Skip to content

Next Gen STEP AP242 Native-ish BREP Geometric Modeling Kernel

8.7 KB
Raw Blame

Temporary fitting from Concept geometry

Experimental. Garment work is research, not part of the supported product. Its APIs, syntax and results may change or be withdrawn. See experimental modules.

Follow-up: the armhole investigation combines these guides with T-pose sewing and A-pose wearing. Final strain improves to 9.23%, while startup checkpoint strain remains an explicit failed gate.

A fitting plane belongs to an ordinary Concept Struct. It is construction geometry, erased from the garment mesh and material scene. A Fitting binding selects named material boundaries and compiles the plane to a bounded, explicit solver command. It does not add permanent pins, alter the flat pattern, weld seams or change the cloth content key.

Concept Struct fitting {
  FigureFit body {
    figure: Anatolia;
    pose: Rest;
    chestHeight: 1280mm;
    hipHeight: 960mm;
  }
  collarFloor: Plane {
    origin: [fitting.body.centerX, fitting.body.centerY,
      fitting.body.shoulderHeight + 45mm]
    normal: [0, 0, 1]
  }
}

// Inside the Garment, alongside Panels, stitch Interfaces and Drape:
Fitting collar {
  on: fitting.collarFloor;
  edges: [front.edges.collar, back.edges.collar];
  compliance: 0.00000001;
  releaseAfter: 20;
}

The body measurement pass and existing Concept plane resolver own construction geometry. Coordinates use the existing compile-time scalar evaluator, including dimensional arithmetic; origins are in Firmament millimetres and compile to metres. Normals must be finite unit vectors. Selected vertices obey dot(normal, position) >= offset, with unilateral XPBD multipliers reset per substep and accumulated across iterations. Compliance is an authoring parameter, not a calibrated textile modulus. Body contact runs after guide projection so a conflicting guide cannot override body clearance; its residual stays visible.

releaseAfter counts gravity-driven settling frames after the 30 sewing frames. Zero supports sewing only. Every guide requires an explicit integer release frame, from 0 through 585. There are at most 32 guides, 64 edge references per binding and 32,768 total selected vertex entries. A drape budget leaving fewer than 15 frames after the final release fails before stepping with garment-fitting-release-budget. Early settling cannot succeed while a guide is active or during those first 15 unsupported frames. The existing 15% strain, 10mm seam gap and 5mm sampled penetration limits stay unchanged.

CLI evidence retains the Concept reference, compiled vertex IDs, numeric plane, compliance, release frame and per-sample active status/residual. It writes the last supported snapshot as guided-garment.obj and guided-body.obj; the usual garment.obj is the unsupported final snapshot. Neither OBJ nor USDA exports the construction plane. A build success establishes compilation only, not a supported or unsupported body fit.

This uses the existing CPU cloth solver. The Vulkan solver and shared GPU header builder reject guides with AUR-CLOTH-005 before GPU work; they must not silently drop constraints. There is no new GPU fitting, friction, CCD or self-contact qualification.

Analytic witness

fixtures/Canonical/Garment/temporary-fitting.firmament uses the same mechanism on a small horizontal sheet. Real draping supports its boundaries at one metre, releases them after five settling frames and shows downward motion during 15 unsupported frames with exact snapshot replay. Tests also check XPBD compliance without artificial iteration stiffening, contact conflicts, unchanged material topology, invalid bindings, unit errors and atomic invalid-command rejection.

Anatolia experiment

The known raised GarmentCode dress is reused without changing its flat cut, material, stitches, initial placement or body. Only measured Concept geometry and the fitting binding are added. The retained source and CLI evidence live under artifacts/local/garment/fitting-round/.

Both trials use the same cloth key 42C204F0BE495216B6CB3C2C3087EC8C786135D3EE71FE2D465BCFB4D468BA8B, 611 vertices, 992 triangles and zero permanent pins. The body is antonia.gameplay-candidate.v1, artifact SHA-256 0a0ad03e86447fc56448940c7aa56b54ea6683e8a874f833e01fd56a65a2d0e5. Only the four named front/back collar boundaries are selected. Each run uses 30 sewing frames, 60 settling frames, eight substeps and 24 iterations. Guides release after settling frame 20, leaving 40 unsupported frames.

