[Workflow] Default classic core tasks to Newton by StafaH · Pull Request #7066 · isaac-sim/IsaacLab

StafaH · 2026-08-13T08:00:53Z

Splits the Cartpole and shared-renderer defaults from #6980.\n\n- Defaults Cartpole to Newton MJWarp.\n- Defaults MultiBackendRendererCfg to the Newton renderer while retaining explicit Isaac RTX selection.\n- Consolidates core environment smoke tests into one three-backend matrix (newton_mjwarp, physx, isaacsim_physx).\n- Removes obsolete Play/MJWarp-discovery test utility logic and the redundant Newton test module.\n\nValidation: uv run python -m pytest source/isaaclab_tasks/test/core/test_preset_kit_decision.py source/isaaclab_tasks/test/core/test_runtime_compatibility.py -q (51 passed).\n\nSupersedes the classic-controls portion of #6980.

greptile-apps · 2026-08-13T08:03:38Z

Greptile Summary

The PR changes Cartpole's default physics and rendering backends to Newton and consolidates core environment smoke coverage around three backend selections.

Defaults direct and manager-based Cartpole configurations to Newton MJWarp.
Defaults Cartpole camera rendering to the Newton renderer while preserving explicit alternatives.
Replaces the separate Newton smoke module with a config-derived backend matrix and updates related tests and golden artifacts.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
source/isaaclab_tasks/test/core/test_environments.py	Consolidates core environment smoke coverage into task/backend parameters derived from registered configurations.
source/isaaclab_tasks/isaaclab_tasks/core/cartpole/cartpole_direct_env_cfg.py	Changes the direct Cartpole physics default from Isaac Sim PhysX to Newton MJWarp.
source/isaaclab_tasks/isaaclab_tasks/core/cartpole/cartpole_manager_env_cfg.py	Changes the manager-based Cartpole physics default to Newton MJWarp.
source/isaaclab_tasks/test/env_test_utils.py	Simplifies environment discovery and supports explicit physics selection in consolidated smoke tests.

_{Reviews (2): Last reviewed commit: "Revert "Split environment smoke tests by..." | Re-trigger Greptile}

greptile-apps · 2026-08-13T08:03:41Z

 # Local imports should be imported last
 from env_test_utils import _run_environments, setup_environment  # isort: skip

+@pytest.mark.parametrize("physics_preset_name", ["newton_mjwarp", "physx", "isaacsim_physx"])


Unsupported presets break smoke matrix

When the Fourbar Pole task is collected, this matrix passes newton_mjwarp, physx, and isaacsim_physx to a configuration that declares only default and newton_kamino; preset resolution raises ValueError, causing the consolidated CI smoke suite to fail.

Knowledge Base Used: isaaclab_tasks: Task Registration and Organization

Fixed in 6c98c4f: core smoke parameters are now generated only for explicitly declared PhysicsCfg presets, so Fourbar is not mislabeled as running unsupported backends.

isaaclab-review-bot

Isaac Lab Review Bot

The change consistently defaults both Cartpole configurations and the shared renderer preset to Newton while retaining explicit Isaac Sim PhysX and RTX selections. The consolidated smoke-test matrix and workflow update account for removal of the separate Newton test module.

Design and architecture: The physics and renderer default changes align across direct and manager-based Cartpole configurations. Backend selection remains explicit through newton_mjwarp, physx, isaacsim_physx, rtx, and isaacsim_rtx, preserving the intended compatibility paths.
API: The changed preset fields retain concrete types, and explicit Isaac RTX selection remains available as a distinct IsaacRtxRendererCfg. Updated test-utility callers no longer use the removed discovery arguments, and the breaking default change includes a changelog fragment with migration guidance.
Implementation: Preset-resolution and runtime-compatibility tests trace the new kitless Newton/OVRTX default as well as explicit Kit-backed selections. The three-backend core smoke-test matrix replaces the removed Newton-specific module, and cleanup now collects Warp-backed objects before simulation teardown. The candidate concern is limited to contradictory test names and docstrings; it does not demonstrate a production behavior or contract failure requiring pre-merge action.

