[Newton] Articulation & Rigid Object Full Tests by AntoineRichard · Pull Request #4406 · isaac-sim/IsaacLab

AntoineRichard · 2026-01-16T13:38:37Z

Description

Brings back all of the original integration tests from the Articulation + the wrench composer.

Also provides UTs, ITs and micro benchmarks for the articulation with newton. Some integration tests are still failing / skipped as the feature set of newton is not complete yet.

Run a unit-test:

./isaaclab.sh -p -m pytest source/isaaclab_newton/test/assets/rigid_object/test_rigid_object.py
./isaaclab.sh -p -m pytest source/isaaclab_newton/test/assets/rigid_object/test_rigid_object_data.py
./isaaclab.sh -p -m pytest source/isaaclab_newton/test/assets/rigid_object/test_articulation.py
./isaaclab.sh -p -m pytest source/isaaclab_newton/test/assets/rigid_object/test_articulation_data.py

Run integration tests:

./isaaclab.sh -p -m pytest source/isaaclab_newton/test/assets/rigid_object/test_integration_rigid_object.py
./isaaclab.sh -p -m pytest source/isaaclab_newton/test/assets/rigid_object/test_integration_articulation.py

Run a micro-benchmark:

./isaaclab.sh -p source/isaaclab_newton/test/assets/rigid_object/benchmark_rigid_object.py
./isaaclab.sh -p source/isaaclab_newton/test/assets/rigid_object/benchmark_rigid_object_data.py
./isaaclab.sh -p source/isaaclab_newton/test/assets/rigid_object/benchmark_articulation.py
./isaaclab.sh -p source/isaaclab_newton/test/assets/rigid_object/benchmark_articulation_data.py

Known slow down:

The methods: make_complete_data_from_torch_dual_index, make_complete_data_from_torch_single_index, make_masks_from_torch_ids, are very slow, and they should be modified to include a out argument, that would allow to preallocate the tensors.
For now most of the warp code was written with modularity in mind, this means that sometimes we launch multiple kernels when we should really just be launching one. For performance optimization we should track these instances and write dedicated kernels.

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

… at least). There are still some errors some will be patched by latest newton.

…h more or less success.

…lean-up session with a Newton Eng to sanity check this.

…position as this is now handled properly by newton. Still a couple of weird cases left.

AntoineRichard · 2026-01-21T10:42:23Z

@greptileai could you give it a review?

greptile-apps · 2026-01-21T10:46:08Z

Greptile Summary

Restores comprehensive integration tests, unit tests, and micro-benchmarks for Articulation and RigidObject classes in the Newton physics backend, bringing back full test coverage that was previously missing
Introduces WrenchComposer utility class for managing external forces and torques on rigid bodies with support for both instantaneous and permanent force application patterns
Implements extensive performance optimizations including lazy buffer allocation, memory view optimizations, and pre-allocated temporary buffers for torch-to-warp array conversions

Important Files Changed

Filename	Overview
`source/isaaclab/utils/wrench_composer.py`	New WrenchComposer utility for external force management with critical initialization bug on line 37
`source/isaaclab/utils/warp/kernels.py`	Added new Warp kernels for force composition with inconsistent torque calculation approach between functions
`source/isaaclab_newton/isaaclab_newton/assets/rigid_object/rigid_object.py`	Major refactoring integrating WrenchComposer with complex lazy allocation and wrench application logic
`source/isaaclab_newton/isaaclab_newton/assets/articulation/articulation.py`	Comprehensive refactor adding full wrench composer support and extensive state validation logic
`source/isaaclab_newton/isaaclab_newton/assets/rigid_object/rigid_object_data.py`	Extensive performance optimizations with removed deprecated decorators and complex memory management changes

Confidence score: 3/5

This PR introduces significant architectural changes and new functionality but contains critical bugs and potential issues requiring careful review
Score lowered due to critical initialization bug in WrenchComposer (line 37), inconsistent torque calculation approaches, many skipped/failing tests indicating incomplete Newton feature set, and extensive API changes with deprecated methods becoming permanent
Pay close attention to source/isaaclab/utils/wrench_composer.py for the initialization bug, force/torque composition kernels for mathematical correctness, and integration test files for understanding Newton backend limitations

greptile-apps

_{38 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-01-21T10:46:07Z

+        if hasattr(asset, "num_bodies"):
+            self.num_bodies = asset.num_bodies
+        else:
+            raise ValueError(f"Unsupported asset type: {self._asset.__class__.__name__}")


logic: self._asset is referenced before being assigned (line 39)

Suggested change

raise ValueError(f"Unsupported asset type: {self._asset.__class__.__name__}")

raise ValueError(f"Unsupported asset type: {asset.__class__.__name__}")

AntoineRichard · 2026-01-21T14:26:53Z

@greptileai let's do another round of review.

greptile-apps

Additional Comments (1)

source/isaaclab/isaaclab/sensors/sensor_base.py, line 206 (link)

logic: The env_mask parameter is defined but never used in the implementation. Should env_mask be implemented alongside env_ids, or is this parameter being added as a placeholder for future development?

