[hougantc-nvda]

Description

Review stack · 4/5: #8787 → #8788 → #8789 → #8790 → #8791

Review only this PR’s changes. The PRs target main; the comparison isolates this layer from its predecessors.

Add optional robot-camera panels (PiP) and a desktop mirror to OVXR teleoperation. Both consume the session's completed images; headset-only operation still needs no desktop window.

• isaaclab_teleop: selects camera feeds and supplies configured CameraPanelConsumer and OwnedDesktopMirror instances through OvxrSessionBinding.frame_consumers.
• isaaclab_ov: owns camera allocation, scene publication and native resource lifetime through the existing render lease.
• Isaac Capture: owns panel presentation and the mirror (#1295, #1296).

GR1T2 and G1 locomanipulation expose a robot-view camera; fixed-base G1 adds two wrist feeds. Its camera/image-observation defaults also affect non-XR use; the documented camera-disable option remains available. Kit Scene UI retains its feed settings. SO101 has no declared PiP camera.

The adapter uses a tuple of auxiliary render-product paths for both one and multiple cameras. The mirror uses OVUI's default Vulkan backend. Neither Lab nor Isaac Capture changes OMNIUI_BACKEND; conflicting initialized backends are rejected. Neither optional consumer is created when disabled.

Panels use the existing renderer and a head-locked layout. The mirror shows completed eye or camera images; eye images exclude the separate PiP layers. Closing the mirror requests orderly application shutdown.

PiP/mirror require the matching CUDA 13 package selection. The mirror also requires Python 3.12+, Capture's desktop-mirror extra and a usable GPU-backed Vulkan display. See camera panels and desktop mirror for selection and display constraints.

Dependencies: compatible Isaac Capture OVXR packages and the provider upgrade in #8858. Native-provider defaults are unchanged.

Type of change

• New feature
• Bug fix
• Documentation update

Unsupported panel placements are rejected at the OVXR resolver boundary; the separate Kit Scene UI placements remain supported.

Testing

Results cover the cumulative five-PR stack on main 6ef40d19a6f8d3770f86b5c4df3f8ccfb6a25f95. Kitless checks use Isaac Capture ae48621f1cc13054c2246210d8f92d30e27fe354 built from source using Python 3.12; Kit checks use Lab's upstream locked Capture selection. They are not independent full-repository results for each PR.

• 1,039 cases across 51 selected Lab/Kit files passed, with no remaining failures or skips. Coverage includes ordinary rendering, native cross-GPU publication, graph capture, startup interruption, failed-drain ownership, retargeting, task selectors, and 140 Kit-selected cases including launch/replay and Scene UI.
• Five Omen IWER scenes passed stereo/task-object inspection, meaningful input, Reset and normal mirror-close shutdown: GR1T2 CPU and split-GPU Newton, G1 locomanipulation CPU, fixed-base G1 Newton and SO101 CPU. Humanoid PiP was checked; SO101 uses mirror only. Reconnect/resumed input was checked for G1 and GR1T2 Newton.
• The ordinary Kit headless Franka XR path passed stereo scene, motion, Reset and reconnect checks. It ended by deliberate owner interrupt after client disconnect; its detached CloudXR service was explicitly stopped afterward.
• Formatting, changelog checks, warning-as-error Sphinx and the kitless dependency/selected-extra checks passed. The installed Capture wheel and staged package files were verified identical.

Limits: these are synthetic IWER checks, not physical-headset acceptance or full task completion. Cold renderer attach, intermittent GR1T2 first-Play recovery and SO101 cube retention remain open. Kit's upstream dependency selection reports nine metadata inconsistencies, including NumPy/Numba; passing the selected Kit tests does not qualify that entire dependency graph. Full-repository, hosted-CI and arm64 coverage are not claimed.

See #8791 for the matched video-rate measurements.

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with uv run isaaclab --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have validated the changed behavior with focused coverage and applied the test-audit criteria to any test changes
• I have added a changelog fragment under source/<pkg>/changelog.d/ for every touched package (do not edit CHANGELOG.rst or bump package versions — CI handles that)
• I have added my name to the CONTRIBUTORS.md or my name already exists there

The warnings checkbox remains open: dependency warnings and the separate Kit metadata audit are unresolved.