No blocking issues. No inline issue met the actionable-evidence threshold; the assessment above records the review feedback.

Automated review; human maintainers own approval decisions.

AntoineRichard · 2026-08-13T12:23:06Z

 @configclass
 class MultiBackendRendererCfg(PresetCfg):
-    default: IsaacRtxRendererCfg = IsaacRtxRendererCfg()
+    default: NewtonWarpRendererCfg = NewtonWarpRendererCfg()


This changes the renderer default for every task that uses MultiBackendRendererCfg, not only the Cartpole family. The failures show that several existing consumers cannot use that default:

Mimic and contrib GR1T2 PickPlace now instantiate NewtonWarpRenderer and fail with ValueError: Failed to load texture: ...GR1T2_albedo.<UDIM>.png, so dataset generation never starts.

The XR-camera and Shadow Hand preset tests resolve default to NewtonWarpRendererCfg instead of their required IsaacRtxRendererCfg.

Registered-task rendering loses the Kit RenderProduct, and Newton returns no output for the MDL shading data types those tasks request.

Please keep the shared default unchanged and introduce a Cartpole-specific renderer preset/default, or migrate and explicitly pin every affected consumer while providing the missing Newton renderer support. The current shared change is a runtime regression outside the PR title scope.

Fixed in 6c98c4f: MultiBackendRendererCfg retains its Isaac RTX default. Newton renderer defaults are now scoped to the Cartpole camera configurations, preserving contrib, Mimic, XR, and Shadow Hand behavior.

AntoineRichard · 2026-08-13T12:23:17Z

 from env_test_utils import _run_environments, setup_environment  # isort: skip


+@pytest.mark.parametrize("physics_preset_name", ["newton_mjwarp", "physx", "isaacsim_physx"])


This matrix does not guarantee the backend named in the test ID. Global preset resolution falls back to default when a task does not expose the requested name. The CI log demonstrates this directly: all three newton_mjwarp, physx, and isaacsim_physx Fourbar cases pass even though Fourbar only declares default/newton_kamino; the labels are therefore not evidence that those backends ran.

Please generate only supported (task, physics preset) pairs, or use a strict physics selector that errors when unavailable. Also restore include_play=False unless every Play variant is intentional: this change collected 252 cases here and 42 stage-in-memory cases, substantially increasing runtime and repeated teardown pressure.

Fixed in 6c98c4f: the matrix now includes only task/backend pairs whose raw config explicitly exposes the requested physics preset. Play filtering remains removed because the current registry has no Play task variants.

AntoineRichard · 2026-08-13T12:23:26Z

+Changed
+^^^^^^^
+
+* **Breaking:** Changed the Cartpole task family and the default renderer preset


This entry explicitly marks both default changes as breaking, but I could not find a prior deprecation for either the Cartpole backend default or the shared renderer default. Please reconcile this with the repository policy that breaking changes require a deprecation first, or document why these default changes are exempt.

This is an intentional 3.0 breaking default change. The fragment now accurately scopes it to Cartpole and provides the explicit-preset migration path; we are not adding deprecation artifacts for this release.

AntoineRichard · 2026-08-13T12:23:45Z

I traced all seven red jobs at 543ee34:

The shared renderer-default change causes deterministic cross-task regressions. isaaclab_mimic and the contrib PickPlace smoke test both fail while Newton loads the GR1T2 UDIM texture; the XR/Shadow preset assertions resolve Newton instead of Isaac RTX; and the two RL summary expectations still describe the former concrete defaults.
The default-rendering baselines were not migrated. test_rendering_registered_tasks has seven failures: the Cartpole stage no longer has the Kit RenderProduct/PhysX prims, RGB differs by 94.01%, albedo by 2.47%, and the Newton renderer produces zero outputs for the three MDL shading variants. Shadow Hand also loses its Kit RenderProduct. The visualizer job separately reports a stable 6.33% pixel difference for newton-cartpole-kit-tiled; that helper leaves the renderer implicit, so the shared default change affects a test labeled Kit.
Two task files terminate natively. test_environments_with_stage_in_memory.py exits 139 while starting Isaac-Lift-Soft-Franka-Camera; test_environments.py exits 139 during Isaac-Lift-Cloth-Franka-Camera with newton_mjwarp. Neither produces a Python traceback or JUnit report. These should be isolated after the renderer scope is corrected.