_{39 files reviewed, 9 comments}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-01-21T14:37:04Z

    def test_reset(self):
        """Test that reset method works properly."""
        articulation, _, _ = create_test_articulation()
+        articulation._create_buffers()


logic: Duplicate _create_buffers() call - already called in factory function on line 125

Suggested change

articulation._create_buffers()

articulation, _, _ = create_test_articulation()

ooctipus · 2026-01-24T09:48:58Z

I gave a rough read and ran the tests myself, beyond making those failed one pass,

Our new tests are a lot more thorough than before, this is good, but it did took me a decent amount of effort to start make sense what is going on. I think some tests could be refactored further, and could potentially help a lot of new

for example:

    assert articulation.is_initialized
    # Check that floating base
    assert not articulation.is_fixed_base
    # Check buffers that exists and have correct shapes
    assert wp.to_torch(articulation.data.root_pos_w).shape == (num_articulations, 3)
    assert wp.to_torch(articulation.data.root_quat_w).shape == (num_articulations, 4)
    assert wp.to_torch(articulation.data.joint_pos).shape == (num_articulations, 12)
    assert wp.to_torch(articulation.data.body_mass).shape == (num_articulations, articulation.num_bodies)
    assert wp.to_torch(articulation.data.body_inertia).shape == (num_articulations, articulation.num_bodies, 3, 3)

    # Check some internal physx data for debugging
    # -- joint related
    assert articulation.num_joints == NewtonManager.get_model().joint_dof_count / num_articulations - 6
    # -- link related
    assert articulation.num_bodies == NewtonManager.get_model().body_count / num_articulations
    # -- link names (check within articulation ordering is correct)
    prim_path_body_names = articulation.root_view.body_names
    assert prim_path_body_names == articulation.body_names
    # -- actuator type
    for actuator_name, actuator in articulation.actuators.items():
        is_implicit_model_cfg = isinstance(articulation_cfg.actuators[actuator_name], ImplicitActuatorCfg)
        assert actuator.is_implicit_model == is_implicit_model_cfg

    # Simulate physics
    for _ in range(10):
        # perform rendering
        sim.step(render=False)
        # update articulation
        articulation.update(sim.cfg.dt)

this pattern repeated many time in articulation_integration_tests.py

all these instances could be replaced with something like

    _assert_initialized_common(articulation, num_articulations, expected_joint_count=12, expected_fixed_base=False, check_body_props=True,)
    _assert_model_counts(articulation, num_articulations, subtract_root_dofs=True)
    _assert_body_names_from_root_view(articulation)
    _assert_actuator_types(articulation, articulation_cfg)
    _step_sim(sim, articulation, steps=10, render=False)

Maybe after these tests passed, let cursor do some of these refactor for us?

ooctipus · 2026-01-24T09:50:09Z

some of the failure seems like requires newton or mujoco_warp to be fixed, is the goal to merge it with failed tests, or merge after those has been fixed?

ooctipus · 2026-01-28T17:41:12Z

    cfg = cfg.to_dict()
    # extract non-USD properties
    fix_root_link = cfg.pop("fix_root_link", None)
+    print("fix_root_link", fix_root_link)


AntoineRichard · 2026-01-29T13:38:57Z

I gave a rough read and ran the tests myself, beyond making those failed one pass,

Our new tests are a lot more thorough than before, this is good, but it did took me a decent amount of effort to start make sense what is going on. I think some tests could be refactored further, and could potentially help a lot of new

for example:

    assert articulation.is_initialized
    # Check that floating base
    assert not articulation.is_fixed_base
    # Check buffers that exists and have correct shapes
    assert wp.to_torch(articulation.data.root_pos_w).shape == (num_articulations, 3)
    assert wp.to_torch(articulation.data.root_quat_w).shape == (num_articulations, 4)
    assert wp.to_torch(articulation.data.joint_pos).shape == (num_articulations, 12)
    assert wp.to_torch(articulation.data.body_mass).shape == (num_articulations, articulation.num_bodies)
    assert wp.to_torch(articulation.data.body_inertia).shape == (num_articulations, articulation.num_bodies, 3, 3)

    # Check some internal physx data for debugging
    # -- joint related
    assert articulation.num_joints == NewtonManager.get_model().joint_dof_count / num_articulations - 6
    # -- link related
    assert articulation.num_bodies == NewtonManager.get_model().body_count / num_articulations
    # -- link names (check within articulation ordering is correct)
    prim_path_body_names = articulation.root_view.body_names
    assert prim_path_body_names == articulation.body_names
    # -- actuator type
    for actuator_name, actuator in articulation.actuators.items():
        is_implicit_model_cfg = isinstance(articulation_cfg.actuators[actuator_name], ImplicitActuatorCfg)
        assert actuator.is_implicit_model == is_implicit_model_cfg

    # Simulate physics
    for _ in range(10):
        # perform rendering
        sim.step(render=False)
        # update articulation
        articulation.update(sim.cfg.dt)

this pattern repeated many time in articulation_integration_tests.py

all these instances could be replaced with something like

    _assert_initialized_common(articulation, num_articulations, expected_joint_count=12, expected_fixed_base=False, check_body_props=True,)
    _assert_model_counts(articulation, num_articulations, subtract_root_dofs=True)
    _assert_body_names_from_root_view(articulation)
    _assert_actuator_types(articulation, articulation_cfg)
    _step_sim(sim, articulation, steps=10, render=False)

Maybe after these tests passed, let cursor do some of these refactor for us?

