Fix mesh-convex margin candidate query by adenzler-nvidia · Pull Request #2916 · newton-physics/newton

adenzler-nvidia · 2026-05-20T11:59:36Z

Description

Fix mesh-vs-convex and heightfield-vs-convex midphase candidate generation to query candidates using the full margin + gap contact envelope, not just gap.

This keeps the mesh BVH query and heightfield grid-cell lookup consistent with broad-phase AABB expansion and final contact acceptance. Without this, valid margin-shell contacts can be culled before narrow phase processing.

Checklist

New or existing tests cover these changes
The documentation is up to date with these changes
CHANGELOG.md has been updated (if user-facing change)

Test plan

uv run --extra dev -m newton.tests -k Midphase
uv run --extra dev ruff check newton/_src/geometry/collision_core.py newton/_src/geometry/narrow_phase.py newton/_src/utils/heightfield.py newton/tests/test_collision_pipeline.py
git diff --check

Bug fix

Steps to reproduce:

Create a mesh-vs-sphere or heightfield-vs-sphere scene with per-shape margin=0.02, gap=0.005.
Place the sphere with true surface separation 0.03.
Run the Newton collision pipeline.
Without this PR, no contact is generated even though 0.03 <= margin_a + margin_b + gap_a + gap_b.

The heightfield regression covers the lateral edge case where grid-cell candidate lookup can cull the contact before narrow phase. Vertical margin-shell contacts above the heightfield already reached narrow phase because the grid lookup is driven by XY cell overlap.

Minimal reproduction:

# Covered by:
# uv run --extra dev -m newton.tests -k Midphase

Summary by CodeRabbit

Bug Fixes
- Fixed missing mesh–convex and heightfield–convex contacts when shapes are separated by margin but still lie within the contact envelope; contact thresholds now include per-shape margin.
Tests
- Added midphase tests verifying contacts are produced for mesh/heightfield vs convex cases when margins/gaps match.

Use the full margin plus gap contact envelope when querying mesh triangle candidates for convex contacts. This keeps the mesh midphase consistent with broad phase expansion and final contact acceptance, preventing valid margin-shell contacts from being culled before narrow phase processing. Add a CUDA regression test where a sphere is separated from a mesh by more than gap but less than margin plus gap.

coderabbitai · 2026-05-20T11:59:51Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Pro

Run ID: 9ffe0099-b96e-4317-a91d-b3437c042f18

📥 Commits

Reviewing files that changed from the base of the PR and between c56385b and f38a747.

📒 Files selected for processing (4)

CHANGELOG.md
newton/_src/geometry/narrow_phase.py
newton/_src/utils/heightfield.py
newton/tests/test_collision_pipeline.py

📝 Walkthrough

Walkthrough

This PR unifies gap and per-shape margins into a single contact_threshold used by mesh- and heightfield-midphase queries; the midphase API and BVH preprocessing were updated, narrow-phase triangle culling supplies the combined threshold, and tests plus a changelog entry were added.

Changes

Mesh–Convex Margin-Shell Contact Generation

Layer / File(s)	Summary
Core midphase API: contact_threshold parameter and BVH logic `newton/_src/geometry/collision_core.py`	`mesh_vs_convex_midphase` signature changed to accept `contact_threshold` (replacing `rigid_gap`); docstring and BVH query preprocessing now compute per-axis margin vectors from the combined threshold, accounting for non-uniform mesh scale.
Triangle overlap culling: gap + per-shape margins threshold `newton/_src/geometry/narrow_phase.py`	narrow-phase kernel computes `contact_threshold = gap_non_mesh + gap_mesh + margin_non_mesh + margin_mesh` (margins from `shape_data[...,3]`) and passes it to `mesh_vs_convex_midphase`; heightfield midphase calls now include `shape_data`.
Heightfield midphase: accept shape_data and inflate AABB by margin `newton/_src/utils/heightfield.py`	`heightfield_vs_convex_midphase` signature updated to accept `shape_data`; midphase AABB inflation uses `contact_threshold = gap_sum + margin_sum` with per-shape margins from `shape_data`.
Tests and module imports for margin-shell validation `newton/tests/test_collision_pipeline.py`	Module-level imports added for `create_mesh_terrain` and `ShapeFlags`. New CUDA tests `TestMeshConvexMidphase` and `TestHeightfieldConvexMidphase` assert rigid contacts occur when shapes are separated by margin but within the contact envelope; tests registered for CUDA devices.
Fix documentation in changelog `CHANGELOG.md`	Unreleased “Fixed” entry added documenting the mesh–convex and heightfield–convex contact fix.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