The assertion/image/texture failures are deterministic and will not be fixed by a rerun. I would first make the Newton renderer default Cartpole-specific (or complete the migration of every shared consumer), then update or explicitly pin the Cartpole/Kit golden tests and rerun the two crash-only cases.

StafaH · 2026-08-13T20:20:15Z

Addressed the deterministic regressions in 6c98c4f and 1e76a29: the generic renderer default is restored to Isaac RTX, while Newton is the default renderer only for Cartpole camera configs; Cartpole visualizer goldens explicitly pin Isaac RTX; and the smoke matrix filters to supported PhysicsCfg presets. The existing teardown release/GC fix remains in place for the native cleanup crashes. The default switch is intentional for 3.0 and documented without deprecation artifacts.

This reverts commit 16f2227.

This reverts commit 389f780.

This reverts commit ed946fa.

…classic # Conflicts: # .github/workflows/build.yaml # source/isaaclab_tasks/test/contrib/test_cartpole_showcase_environments_with_stage_in_memory.py # source/isaaclab_tasks/test/contrib/test_contrib_environments_smoke.py # source/isaaclab_tasks/test/contrib/test_pickplace_stack_environments.py # source/isaaclab_tasks/test/contrib/test_teleop_environments.py # source/isaaclab_tasks/test/contrib/test_teleop_environments_with_stage_in_memory.py # source/isaaclab_tasks/test/core/test_environments_isaacsim_physx.py # source/isaaclab_tasks/test/core/test_environments_newton.py # source/isaaclab_tasks/test/core/test_environments_ovphysx.py # source/isaaclab_tasks/test/core/test_environments_with_stage_in_memory.py # source/isaaclab_tasks/test/core/test_preset_kit_decision.py # source/isaaclab_tasks/test/core/test_record_video.py # source/isaaclab_tasks/test/env_test_utils.py