Yeah these are the OG tests. I didn't want to alter them too much to see what kind of coverage we have in newton. I think they would need to be reworked in general to be more thorough.

…d paradigm

github-actions Bot added documentation Improvements or additions to documentation asset New asset feature or request infrastructure labels Jan 16, 2026

AntoineRichard marked this pull request as ready for review January 20, 2026 12:47

AntoineRichard requested review from Mayankm96, hhansen-bdai, jtigue-bdai, kellyguo11 and pascal-roth as code owners January 20, 2026 12:47

AntoineRichard changed the title ~~[Newton] Articulation Upgrade~~ [Newton] Articulation & Rigid Object Full Tests Jan 21, 2026

AntoineRichard added 14 commits January 21, 2026 11:36

Rigid object is passing both unit and integration tests (most of them…

cbe24bb

… at least). There are still some errors some will be patched by latest newton.

Changed Newton version to latest

599739e

pre-commits

8aa4c6d

test articulation WIP

80282da

improved CPU support, fixed some minor bugs and ran all the tests wit…

eb7845c

…h more or less success.

moved and renamed

308e1c9

bringing more of the rigid object back.

9dca442

Fix some tolerances in the rigid object test. Might be good to do a c…

9af2fea

…lean-up session with a Newton Eng to sanity check this.

tested friction Ok. Restitution not OK.

fd5eb20

various fixes.

3653d3d

WIP

b3328ff

addressed some of our Lizard overloard comments.

f8f51ea

pre-commits

14ffe75

Skipping failing tests + various test fix. We no longer offset robot …

1befd82

…position as this is now handled properly by newton. Still a couple of weird cases left.

AntoineRichard force-pushed the antoiner/articulation_upgrade branch from 2fc4559 to 1befd82 Compare January 21, 2026 10:36

github-actions Bot added the isaac-lab Related to Isaac Lab team label Jan 21, 2026

AntoineRichard added 2 commits January 21, 2026 11:36

pre-commits

2b19553

removing prints

7cf49ff

greptile-apps Bot reviewed Jan 21, 2026

View reviewed changes

AntoineRichard added 4 commits January 21, 2026 15:00

perf opt

9bc30d0

Improved doc-strings

11da408

pre-commits

7749b3b

Should be ready to merge / test

5974d98

greptile-apps Bot reviewed Jan 21, 2026

View reviewed changes

AntoineRichard added 3 commits January 21, 2026 15:54

Fixed some more Nits.

79b0949

forgot to include common...

b78e9b1

pre-commits

89c1b27

ooctipus reviewed Jan 28, 2026

View reviewed changes

ooctipus mentioned this pull request Jan 28, 2026

[Newton] contact behavior tests and contact sensor tests #4483

Closed

ooctipus approved these changes Jan 29, 2026

View reviewed changes

AntoineRichard added 6 commits February 2, 2026 13:56

Fixed issue with latest Newton.

1183da7

updated deps

55258be

fixing bugs here and here + improved spawning logic to match new worl…

2572722

…d paradigm

Let's go latest flashiest Newton!

43e904b

Merge branch 'dev/newton' into antoiner/articulation_upgrade

9171347

pre-commits

49abb53

AntoineRichard merged commit 48e23d2 into isaac-sim:dev/newton Feb 3, 2026
3 of 9 checks passed

	raise ValueError(f"Unsupported asset type: {self._asset.__class__.__name__}")
	raise ValueError(f"Unsupported asset type: {asset.__class__.__name__}")

	articulation._create_buffers()
	articulation, _, _ = create_test_articulation()

Conversation

AntoineRichard commented Jan 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Checklist

Uh oh!

AntoineRichard commented Jan 21, 2026

Uh oh!

greptile-apps Bot commented Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Important Files Changed

Confidence score: 3/5

Uh oh!

greptile-apps Bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

AntoineRichard commented Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greptile-apps Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Additional Comments (1)

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

greptile-apps Bot Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

ooctipus commented Jan 24, 2026

Uh oh!

ooctipus commented Jan 24, 2026

Uh oh!

ooctipus Jan 28, 2026

Choose a reason for hiding this comment

Uh oh!

AntoineRichard commented Jan 29, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

AntoineRichard commented Jan 16, 2026 •

edited

Loading

greptile-apps Bot commented Jan 21, 2026 •

edited

Loading

AntoineRichard commented Jan 21, 2026 •

edited

Loading

greptile-apps Bot left a comment •

edited

Loading