IsaacLab 3.0 Locomanipulation SDG support

[jaybdub]

Description

This PR (implemented by Yuchen Deng / Chirag Majithia) handles quat conversion in the locomanipulation SDG workflow, enabling support for Isaac Lab 3.0. This also updates the documentation to add the --visualizer kit command required to run the workflow.

Locomanipulation SDG was broken with Isaac Lab 3.0+ due to CUDA device semantics / quaternion conversion. This is fixed by this PR.

Type of change

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

Screenshots

Please attach before and after screenshots of the change if applicable.

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 Isaac Lab 3.0 compatibility issues in the locomanipulation SDG workflow by addressing CUDA device placement and quaternion convention changes.

Key changes:

• Fixed device placement by adding device= parameters to torch tensor creation throughout the codebase
• Added .to(self.device) calls when cloning state tensors in interactive_scene.py to ensure tensors are on correct device
• Added warp.to_torch() conversion for body_link_state_w in scene_utils.py to handle warp tensor types properly
• Implemented quaternion conversion from legacy WXYZ to XYZW format for action data loaded from older datasets
• Updated hardcoded quaternion literals from WXYZ (e.g., [1.0, 0.0, 0.0, 0.0]) to XYZW format (e.g., [0.0, 0.0, 0.0, 1.0])
• Added --visualizer kit flag to documentation example

Confidence Score: 5/5

• Safe to merge - addresses critical bug fixes for Isaac Lab 3.0 compatibility with consistent patterns
• The changes are well-scoped bug fixes that systematically address device placement and quaternion convention issues. The patterns are consistent throughout (adding device parameters, .to(device) calls, and quaternion format updates). The warp-to-torch conversion is necessary for compatibility. The changes align with the PR description and fix the stated issues.
• No files require special attention

Important Files Changed

Filename	Overview
scripts/imitation_learning/locomanipulation_sdg/generate_data.py	Added device parameter to tensor creation and quaternion conversion for legacy dataset support
source/isaaclab/isaaclab/scene/interactive_scene.py	Added .to(self.device) calls when cloning state tensors to ensure proper device placement
source/isaaclab_mimic/isaaclab_mimic/locomanipulation_sdg/envs/g1_locomanipulation_sdg_env.py	Fixed quaternion convention from WXYZ to XYZW and added device parameter to fixture_pose tensor
source/isaaclab_mimic/isaaclab_mimic/locomanipulation_sdg/scene_utils.py	Added warp to torch conversion for body_link_state_w and device handling for relative_pose

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Dataset Loading] --> B{Legacy Quat Format?}
    B -->|Yes| C[Convert WXYZ to XYZW]
    B -->|No| D[Use XYZW directly]
    C --> E[Episode Data]
    D --> E
    E --> F[Scene Reset]
    F --> G[Clone State Tensors]
    G --> H[Apply .to device]
    H --> I[Warp to Torch Conversion]
    I --> J[Transform Operations]
    J --> K[Device-aware Tensor Creation]
    K --> L[Locomanipulation SDG]

Loading

_{Last reviewed commit: 812a3fb}

[greptile-apps]

_{7 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[peterd-NV]

Following the slack on thread, can the dataset download links be updated to content-staging and 6.0

[ShevaDavid]

@jaybdub

Could you please open access to ONNX in this repository: https://github.com/NVIDIA-AI-IOT/sdg_pallet_model

People have requested access, but nobody has received it according to the GitHub issues.

Thanks in advance!