newton-physics/newton#2745: Both PRs modify mesh_vs_convex_midphase in newton/_src/geometry/collision_core.py—Fix bug with nonuniform scaling in mesh vs convex collisions #2745 changes the BVH AABB query to unscaled mesh-local space and introduces per-axis gap/margin handling, while the main PR reworks the threshold plumbing (rigid_gap → contact_threshold) and uses it to compute the same kind of anisotropic per-axis contact expansion.
newton-physics/newton#2527: Main PR updates mesh_vs_convex_midphase’s threshold semantics/signature and how per-axis margins are folded into contact_threshold, while retrieved PR also modifies mesh_vs_convex_midphase’s internal tiled BVH triangle query draining loop control flow—so both change the same midphase function at the code level.
newton-physics/newton#2651: Both PRs modify the heightfield-vs-convex midphase pipeline—newton/_src/utils/heightfield.py::heightfield_vs_convex_midphase and its narrow_phase.py call sites—by changing midphase culling/overlap threshold inputs (one via convex-local AABB, the other by adding per-shape margin-based contact_threshold/shape_data).

Suggested reviewers

eric-heiden
Milad-Rakhsha-NV

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 71.43% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly and clearly summarizes the main change: fixing the mesh-convex margin candidate query by incorporating margin into the midphase BVH query.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

codecov · 2026-05-20T12:20:09Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!

nvtw · 2026-05-20T14:29:34Z

I think the bugfix is correct. But can you double check - it's very likely that the heightfield code has the same problem.

Use the full margin plus gap contact envelope when querying heightfield cells for convex contacts. This keeps heightfield midphase candidate generation consistent with broad phase expansion and final contact acceptance. Add a CUDA regression covering a sphere just outside a heightfield lateral edge, where the surface separation is larger than gap but still inside the margin shell.

adenzler-nvidia requested a review from nvtw May 20, 2026 12:05

nvtw approved these changes May 20, 2026

View reviewed changes

adenzler-nvidia enabled auto-merge May 20, 2026 15:27

adenzler-nvidia added this pull request to the merge queue May 20, 2026

Merged via the queue into newton-physics:main with commit bf9a8f1 May 20, 2026
25 checks passed

jcarius-nv added this to the 1.2.1 Patch Release milestone May 21, 2026

jcarius-nv mentioned this pull request May 26, 2026

Cherry-picks for 1.2.1 patch #2958

Merged

adenzler-nvidia added a commit that referenced this pull request May 27, 2026

Fix mesh-convex margin candidate query (#2916)

89fe728

adenzler-nvidia mentioned this pull request Jun 1, 2026

Cherry-pick 1.2.1 PRs onto release-1.2 #3021

Merged

jkuzmeski pushed a commit to jkuzmeski/newton that referenced this pull request Jun 2, 2026

Fix mesh-convex margin candidate query (newton-physics#2916)

7d3da87

This was referenced Jun 3, 2026

Cherry-pick 1.2.1 PRs onto release-1.2 #3047

Merged

Merge back 1.2.1 changelog and reconcile 1.2.0 #3080

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix mesh-convex margin candidate query#2916

Fix mesh-convex margin candidate query#2916
adenzler-nvidia merged 2 commits into
newton-physics:mainfrom
adenzler-nvidia:adenzler/fix-mesh-convex-margin-shell

adenzler-nvidia commented May 20, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented May 20, 2026 •

edited

Loading

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Suggested reviewers

❌ Failed checks (1 warning)

Uh oh!

codecov Bot commented May 20, 2026

Uh oh!

nvtw commented May 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

adenzler-nvidia commented May 20, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Checklist

Test plan

Bug fix

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Suggested reviewers

❌ Failed checks (1 warning)

Uh oh!

codecov Bot commented May 20, 2026

Codecov Report

Uh oh!

nvtw commented May 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

adenzler-nvidia commented May 20, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented May 20, 2026 •

edited

Loading