thanhndv212/figaroh-plus/figaroh-setup-env

FIGAROH Environment Setup

_Arguments: none needed. Optionally pass a workspace root if it is not $FIGAROH_WS._

Paths. $FIGAROH_WS is the workspace root — the directory holding the

figaroh and figaroh-examples repos side by side. Set it once per shell:

export FIGAROH_WS="$(cd /path/to/your/workspace && pwd)".

When to Use

• First contact with this workspace, or a new machine.
• ModuleNotFoundError: figaroh, No module named 'cyipopt', mesh/package-dir errors.
• Library edits in figaroh/src/ seem to have no effect (→ shadowed by a

site-packages copy).

• Before invoking figaroh-setup-calibration, -identification, -optimal, or

-new-robot.

Step 1 — Run the doctor (always start here)

bash "$FIGAROH_WS"/figaroh/skills/figaroh-setup-env/scripts/doctor.sh

Read-only; mutates nothing. Exit 0 = ready, 1 = blocked. It checks workspace layout, the conda env, import figaroh and whether it resolves to the local source tree, core deps, each optional dep (with the single task it blocks), and that the UR10 reference example is intact. Every FAIL prints its own fix: line — apply those, then re-run until exit 0. Do not proceed to a task skill while it exits 1.

--ws <root> and --env <name> override the defaults.

Step 2 — The environment contract

`figaroh-dev` is the only supported environment. Python 3.12, conda-forge, defined by figaroh/environment.yml:

conda env create -f $FIGAROH_WS/figaroh/environment.yml
conda activate figaroh-dev

environment.yml already does pip install -e ., so creating the env gives you the editable install of the local library. To repair or refresh it:

conda run -n figaroh-dev pip install -e $FIGAROH_WS/figaroh
conda env update -n figaroh-dev -f $FIGAROH_WS/figaroh/environment.yml

Two ways to run commands — both fine, be consistent within a session:

conda activate figaroh-dev && python calibration.py     # interactive
conda run -n figaroh-dev python calibration.py          # scripted / CI

Trap: figaroh-examples/environment.yml declares an env named figaroh-examples. It is stale and unused — ignore it. The real env lives in the sibling figaroh/ repo.

Step 3 — Optional dependencies, and exactly what each one gates

Install these only when the task needs them; none of them block calibration or identification.

Step 4 — Confirm with the reference example

The install is not proven until a real task runs. Use UR10 as the bisect point:

conda activate figaroh-dev
cd $FIGAROH_WS/figaroh-examples/examples/ur10
python calibration.py --calibrate-only --no-plot

Expect post-calibration position/orientation RMSE and a saved data/calibration/calibration_results_<timestamp>.npz. If this passes, the environment is not the problem — route any later failure to the user's model, config, or data.

Broader gate, when you want the whole suite:

cd $FIGAROH_WS/figaroh-examples
python validate.py --quick      # skips IPOPT-heavy scripts
python validate.py              # full: pytest + every example script

Known-benign noise — do not chase these

• `RuntimeWarning: to-Python converter for std::__1::shared_ptr<ndcurves::curve_abc…>

already registered; second conversion method ignored on import figaroh. Harmless Boost.Python double-registration from ndcurves/pinocchio`.

• conda run appends a trailing blank line to captured stdout — ... | tail -1

silently yields the empty line instead of your value. Grep for a sentinel instead (this is why doctor.sh has a probe() helper).

• IPOPT timeouts in validate.py are reported separately from failures and are

expected, not bugs.

• Pre-existing test failures in figaroh/: test_print_collision_pairs*,

test_verbose_output (logging-related). Not regressions.

Failure → cause → fix

Next

Environment green → return to figaroh-start's Dispatch table, or go straight to figaroh-setup-calibration, figaroh-setup-identification, figaroh-setup-optimal, or figaroh-setup-new-robot.