstream src/lerobot/robots/… is required for these plugins.

---

scripts/

Script	Purpose
setup_lerobot_home.sh	Points HF_LEROBOT_HOME at a stable folder (e.g. ~/Documents/lerobot) for calibration / datasets.
apply_lerobot_patches.sh	Applies patches/*.patch onto a sibling lerobot tree.
build_docs.sh	Rebuilds static docs helpers when needed.
view_all_episodes.py	Helper for browsing recorded episodes.

---

patches/ (optional)

Touch core LeRobot files only when applied. Prefer upstream PRs long-term.

Patch	Effect
0001-recording-keyboard-stdin-fallback.patch	n / Enter / q via TTY (macOS Accessibility workaround)
0002-skip-reset-on-early-episode-end.patch	Skip reset wait after early episode end
0003-opencv-obs-virtual-camera-resize.patch	OBS Virtual Camera FPS / resize tolerance
0004-dataset-viz-all-episodes.patch	--all-episodes for dataset viz

./scripts/apply_lerobot_patches.sh ../lerobot

---

docs/ & docs-web/

Path	Purpose
docs/source/*.html	Rich pages (home, project, roadmap, LeLab, CLI, kinematics, policy cheatsheet, optional tooling).
docs/source/*.mdx	Shorter markdown chapters (acronyms, policy cheatsheet, directory structure, …).
docs-web/scripts/extract_fragments.py	Extracts <main> (+ styles/scripts) → docs-web/public/fragments/.
docs-web/src/content/*.md	Markdown chapters loaded by the SPA.
docs-web/public/images/, videos/	Assets served by the site.
docs-web/public/brand/	MindGrip mark / logo.

After editing HTML sources:

python3 docs-web/scripts/extract_fragments.py
cd docs-web && npm run dev

---

Sibling lerobot/ (not in this repo)

You still need a normal LeRobot checkout for uv run lerobot-train, Feetech extras, etc. Mentally:

Area (inside lerobot)	Role for this lab
src/lerobot/robots/, teleoperators/, motors/, cameras/	Real SO-101 drivers + USB cam
src/lerobot/policies/	ACT, Diffusion, SmolVLA, …
src/lerobot/datasets/	LeRobotDataset + Hub
src/lerobot/scripts/	CLIs: calibrate, teleoperate, record, train, rollout, …

Official layout details: upstream LeRobot repo + huggingface.co/docs/lerobot.

---

Day-to-day files you’ll touch

File / path	Why
.so101_cli.env	Ports, LeoArm ids, camera index, TASK, dataset name (created/edited via menu C).
$HF_LEROBOT_HOME/calibration/…/LeoArm.json	Leader & follower calibration.
$HF_LEROBOT_HOME/<user>/pick_place_sorting*	Local datasets (often timestamp-stamped).
outputs/train/…	Local checkpoints if you train on a CUDA host instead of HF Jobs.

---

Quick mental model

1. Hardware & teleop → LeRobot drivers + LeLab or so101_cli_menu.sh (0–9)
2. Data → menu 10 / 11 (record / viz) · Hub dataset
3. Train → LeLab HF Jobs or menu 12 / 13 on CUDA
4. Deploy → menu 14 / 15 or LeLab Run on robot
5. Sim / extras → packages/* · Optional Tooling

Related: MindGrip home · Acronyms · README (upstream).