[Backport release/3.0.0] Clear self-collision filter pairs before finalizing shadow Newton model (#7505)

[kellyguo11]

Description

Backports #7505 to release/3.0.0 by cherry-picking the merged commit 4b9ba22508895906bb856845153b6409b5e86b3c with -x provenance.

The automatic backport run encountered the expected visualization-builder conflict, but its inferred resolution dropped the release-only rename_builder_labels import and failed Ruff with F821.

This manual backport preserves the source change while retaining the release branch's API sequence: it strips collision filters and groups before composition and replication, calls the release signature of replicate_builder_mapping, and then calls rename_builder_labels separately. No runtime dependencies are added.

Field	Commit
Original merged change	4b9ba22508895906bb856845153b6409b5e86b3c
Release base used	355dc9ba107527d7baae7600e89229a89d4e4628
Proposed backport	64c6d483803bf706404236516d92c8ec77ab8211

Type of change

• Bug fix (non-breaking change which fixes an issue)

Release backport

• This PR already targets the active release branch.

Validation

• Repository backport candidate validation passed with all five source paths preserved and no extra paths changed.
• Full repository pre-commit suite passed against release/3.0.0, including changelog and Git LFS checks.
• uv run --no-project python -m compileall -q passed for all three modified Python files.
• git diff --check upstream/release/3.0.0...HEAD passed.
• Focused Newton regression tests are pending Linux CI because the repository lockfile does not support the local macOS/arm64 host. The original source PR completed 53 checks successfully.

Checklist

• I have read and understood the contribution guidelines
• I have run the available pre-commit checks
• Documentation changes are not needed for this backport
• My changes generate no new runtime warnings (pending Linux CI)
• The original regression test is preserved in the backport
• Changelog fragments are preserved for both touched packages
• The original contributor is already listed in CONTRIBUTORS.md

[kellyguo11]

run-ci

[greptile-apps]

Greptile Summary

This backport prevents PhysX-backed Newton shadow visualization models from retaining and replicating unused collision-filter metadata before finalization.

• Clears collision-filter pairs and resets collision groups in standalone and cloned visualization builders.
• Rebuilds cloned output from a compact destination builder while preserving the release branch’s replication and label-renaming sequence.
• Adds regression coverage for finalized shape counts and empty collision/contact-pair stores.
• Includes the corresponding Isaac Lab and Isaac Lab Newton changelog fragments.

Confidence Score: 5/5

The PR appears safe to merge with no concrete correctness or security defects identified.

The changed path removes collision metadata from a visualization-only shadow model before replication, while the regression test confirms that finalized standalone and cloned models retain expected shape counts and produce no collision-filter or contact pairs.

Important Files Changed

Filename	Overview
source/isaaclab_newton/isaaclab_newton/physics/visualization_builder.py	Clears unused collision metadata before composing and finalizing standalone or replicated shadow visualization models.
source/isaaclab_newton/test/cloner/test_rename_builder_labels.py	Extends builder test doubles for collision metadata and adds a real-builder regression covering standalone and cloned finalization.
source/isaaclab/test/sim/test_newton_manager_visualization_state.py	Updates the fake shadow builder to satisfy the new collision-state and composition interface.
source/isaaclab_newton/changelog.d/fix-shadow-model-collision-filter-oom.rst	Documents the fix for memory exhaustion caused by replicated unused collision metadata.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    A[USD stage] --> B[Global and source visualization builders]
    B --> C[Clear collision-filter pairs and reset groups]
    C --> D[Compact destination ModelBuilder]
    D --> E[Compose global geometry]
    E --> F[Replicate source builders by clone plan]
    F --> G[Rename labels]
    G --> H[Add shadow deformables]
    H --> I[Finalize visualization-only Newton model]

Loading

_{Reviews (1): Last reviewed commit: "Clear self-collision filter pairs before..." | Re-trigger Greptile}

[isaaclab-review-bot]

Isaac Lab Review Bot

Reviewed the backport’s collision-filter reset in both standalone and cloned Newton visualization-builder paths, including composition order, release-branch clone helper calls, updated test doubles, regression coverage, and package changelog fragments. No actionable defect is established by the supplied patch and context.

• Design and architecture: The clone path rebuilds into a fresh ModelBuilder after clearing collision metadata on the global and source builders, then preserves the existing global composition, replication, label-renaming, and shadow-deformable sequence. The standalone path clears the same metadata before returning. This is a focused change, though it remains coupled to Newton’s mutable builder internals and their composition behavior.
• API: The public function signature and return shape remain unchanged. The backport retains the release branch’s separate replicate_builder_mapping and rename_builder_labels calls, and no dependency or public API change is introduced.
• Implementation: The new attribute accesses are reflected in affected mocks and fakes, and the regression test finalizes real standalone and cloned builders while checking shape counts and empty collision/contact-pair stores. Clearing occurs after shadow deformables in the standalone path but before their addition in the clone path; the current evidence does not show that shadow deformables repopulate these stores, so this is only a future maintenance consideration rather than a finding.

No blocking issues. No inline issue met the actionable-evidence threshold; the assessment above records the review feedback.

Automated review; human maintainers own approval decisions.