…classic

… physics backend (#7334) # Description `--deterministic` configured PyTorch and the Isaac RTX renderer but never reached the physics solver, so training on Newton backends was not reproducible even with the flag passed. Two defaults moved out from under the flag between 3.0beta2 and GA: | Default | v3.0.0-beta2 | GA | |---|---|---| | `CartpolePhysicsCfg.default` | `PhysxCfg()` | `NewtonCfg` MJWarp, `deterministic_mode="not_guaranteed"` | | `MultiBackendRendererCfg.default` | `IsaacRtxRendererCfg()` | `NewtonWarpRendererCfg()` | The physics switch landed in `0caae64dc7c` (#7066), absent from all three `v3.0.0-beta*` tags. At beta2 the flag worked because PhysX is run-to-run deterministic for rigid bodies **and** Isaac RTX was the default renderer; both premises were removed without re-wiring the flag. `--deterministic` now sets `PhysicsCfg.deterministic` on the resolved physics config, in `apply_env_overrides()` — the existing CLI-to-cfg seam, after `scan()` resolves the backend and before the solver is built. That field is the backend-agnostic request; each physics manager translates it when the simulation starts: - **Newton** derives `deterministic_mode="run_to_run"`, applies the MJWarp `disable_sensors` prerequisite on the GPU path, and leaves MuJoCo-CPU alone. An explicitly set `deterministic_mode` wins. - **PhysX / OvPhysX** enable `enable_enhanced_determinism`. OvPhysX is best-effort and not verified end to end. Validation stays with the backend: `NewtonManager._validate_deterministic_solver_cfg()` rejects an unsupported solver at solver initialization, so there is one policy and one set of error messages rather than a copy in the RL layer. Adding a backend no longer means editing `isaaclab_rl`. **A determinism request that would starve a sensor is now refused.** Disabling MuJoCo Warp's sensors also skips its `rne_postconstraint` stage, which fills Newton's `body_qdd` / `body_parent_f`. The IMU, PVA and joint-wrench sensors read that state, so `Isaac-Ant`, `Isaac-Humanoid` and `Isaac-Repose-Cube-Shadow` — all defaulting to `newton_mjwarp` and feeding `joint_wrench` into their policy observations — would have trained on values that are never refreshed, with no error. `NewtonManager` raises at solver initialization, the only point where both the solver config and the registered sensors are visible. The guard restates `{"body_qdd", "body_parent_f"}`, which Newton also states internally (`solver_mujoco.py:4360` as a set, `:5179` as an equivalent `or`-chain). That duplication is tracked upstream in newton-physics/newton#4109, which asks for two things: Newton refusing the combination at the source, and exporting the field set as a public constant. Either lets this guard shrink — the constant and its sensor-tracking delete entirely once Isaac Lab pins a Newton that raises. Until then the guard is what prevents the silent case, so it stays. Fixes NVBug 6658578 (P0, Isaac Lab 3.0 GA). ## Type of change - Bug fix (non-breaking change which fixes an issue) ## Validation Three `--deterministic` runs of `Isaac-Cartpole-Camera` (50 epochs, task defaults) on one L40, each in its own container: ``` det1 vs det2 55/55 checkpoint tensors bitwise identical det1 vs det3 55/55 checkpoint tensors bitwise identical ctrl1 vs ctrl2 41/55 differ <- same task, no flag ``` The unflagged controls diverge, so the agreement above is the flag's doing rather than a task that is trivially reproducible. `Isaac-Ant --deterministic` fails at startup with the sensor message. MuJoCo-CPU could not be exercised: `SolverMuJoCo.get_max_contact_count()` raises `NotImplementedError` on that path, so it is unreachable in Isaac Lab today. 8 unit tests added; 200 pass across `test_entrypoints_common.py` and `test_newton_manager_abstraction.py`. ## Release backport - [x]  Backport this pull request to the active release branch after it merges into `develop` ## Checklist - [x] I have read and understood the [contribution guidelines](https://isaac-sim.github.io/IsaacLab/main/source/refs/contributing.html) - [x] I have run the [`pre-commit` checks](https://pre-commit.com/) with `./isaaclab.sh --format` - [x] I have made corresponding changes to the documentation - [x] My changes generate no new warnings - [x] I have added tests that prove my fix is effective or that my feature works - [x] I have added a changelog fragment under `source/<pkg>/changelog.d/` for every touched package - [x] I have added my name to the `CONTRIBUTORS.md` or my name already exists there

