Hydra SDG rendering fix

[pbarejko]

Description

This MR is a continuation of #5375 without omni.ui as a dependency, only minimal dependencies are added and required code was pushed to the render backend that requires it.

Fixes # (issue)

cudaErrorIllegalAddress

./isaaclab.sh -p -m  pytest source/isaaclab/test/sensors/test_multi_tiled_camera.py -v

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

[isaaclab-review-bot]

🤖 Isaac Lab Review Bot

Summary

This PR fixes a CUDA memory error (cudaErrorIllegalAddress) in headless SDG rendering by ensuring the RTX Hydra engine is properly attached to the USD context before Replicator render products are created. The fix adds a new dependency (omni.kit.hydra_texture) to the headless rendering kit file and introduces an ensure_rtx_hydra_engine_attached() utility function called during renderer initialization.

Architecture Impact

The change is well-contained within the Isaac RTX renderer subsystem:

• IsaacRtxRenderer.__init__() now calls ensure_rtx_hydra_engine_attached() at construction time
• Any code path that creates an IsaacRtxRenderer (primarily TiledCamera sensors) will trigger this initialization
• The function is idempotent, so GUI workflows with omni.kit.viewport.window already attached will see no change
• No breaking API changes

Implementation Verdict

Minor fixes needed — The core logic is sound but there's an inconsistency in error handling and a potential issue with import timing.

Test Coverage

No tests added. This is a bug fix for a headless rendering crash that would be difficult to unit test (requires specific GPU behavior). However, the PR description doesn't reference the original issue number or provide reproduction steps, making it hard to verify the fix works. The checklist item for tests is unchecked.

CI Status

No CI checks available yet — cannot verify this doesn't regress existing rendering tests.

Findings

🔴 Critical: source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer_utils.py:14 — Top-level import omni.usd will fail in non-Kit environments

The module now unconditionally imports omni.usd at the top level (line 14). This will cause ImportError when this module is imported in environments without Omniverse (e.g., some unit tests, pure-Python tooling). The other Omniverse imports in this file (omni.kit.app) are done inside functions specifically to avoid this issue.

# Line 14 - problematic top-level import
import omni.usd

Should be moved inside ensure_rtx_hydra_engine_attached() or guarded:

def ensure_rtx_hydra_engine_attached() -> None:
    try:
        import omni.usd
        ctx = omni.usd.get_context()
        # ...

🟡 Warning: source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer_utils.py:91 — Docstring says "warnings" but implementation uses logger.error

The docstring at line 78 states "Failures are logged as warnings and do not propagate" but line 91 uses logger.error(). This inconsistency could mislead readers and affects log filtering in production:

# Line 78 docstring says:
# "Failures are logged as warnings and do not propagate"

# Line 91 uses error:
except Exception as e:  # noqa: BLE001
    logger.error("RTX Hydra engine attach failed: %s", e)

Either update the docstring to say "errors" or change to logger.warning() as documented.

🟡 Warning: source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer.py:75 — ensure_rtx_hydra_engine_attached() called before stage is available

IsaacRtxRenderer.__init__() calls ensure_rtx_hydra_engine_attached() which queries omni.usd.get_context(). If the renderer is constructed before a Kit app is fully initialized or a USD context exists, get_context() returns None and the function silently returns without attaching the engine. Later, create_render_data() will still fail.

Consider whether this should be called in create_render_data() instead (line 78), where we know the stage/context must exist since the sensor view is being accessed:

def create_render_data(self, sensor: SensorBase) -> IsaacRtxRenderData:
    ensure_rtx_hydra_engine_attached()  # Context guaranteed to exist here
    import omni.replicator.core as rep
    # ...

🔵 Improvement: apps/isaaclab.python.headless.rendering.kit:25 — Extra blank line after dependency

Minor: There's a double blank line after the new omni.kit.hydra_texture dependency which is inconsistent with the rest of the file's formatting:

+"omni.kit.hydra_texture" = {}  # required for headless rendering and rtx instantiation
+
+
 [settings.isaaclab]

🔵 Improvement: source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer_utils.py:60-92 — Function placement breaks logical grouping

The new ensure_rtx_hydra_engine_attached() function is inserted between _wait_for_streaming_complete() (a private helper) and ensure_isaac_rtx_render_update() (a public function). For better code organization, consider placing it either:

• Before all the streaming-related private helpers (after the module-level constants), or
• After ensure_isaac_rtx_render_update() since they're both public API functions

This is minor but affects readability when navigating the module.

[greptile-apps]

Greptile Summary

This PR fixes headless RTX rendering by ensuring the RTX Hydra engine is attached to the USD context at IsaacRtxRenderer construction time, replicating the activation step normally performed by ViewportWindow (which is intentionally omitted in headless .kit files). The omni.kit.hydra_texture extension is also added to isaaclab.python.headless.rendering.kit to make RTX engine instantiation possible without the UI viewport stack.

Confidence Score: 5/5

Safe to merge; only P2 findings remain (log level mismatch in docstring, redundant local import).

The fix is minimal and well-scoped: the new helper is idempotent, errors are swallowed rather than propagated, and the .kit change adds only one targeted extension. Both remaining findings are style/documentation level (P2) and do not affect correctness or runtime behaviour.

No files require special attention.

Important Files Changed

Filename	Overview
source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer_utils.py	Adds ensure_rtx_hydra_engine_attached to idempotently bind the RTX Hydra engine; minor docstring/log-level mismatch and a redundant local omni.usd import remain.
source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer.py	Calls ensure_rtx_hydra_engine_attached() in __init__; straightforward one-liner addition, no issues.
apps/isaaclab.python.headless.rendering.kit	Adds omni.kit.hydra_texture extension to enable headless RTX Hydra engine instantiation; well-commented and minimal.

Sequence Diagram

sequenceDiagram
    participant App as Headless Kit App
    participant Renderer as IsaacRtxRenderer.__init__
    participant Utils as isaac_rtx_renderer_utils
    participant USD as omni.usd.UsdContext

    App->>Renderer: instantiate(cfg)
    Renderer->>Utils: ensure_rtx_hydra_engine_attached()
    Utils->>USD: get_context()
    USD-->>Utils: ctx
    Utils->>USD: ctx.get_attached_hydra_engine_names()
    USD-->>Utils: engine names
    alt "rtx" not already attached
        Utils->>USD: create_hydra_engine("rtx", ctx)
        USD-->>Utils: engine attached
    else already attached
        Utils-->>Renderer: no-op
    end
    Renderer-->>App: ready

Loading

Comments Outside Diff (1)

1. source/isaaclab_physx/isaaclab_physx/renderers/isaac_rtx_renderer_utils.py, line 29-36 (link)

   Redundant local import omni.usd inside _get_stage_streaming_busy

   omni.usd is now imported at module level (line 14), so the import omni.usd inside _get_stage_streaming_busy is redundant. The local import was presumably there to keep the module importable in non-Kit environments, but that guard is now moot. The redundant import can be removed.

_{Reviews (1): Last reviewed commit: "Hydra SDG rendering fix" | Re-trigger Greptile}