Fixes backend deprecation warning call sites

[AntoineRichard]

Summary

• Updated PhysX and Newton backend tests to use the current root-state, joint-state, contact-sensor, and wrench-composer API names.
• Updated the Newton contact sensor adapter to use the current SensorContact constructor and force/metadata fields.
• Bumped matching PhysX and Newton extension changelog/version files.

Test Plan

• ./isaaclab.sh -p -m py_compile source/isaaclab_physx/test/sensors/test_frame_transformer.py source/isaaclab_newton/test/sensors/test_frame_transformer.py source/isaaclab_physx/test/sensors/test_contact_sensor.py source/isaaclab_newton/isaaclab_newton/physics/newton_manager.py source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor.py source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor_kernels.py
• ./isaaclab.sh -f
• Focused deprecation scan: 118 matches on origin/develop, 0 matches on this branch
• Targeted GPU pytest on NvidiaWorkstation-WiFi: attempted in isaac-lab-base-pr5304:latest, but the PhysX container timed out after 3600s during pytest collection before tests ran

[greptile-apps]

Greptile Summary

This PR fixes deprecation warnings across both PhysX and Newton backends by updating call sites to use current API names for root state, joint state, wrench composer fields, and the Newton SensorContact interface. The most substantive change is in the Newton contact sensor, which migrates from a single 3D net_force array (where column 0 held the total and columns 1+ held per-filter forces) to two separate arrays (total_force 1D and force_matrix 2D), including a new flatten_metadata helper and a corresponding kernel rewrite.

Confidence Score: 4/5

Safe to merge; the API migration is consistent and the kernel logic is correct, though GPU tests timed out before running.

No P0 or P1 findings. All renamed fields and method calls are applied uniformly across both backends and their tests. The Newton contact sensor refactor correctly separates the old single 3D force array into distinct total_force and force_matrix handles, with matching kernel and buffer-initialization logic. The only caveat is that targeted GPU pytest timed out during collection, so runtime correctness of the new sensor data paths was not directly validated.

source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor.py and contact_sensor_kernels.py contain the substantive logic change and merit the most scrutiny when GPU tests are available.

Important Files Changed

Filename	Overview
source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor.py	Migrates from old Newton SensorContact API (shape, sensing_objs, net_force) to new split API (total_force, sensing_obj_idx/type, force_matrix); adds flatten_metadata helper and int(idx) guard for numpy index safety
source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor_kernels.py	Kernel updated to accept separate total_force (1D) and force_matrix (2D) arrays instead of a single 3D array; indexing logic is consistent with the new flat layout
source/isaaclab_newton/isaaclab_newton/physics/newton_manager.py	Single rename: include_total=True → measure_total=True in contact sensor constructor call
source/isaaclab_physx/test/sensors/test_contact_sensor.py	Replaces write_root_pose_to_sim with write_root_pose_to_sim_index; renames num_bodies to num_sensors in friction/contact position test helpers
source/isaaclab_newton/test/sensors/test_frame_transformer.py	Replaces deprecated default_root_state, write_root_pose_to_sim, write_joint_state_to_sim, and set_joint_position_target with their _index suffixed counterparts throughout all test scenarios
source/isaaclab_physx/test/sensors/test_frame_transformer.py	Mirrors the Newton frame transformer test updates: deprecated root-state and joint-state write APIs replaced with current _index variants

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[NewtonManager.add_contact_sensor\nmeasure_total=True] --> B[contact_view]

    B --> C[total_force\nshape: n_envs x n_sensors]
    B --> D[force_matrix\nshape: n_envs x n_sensors, n_filter_objects\nor None]
    B --> E[sensing_obj_idx / sensing_obj_type]
    B --> F[counterpart_indices / counterpart_type]

    E --> H[flatten_metadata -> _sensor_names]
    F --> I[flatten_metadata -> _filter_object_names]

    C --> J[copy_from_newton_kernel\ndim = n_envs x n_sensors x max filter_objs,1]
    D --> J

    J --> K[net_force_total 2D\nn_envs, n_sensors]
    J --> L[force_matrix_w 3D\nn_envs, n_sensors, n_filter_objects\nor None]