… physics backend (#7334) # Description `--deterministic` configured PyTorch and the Isaac RTX renderer but never reached the physics solver, so training on Newton backends was not reproducible even with the flag passed. Two defaults moved out from under the flag between 3.0beta2 and GA: | Default | v3.0.0-beta2 | GA | |---|---|---| | `CartpolePhysicsCfg.default` | `PhysxCfg()` | `NewtonCfg` MJWarp, `deterministic_mode="not_guaranteed"` | | `MultiBackendRendererCfg.default` | `IsaacRtxRendererCfg()` | `NewtonWarpRendererCfg()` | The physics switch landed in `0caae64dc7c` (#7066), absent from all three `v3.0.0-beta*` tags. At beta2 the flag worked because PhysX is run-to-run deterministic for rigid bodies **and** Isaac RTX was the default renderer; both premises were removed without re-wiring the flag. `--deterministic` now sets `PhysicsCfg.deterministic` on the resolved physics config, in `apply_env_overrides()` — the existing CLI-to-cfg seam, after `scan()` resolves the backend and before the solver is built. That field is the backend-agnostic request; each physics manager translates it when the simulation starts: - **Newton** derives `deterministic_mode="run_to_run"`, applies the MJWarp `disable_sensors` prerequisite on the GPU path, and leaves MuJoCo-CPU alone. An explicitly set `deterministic_mode` wins. - **PhysX / OvPhysX** enable `enable_enhanced_determinism`. OvPhysX is best-effort and not verified end to end. Validation stays with the backend: `NewtonManager._validate_deterministic_solver_cfg()` rejects an unsupported solver at solver initialization, so there is one policy and one set of error messages rather than a copy in the RL layer. Adding a backend no longer means editing `isaaclab_rl`. **A determinism request that would starve a sensor is now refused.** Disabling MuJoCo Warp's sensors also skips its `rne_postconstraint` stage, which fills Newton's `body_qdd` / `body_parent_f`. The IMU, PVA and joint-wrench sensors read that state, so `Isaac-Ant`, `Isaac-Humanoid` and `Isaac-Repose-Cube-Shadow` — all defaulting to `newton_mjwarp` and feeding `joint_wrench` into their policy observations — would have trained on values that are never refreshed, with no error. `NewtonManager` raises at solver initialization, the only point where both the solver config and the registered sensors are visible. The guard restates `{"body_qdd", "body_parent_f"}`, which Newton also states internally (`solver_mujoco.py:4360` as a set, `:5179` as an equivalent `or`-chain). That duplication is tracked upstream in newton-physics/newton#4109, which asks for two things: Newton refusing the combination at the source, and exporting the field set as a public constant. Either lets this guard shrink — the constant and its sensor-tracking delete entirely once Isaac Lab pins a Newton that raises. Until then the guard is what prevents the silent case, so it stays. Fixes NVBug 6658578 (P0, Isaac Lab 3.0 GA). ## Type of change - Bug fix (non-breaking change which fixes an issue) ## Validation Three `--deterministic` runs of `Isaac-Cartpole-Camera` (50 epochs, task defaults) on one L40, each in its own container: ``` det1 vs det2 55/55 checkpoint tensors bitwise identical det1 vs det3 55/55 checkpoint tensors bitwise identical ctrl1 vs ctrl2 41/55 differ <- same task, no flag ``` The unflagged controls diverge, so the agreement above is the flag's doing rather than a task that is trivially reproducible. `Isaac-Ant --deterministic` fails at startup with the sensor message. MuJoCo-CPU could not be exercised: `SolverMuJoCo.get_max_contact_count()` raises `NotImplementedError` on that path, so it is unreachable in Isaac Lab today. 8 unit tests added; 200 pass across `test_entrypoints_common.py` and `test_newton_manager_abstraction.py`. ## Release backport - [x]  Backport this pull request to the active release branch after it merges into `develop` ## Checklist - [x] I have read and understood the [contribution guidelines](https://isaac-sim.github.io/IsaacLab/main/source/refs/contributing.html) - [x] I have run the [`pre-commit` checks](https://pre-commit.com/) with `./isaaclab.sh --format` - [x] I have made corresponding changes to the documentation - [x] My changes generate no new warnings - [x] I have added tests that prove my fix is effective or that my feature works - [x] I have added a changelog fragment under `source/<pkg>/changelog.d/` for every touched package - [x] I have added my name to the `CONTRIBUTORS.md` or my name already exists there (cherry picked from commit 3748a72)

StafaH added 4 commits August 12, 2026 21:03

Consolidate environment backend smoke tests

f0eaf24

Default classic tasks to Newton

c3329cb

Collect environment objects before teardown

742fbf2

Update preset tests for Newton defaults

f416d9e

StafaH requested a review from a team August 13, 2026 08:00

StafaH requested review from hujc7, kellyguo11, myurasov-nv and ooctipus as code owners August 13, 2026 08:00

