agentleFS
Sign inStart free
Seeed-ProjectsSeeed-Projects/Seeed-Jetson-DevelopTool/skills/codex/lerobot-env-setup/AGENTS.md

Seeed-Jetson-DevelopTool / lerobot-env-setup

AGENTS.md57 starsChanged 4 months agoMITView on GitHub

Set up the LeRobot environment on NVIDIA Jetson (SO-ARM / JetPack 6.0+). Installs Seeed-fork LeRobot with GPU-enabled torch 2.8, pinned numpy 1.26, OpenCV 4.10, and resolves the torch-overwrite trap from editable install. Use when the user asks to install, rebuild, or fix LeRobot on Jetson.

The file

We split this AGENTS.md into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run the install script one phase at a time.

    line 12
  2. Very specific:

    If it printed [STOP] → stop, follow the failure decision tree below

    line 14
  3. Very specific:

    If it ended with [install] Step N: or [OK] → proceed to next phase

    line 15
  4. Distinctive:

    All phases use the same script — the script is idempotent.

    line 26

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

LeRobot Environment Setup (Jetson / SO-ARM)

Execution model

Run the install script one phase at a time. After each phase:

  • Relay the output to the user (show [install] lines)
  • If it printed [STOP] → stop, follow the failure decision tree below
  • If it ended with [install] Step N: or [OK] → proceed to next phase

Why phases? The torch download is ~254 MB and takes 10–20 min. Running everything in one call risks silent timeout. Phases give the user visibility and a recovery point at each step.

Phase commands

Replace <ENV> with the env name the user requested (default: lerobot). All phases use the same script — the script is idempotent; already-done steps are skipped automatically.

Phase 1 — Environment + clone (fast, ~1 min)

bash ~/.codex/skills/lerobot-env-setup/scripts/install_lerobot.sh \
  --env <ENV> --robot-type so-arm --phase env

Phase 2 — Download wheels (slow, 10–20 min, shows % progress)

bash ~/.codex/skills/lerobot-env-setup/scripts/install_lerobot.sh \
  --env <ENV> --robot-type so-arm --phase download

Phase 3 — Install torch + verify CUDA (fast, ~2 min)

bash ~/.codex/skills/lerobot-env-setup/scripts/install_lerobot.sh \
  --env <ENV> --robot-type so-arm --phase torch-pre

Phase 4 — opencv / ffmpeg / numpy + lerobot editable (medium, ~5 min)

bash ~/.codex/skills/lerobot-env-setup/scripts/install_lerobot.sh \
  --env <ENV> --robot-type so-arm --phase deps

Phase 5 — Reinstall torch + validate (fast, ~2 min)

bash ~/.codex/skills/lerobot-env-setup/scripts/install_lerobot.sh \
  --env <ENV> --robot-type so-arm --phase torch-post

Failure decision tree

OutputAction
[STOP] conda not foundAsk user: "Please confirm miniconda is installed at ~/miniconda3 or give path"
[STOP] conda env ... exists but has Python X.YShould not happen (script auto-fixes). If seen, run conda env remove -n <ENV> then re-run phase 1
[STOP] torch wheel not found after downloadSharePoint URL expired. Ask user to re-share wheel files or place them manually in ~/wheels/
[STOP] torch.cuda.is_available() = FalseWheel may be corrupt. Run: rm ~/wheels/*.whl then re-run phase 2 and 3
[FAIL] Some checks failedRead which line shows ✗, then check references/conflict_resolution_playbook.md for F1–F7
Script hangs with no output >5 minReport to user, ask whether to cancel and retry

After all phases pass — Serial port (SO-ARM only)

if dpkg -l brltty 2>/dev/null | grep -q '^ii'; then sudo apt remove brltty; fi
if ! id -nG | grep -qw dialout; then sudo usermod -aG dialout $USER; fi
if [[ ! -f /etc/udev/rules.d/99-serial-ports.rules ]]; then
  sudo tee /etc/udev/rules.d/99-serial-ports.rules > /dev/null <<'EOF'
KERNEL=="ttyUSB[0-9]*", MODE="0666"
KERNEL=="ttyACM[0-9]*", MODE="0666"
EOF
  sudo udevadm control --reload-rules && sudo udevadm trigger
fi

Reference files

  • references/conflict_resolution_playbook.md — F1–F7 fix playbook
  • references/lerobot_env_setup.md — detailed install notes
  • references/jetpack_compatibility_matrix.json — version gates
  • references/dependency_mapping_rules.md — install order rules

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free