Updates CI test setup and fixes camera tests

[kellyguo11]

Description

• Fixes some issues in unit tests
• Fixes rendering in recent visualizer change, which broke the render update
• Moves curobo into its own CI job so that we can switch back to the base container for default lab testing
• Adds assembly/disassembly automate environments to curobo test suite; fixes syntax issues in environments

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 multiple test failures and infrastructure issues:

Core fixes:

• Rendering fix: Resolves double-rendering bug where both KitVisualizer.step() and SimulationContext.render() were calling app.update(). Adds pumps_app_update() method to coordinate rendering responsibility and pauses simulation during updates to prevent frame skipping.
• Warp tensor conversions: Adds missing wp.to_torch() calls in AutoMate environments (assembly_env.py, disassembly_env.py) to fix type errors when accessing jacobians, mass matrices, and default states.
• Test robustness: Replaces brittle absolute tolerances with statistical checks (median/max) in force control tests to handle physics solver variability. Fixes flaky camera tests with deterministic cube teleportation, disabled gravity, and relaxed thresholds.

CI/Infrastructure:

• Moves cuRobo-dependent tests (AutoMate, SkillGen, Pink IK) to dedicated test-curobo job using separate Docker image, allowing base container to revert to standard dependencies.
• Adds isaaclab_teleop to cuRobo container for Pink IK dependencies.
• Makes isaaclab_teleop imports optional to prevent failures when not installed.
• Increases test timeouts and flaky test retry limits to reduce non-deterministic failures.

Confidence Score: 4/5

• This PR is safe to merge with minor concerns about disabled motion vector validation
• Score reflects well-understood bug fixes with clear explanations, but one test has disabled validation (motion vectors) that should be re-enabled. The rendering fix addresses a real double-update bug, the warp conversions fix actual type errors, and the CI restructuring is clean. The continue-on-error: true on cuRobo tests is intentional to prevent blocking on unstable tests.
• Pay attention to source/isaaclab/test/sensors/test_tiled_camera.py where motion vector validation is commented out with a TODO

Important Files Changed

Filename	Overview
source/isaaclab/isaaclab/visualizers/kit_visualizer.py	Adds pumps_app_update() method and pauses simulation during app.update() to fix rendering issue
source/isaaclab/isaaclab/visualizers/visualizer.py	Adds abstract pumps_app_update() method to prevent double-rendering in visualizer hierarchy
source/isaaclab/isaaclab/sim/simulation_context.py	Checks visualizers for pumps_app_update() and calls physics_manager.forward() to fix render lag
source/isaaclab/test/sensors/test_tiled_camera.py	Fixes flaky camera tests with deterministic cube motion, disabled gravity, and relaxed thresholds
source/isaaclab/test/controllers/test_operational_space.py	Replaces absolute tolerance with statistical median/max checks to handle physics solver variability
source/isaaclab_tasks/isaaclab_tasks/direct/automate/assembly_env.py	Adds missing wp.to_torch() conversions for warp tensors to prevent type errors
source/isaaclab_tasks/isaaclab_tasks/direct/automate/disassembly_env.py	Adds missing wp.to_torch() conversions for warp tensors to prevent type errors
.github/workflows/build.yml	Adds dedicated test-curobo CI job to run cuRobo-dependent tests in separate Docker container

Sequence Diagram

sequenceDiagram
    participant Test as Test Runner
    participant SimCtx as SimulationContext
    participant KitViz as KitVisualizer
    participant App as omni.kit.app
    
    Note over Test,App: Before Fix (Double Rendering)
    Test->>SimCtx: step(render=True)
    SimCtx->>SimCtx: physics_manager.step()
    SimCtx->>KitViz: update_visualizers()
    KitViz->>App: app.update() [1st render]
    SimCtx->>App: app.update() [2nd render - BUG!]
    
    Note over Test,App: After Fix (Single Rendering)
    Test->>SimCtx: step(render=True)
    SimCtx->>SimCtx: physics_manager.step()
    SimCtx->>KitViz: update_visualizers()
    KitViz->>App: set playSimulations=False
    KitViz->>App: app.update()
    KitViz->>App: set playSimulations=True
    SimCtx->>KitViz: pumps_app_update()?
    KitViz-->>SimCtx: True
    Note over SimCtx: Skip redundant app.update()

Loading

_{Last reviewed commit: f3c02cc}

[greptile-apps]

_{9 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

tolerance doubled from 5.0 to 10.0 N - verify this still catches force control regressions against 10 N targets

[greptile-apps]

_{19 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Motion vector validation is commented out with a TODO note. This means the test isn't actually verifying motion vectors work correctly - it only checks the shape but silently passes even if motion vectors are all zeros.

Suggested change

	# TODO: this looks broken on tot
	# assert im_data[i].abs().mean() > 0.001
	print(im_data[i].abs().mean())
	# Motion vectors should be non-zero since cubes are moving 0.5 units/frame
	assert im_data[i].abs().mean() > 0.001, f"Motion vectors appear broken: mean={im_data[i].abs().mean()}"

[greptile-apps]

continue-on-error: true means cuRobo test failures won't block the workflow. Combined with the if: always() on combine-results, this allows the PR to pass even if cuRobo tests fail.

[ooctipus]

we can use setting manager here

[ooctipus]

is this kit visualizer specific? if is, would be better to implement in kit visualizer instead of simulation context

[kellyguo11]

but that would mean we need to enable the kit visualizer if we want to do rendering? I think eventually this should be moved to the renderer class, but we need that separation first from Piotr and team

[ooctipus]

i see this is for sensor rendering?
I thought sensor rendering responsibility if compeletely owned by sensor, if thats then, maybe this code needs come from piotr? Or could it be implemented in tiledCamera implementation instead?

[kellyguo11]

yes it's sensor rendering. I agree it should be refactored out with Piotr's changes, but I don't know what that looks like yet, so I just left it here for now. the issue with putting it in TiledCamera right now is I'm not sure how to handle synchronizing the step call across multiple cameras. This change was mainly to bring back rendering so that at least we can still train perception environments on the current develop branch.