Add sibling FANUC CRX SKUs (CRX-3iA, 5iA, 10iA, 10iA/LP, 20iA/L, 30iA)

[siddhss5]

Follow-on coverage for the FANUC CRX collaborative-robot line. CRX-10iA/L landed in #300 / v1.2.1 and confirmed the solver class (non-Pieper 6R via ikgeo.general_6r, 991 µs median, AE-3 leftvar=1, max cond(m_quad) 1.3e+03). The remaining six SKUs in the family share — by inspection of the official FANUC xacro macros — the same kinematic skeleton with scaled link parameters.

SKUs to add

Each row is one PR per the established arm-add cadence (#284 Z1, #285 PiPER, #289 Rizon 10, #300 CRX-10iA/L). One manifest entry plus a vendored URDF.

• CRX-3iA — 3 kg payload, ~540 mm reach
• CRX-5iA — 5 kg payload, ~994 mm reach
• CRX-10iA — 10 kg payload, ~1249 mm reach (the non-/L base model)
• CRX-10iA/LP — 10 kg payload, ~1418 mm reach (long + paint-class variant)
• CRX-20iA/L — 20 kg payload, ~1418 mm reach
• CRX-30iA — 30 kg payload, ~1888 mm reach (flagship)

URDF sources

The ros-industrial mirror only ships CRX-10iA (used in #300). For the other six, the canonical source is FANUC-CORPORATION/fanuc_description/fanuc_crx_description/urdf/, published as xacro macros with link-length parameters. Per-SKU files:

• crx3ia_urdf_macro.xacro
• crx5ia_urdf_macro.xacro
• crx10ia_urdf_macro.xacro
• crx10ia_lp_urdf_macro.xacro
• crx20ia_l_urdf_macro.xacro
• crx30ia_urdf_macro.xacro

Each needs xacro rendering before vendoring into tests/fixtures/. Standard xacro invocation against a top-level wrapper that instantiates the macro at prefix=\"\" is enough; no real reason to import the entire fanuc_crx_description ROS package.

Shared-skeleton hypothesis

All six are expected to route to ikgeo.general_6r with the same dispatch reasoning as CRX-10iA/L. To verify for each SKU before merging:

1. Dispatcher reasoning string matches "Closes the EAIK coverage gap" (tier-2 RR, not tier-0 Pieper).
2. AE-3 _cached_best_leftvar lands on _LINEARITY_JOINT = 1 (same as CRX-10iA/L). If a SKU picks a different leftvar, that's a signal the kinematic skeleton diverges meaningfully — investigate before adding.
3. cond(m_quad) on a representative non-singular pose stays under the 1e+04 gate (CRX-10iA/L max was 1.3e+03).

If all three hold across the six SKUs, the codegen path is fully reused. If not, file a sub-issue and treat that SKU as its own investigation.

Per-SKU PR template

Modeled on #300:

1. Render xacro → tests/fixtures/fanuc_crx<sku>.urdf.
2. uv run ssik add-arm tests/fixtures/fanuc_crx<sku>.urdf --base base_link --ee tool0 --name fanuc_crx<sku>.
3. Add [arms.fanuc_crx<sku>_ik] MANIFEST entry — copy CRX-10iA/L's block, swap fixture path + display_name + kinematic_class reach annotation + sample_q to an in-limits non-singular pose.
4. uv run python scripts/regen_artifacts.py → confirm tier-2 + leftvar=1 + cond gate.
5. 500-pose Hypothesis sweep — passes at fk_ceiling_fuzz=1e-4.
6. Bench via examples/04_compare_vs_eaik.py → populate [arms.fanuc_crx<sku>_ik.bench].
7. scripts/regen_docs.py → AUTOGEN tables (README, prebuilt, quickstart).
8. PR title: "Add FANUC CRX- prebuilt artifact (non-Pieper 6R)".

Bench-table comparison

Each SKU should produce a fresh "refuses (6R-Unknown Kinematic Class)" row from EAIK and a sub-ms ssik measurement — same shape as the CRX-10iA/L row already in the README EAIK table.

Out of scope

• Older FANUC cobots (CR-7iA, CR-35iA) — different product line that predates CRX. ros-industrial does carry them (fanuc_cr7ia_support, fanuc_cr35ia_support). Probably Pieper-class spherical-wrist; would route to tier-0, not the same wedge story. Worth a separate issue if and when they become relevant for coverage.
• A combined PR. Per the established cadence, one PR per SKU. Bundling muddies the artifact-snapshot review and makes any per-SKU conditioning surprise harder to bisect.
• Tuning the _cached_best_leftvar heuristic if any sibling picks a different leftvar — that's RR algebraic-precision gap on mixed-alpha synthetic geometries (e.g. MC Table I) #82-class work, surface it via a sub-issue with the offending SKU and the cond(m_quad) measurements.

Acceptance

• Each of the six SKUs lands as its own merged PR with a green bulletproof sweep
• All six route to ikgeo.general_6r tier-2 with _LINEARITY_JOINT = 1 (or a documented investigation if any diverge)
• README EAIK comparison table grows by six rows, each "refuses" / sub-ms
• from ssik.prebuilt import fanuc_crx<sku>_ik works for every variant