The matching no-guide control (control-60/) uses the same 30+60 frame budget, cloth, body and solver settings. It slips below the shoulders, with 111.24% peak strain and 5.618mm seam gap. All three runs have exact replay, finite metrics and no guides active in their final snapshot. Comparison values and matching cloth/body hashes are retained in comparison.json:

Trial Final strain Seam gap Sampled penetration Appearance
No guide 111.24% 5.618mm 0.0000964mm Slips below shoulders
Low, soft guide 48.50% 5.658mm 0.858mm Still strapless
Higher, firmer guide 87.12% 3.144mm 0.00872mm Remains around shoulders

None passes the strain gate. The higher guide improves retention without permanent support; the low guide demonstrates why optimizing peak strain in isolation would choose the wrong visual result.

The first trial (anatolia-guided.firmament, guided-60/) sets the plane 30mm below the measured shoulder height, at 1.3848487m, with compliance 1e-6. Its final strain is 48.50%, seam gap 5.658mm and sampled penetration 0.858mm. It still looks strapless while supported and after release. A lower strain alone is not a better fit.

The second (anatolia-guided-high.firmament, high-60/) uses shoulder height plus 45mm, at 1.4598488m, with compliance 1e-8. Source SHA-256 is 9b912b0e7e30de828d854d44a111c83bda95daa43a4977c756af6767384f014a. The collar remains around the shoulders after release; the guide violation is zero in the last supported sample. Final seam gap is 3.144mm and sampled penetration is 0.00872mm, with exact replay and all guides removed. However, peak strain is 84.46% already at the last supported sample and 87.12% after release: the dress still fails the unchanged 15% strain limit.

The worst final constraint is the right-back armhole diagonal 336/337, rest length 14.275mm. Its material coordinates are retained in stretchOutliers. This isolates the next investigation to sewing/contact/material strain at the armhole rather than simply keeping the collar from falling. It does not prove that pattern sizing alone, contact alone or a particular textile parameter is the root cause. No acceptance threshold or pattern dimension was adjusted to make the trial pass.

dotnet Aetheris.CLI/bin/Release/net10.0/aetheris.dll garment drape `
  artifacts/local/garment/fitting-round/anatolia-guided-high.firmament `
  --body artifacts/local/humanoid-production/antonia.gameplay-body.json `
  --ticks 60 --substeps 8 --iterations 24 `
  --out-dir artifacts/local/garment/fitting-round/high-60 --json

& 'C:/Program Files/Blender Foundation/Blender 5.2/blender.exe' `
  --background --python scripts/render-garment.py -- `
  --garment fitting-round-high-supported `
  --input-dir artifacts/local/garment/fitting-round/high-60 --frame guided --samples 12

Render the unsupported result by omitting --frame guided and choosing a different --garment output name. All three views use the actual solver OBJ meshes, with no Blender fitting, subdivision or cloth simulation. Each render records its input hashes. The first trial preceded the dedicated checkpoint-body export; its Rest body OBJ was copied to guided-body.obj for presentation only. The second trial exports both checkpoint meshes directly through the CLI.

Validation

The Release Aetheris solution build passed with the existing nullable/obsolete and WebAssembly SQLite interop warnings. Its complete idle, serial solution lane passed 4,517 tests, including 1,195 kernel and 2,108 Firmament tests. The 15 fitting tests also passed after making the scalar-expression test input independent of checkout line endings. The generated STEP test output was restored after validation. Logs are retained as artifacts/local/garment/fitting-round/{build-final,tests-full,tests-focused-final}.log.

The Release Aurelian consumer solution built with zero warnings and errors. Its full serial test lane passed 1,218 tests with the installed Naga selected through AURELIAN_NAGA. Copeland logs are artifacts/aurelian-beacon3d/garment-fitting-{build,tests}.log. This establishes consumer compatibility and explicit GPU refusal, not GPU fitting support or a new native garment animation qualification. No dependency was downloaded and no third-party solver code was adopted.