StafaH mentioned this pull request Aug 13, 2026

[Worflow] Make newton and rsl_rl default across core tasks #6980

Closed

7 tasks

github-actions Bot added isaac-lab Related to Isaac Lab team infrastructure labels Aug 13, 2026

greptile-apps Bot reviewed Aug 13, 2026

View reviewed changes

Format environment test updates

543ee34

isaaclab-review-bot Bot reviewed Aug 13, 2026

View reviewed changes

AntoineRichard approved these changes Aug 13, 2026

View reviewed changes

AntoineRichard reviewed Aug 13, 2026

View reviewed changes

StafaH added 3 commits August 13, 2026 11:17

Pin registered rendering tests to Isaac Sim

0fddcfe

Release environments before simulation teardown

c0e5f45

Scope Newton renderer defaults to Cartpole

6c98c4f

StafaH requested a review from matthewtrepte as a code owner August 13, 2026 20:18

Clarify Cartpole renderer release note

1e76a29

Fix Cartpole backend test baselines

20955dc

StafaH requested a review from ClemensSchwarke as a code owner August 13, 2026 22:16

Split environment smoke tests by backend

3e5b9be

StafaH added 5 commits August 13, 2026 17:10

Retain stage-in-memory test utilities

9ad93dd

Test OV PhysX environment presets

ac93fbd

Revert "Test OV PhysX environment presets"

d1f4ef9

This reverts commit 16f2227.

Revert "Retain stage-in-memory test utilities"

d0a8017

This reverts commit 389f780.

Revert "Split environment smoke tests by backend"

e8cf939

This reverts commit ed946fa.

StafaH closed this Aug 14, 2026

StafaH reopened this Aug 14, 2026

kellyguo11 approved these changes Aug 14, 2026

View reviewed changes

StafaH and others added 4 commits August 14, 2026 12:33

Merge remote-tracking branch 'origin/develop' into mh/default-newton-…

07000a9

…classic

Fix classic default backend tests

5ffde71

Use RTX renderer for PhysX video test

46d2aac

Merge branch 'develop' into mh/default-newton-classic

48065ef

ooctipus approved these changes Aug 15, 2026

View reviewed changes

ooctipus merged commit 0caae64 into isaac-sim:develop Aug 15, 2026
73 of 77 checks passed

ooctipus deleted the mh/default-newton-classic branch August 15, 2026 11:07

hujc7 mentioned this pull request Aug 25, 2026

[Fix] Make --deterministic reproduce training runs by configuring the physics backend #7334

Merged

8 tasks

Navigation Menu

Uh oh!

[Workflow] Default classic core tasks to Newton - #7066

[Workflow] Default classic core tasks to Newton#7066
ooctipus merged 21 commits into
isaac-sim:developfrom
StafaH:mh/default-newton-classic

StafaH commented Aug 13, 2026

Uh oh!

greptile-apps Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot Aug 13, 2026

Uh oh!

StafaH Aug 13, 2026

Uh oh!

isaaclab-review-bot Bot left a comment

Uh oh!

AntoineRichard Aug 13, 2026

Uh oh!

StafaH Aug 13, 2026

Uh oh!

AntoineRichard Aug 13, 2026

Uh oh!

StafaH Aug 13, 2026

Uh oh!

AntoineRichard Aug 13, 2026

Uh oh!

StafaH Aug 13, 2026

Uh oh!

AntoineRichard commented Aug 13, 2026

Uh oh!

StafaH commented Aug 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

		from env_test_utils import _run_environments, setup_environment # isort: skip


		@pytest.mark.parametrize("physics_preset_name", ["newton_mjwarp", "physx", "isaacsim_physx"])

Uh oh!

Conversation

StafaH commented Aug 13, 2026

Uh oh!

greptile-apps Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

greptile-apps Bot Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

StafaH Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

isaaclab-review-bot Bot left a comment

Choose a reason for hiding this comment

Isaac Lab Review Bot

Uh oh!

AntoineRichard Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

StafaH Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

AntoineRichard Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

StafaH Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

AntoineRichard Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

StafaH Aug 13, 2026

Choose a reason for hiding this comment

Uh oh!

AntoineRichard commented Aug 13, 2026

Uh oh!

StafaH commented Aug 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

greptile-apps Bot commented Aug 13, 2026 •

edited

Loading