Fixes CPU pipeline stage initialization in PhysxManager

[kellyguo11]

Description

In recent PhysicsManager refactoring, behavior in the PhysxManager changed to call attatch_stage by default for initialization.
This triggered a side effect in CPU pipeline that caused objects to miss contacts and collisions.

It appears that attach stage to PhysX before loading/starting is only needed for GPU pipeline. So in this change, we are guarding this call to run on GPU pipeline only and avoid it on CPU pipeline to avoid issues in collisions.

Type of change

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

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have updated the changelog and the corresponding version in the extension's config/extension.toml file
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[greptile-apps]

Greptile Summary

This PR fixes a CPU collision detection bug introduced during the PhysicsManager refactoring (#4564). The recent refactoring added an unconditional call to attach_stage() in the warmup process, which was intended to fix GPU pipeline initialization. However, this caused a double-initialization issue on the CPU pipeline where attach_stage() combined with force_load_physics_from_usd() corrupts the CPU broadphase (MBP) collision setup, causing objects to fall through surfaces non-deterministically.

Key changes:

• Added device detection to check if GPU pipeline is in use (is_gpu = "cuda" in PhysicsManager.get_device())
• Wrapped the attach_stage() call in a conditional that only executes for GPU pipeline
• Added detailed comments explaining why this separation is necessary and the root cause of the CPU collision bug

The fix aligns with the original behavior where the old SimulationManager never called attach_stage() explicitly on CPU pipeline.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The fix is well-targeted and addresses a specific regression introduced in a recent refactoring. The change is minimal (7 lines modified), follows the same device detection pattern used elsewhere in the codebase (line 448), and includes clear explanatory comments. The fix restores the original CPU pipeline behavior while preserving the GPU pipeline fix.
• No files require special attention

Important Files Changed

Filename	Overview
source/isaaclab_physx/isaaclab_physx/physics/physx_manager.py	Conditionally calls attach_stage() only for GPU pipeline to fix CPU collision detection issue caused by double-initialization of broadphase collision setup

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[_warmup_and_create_views called] --> B{Check device type}
    B -->|GPU pipeline<br/>cuda in device| C[attach_stage]
    B -->|CPU pipeline<br/>cpu device| D[Skip attach_stage]
    C --> E[force_load_physics_from_usd]
    D --> E
    E --> F[start_simulation]
    F --> G[update_simulation]
    G --> H[Create SimulationView]
    H --> I[Physics Ready]
    
    C -.->|Prevents| J[GPU buffer allocation issues]
    D -.->|Prevents| K[CPU broadphase MBP<br/>double-initialization]
    K -.->|Would cause| L[Objects fall through surfaces]

Loading

_{Last reviewed commit: 90894d0}

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}