Loading

_{Reviews (1): Last reviewed commit: "Merge branch 'develop' into antoiner/bac..." | Re-trigger Greptile}

[isaaclab-review-bot]

🤖 Isaac Lab Review Bot

Summary

This PR fixes deprecation warnings by updating call sites in PhysX and Newton backend tests to use current API names for root-state, joint-state, contact-sensor, and wrench-composer methods. It also updates the Newton contact sensor adapter to use the current SensorContact constructor and force/metadata fields. The changes are mechanically correct and follow the established deprecation migration patterns.

Architecture Impact

Self-contained. The changes are limited to:

1. Test files updating deprecated method calls to their non-deprecated equivalents
2. Newton contact sensor adapter updating to use new Newton SensorContact API fields
3. Changelog and version bumps for both isaaclab_newton and isaaclab_physx

The core framework APIs remain unchanged; this PR only migrates consumers to the current API.

Implementation Verdict

Ship it — with one minor clarification suggestion.

Test Coverage

The PR modifies test files to use non-deprecated APIs, which is appropriate. The author reports running a focused deprecation scan showing 118 matches on origin/develop and 0 matches on the branch, confirming comprehensive cleanup. CI shows isaaclab_newton and isaaclab_physx tests passing. The GPU pytest timeout noted in the test plan appears to be infrastructure-related, not a code issue.

CI Status

✅ isaaclab_newton: success
✅ isaaclab_physx: success
✅ pre-commit: success
✅ license-check: success
⏳ Other checks skipped (expected for scoped changes)

Findings

🔵 Improvement: source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor.py:371-377 — Defensive handling of empty counterpart metadata

The flatten_metadata function handles nested structures well, but when self.contact_view.counterpart_indices and self.contact_view.counterpart_type have mismatched lengths, the zip() will silently truncate. Consider adding a length assertion:

flat_counterpart_idx = flatten_metadata(self.contact_view.counterpart_indices)
flat_counterpart_type = flatten_metadata(self.contact_view.counterpart_type)
assert len(flat_counterpart_idx) == len(flat_counterpart_type), "Counterpart metadata length mismatch"
flat_counterparts = list(zip(flat_counterpart_idx, flat_counterpart_type))

This would surface data corruption issues earlier rather than producing subtly wrong filter names.

🔵 Improvement: source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor.py:380-381 — Clarify force_matrix shape handling

The code correctly handles force_matrix being None, but the shape extraction logic could be clearer:

force_matrix_shape = force_matrix.shape if force_matrix is not None else (total_sensor_count, 0)
self._num_filter_objects = force_matrix_shape[1] if len(force_matrix_shape) > 1 else 0

If force_matrix is a 1D array (unexpected), len(force_matrix_shape) > 1 would be False, resulting in _num_filter_objects = 0. This is likely the desired behavior, but a comment explaining the expected shapes would improve maintainability.

🔵 Improvement: source/isaaclab_newton/docs/CHANGELOG.rst:4 — Changelog date format

The changelog entry uses 2026-04-28 which appears to be a future date (or placeholder). Verify this is intentional or update to the actual release date.

🔵 Improvement: source/isaaclab_physx/test/sensors/test_contact_sensor.py:750-762 — Consistent naming after API migration

Good catch updating num_bodies → num_sensors in the friction test helper. The variable rename correctly reflects the semantic meaning after the API migration. However, the loop variable body_idx → sensor_idx rename is incomplete in the comment on line 760 which still references the old name pattern implicitly. Consider updating the loop structure comment for consistency.

🔵 Improvement: source/isaaclab_newton/isaaclab_newton/sensors/contact_sensor/contact_sensor_kernels.py:18-19 — Kernel signature documentation

The kernel comment says (n_envs * n_sensors) for newton_total_force but the launch dimension comment says dim=(num_envs, num_sensors, max(num_filter_objects, 1)). These are consistent (flat vs 3D launch), but adding a note that the flat array is indexed via env * num_sensors + sensor would help readers understand the mapping immediately.