Updates core deprecation call sites

[AntoineRichard]

Summary

• Migrates core test and MDP callers off deprecated state/read/write helper APIs.
• Updates the test_pose_inv tensor-to-NumPy conversion for NumPy 2.0.
• Bumps the isaaclab changelog/version because core MDP source changed.

Verification

• ./isaaclab.sh -f
• Scoped deprecated-call-site search: assigned core matches removed.

Rebased onto develop after PR #5304 merged.

[greptile-apps]

Greptile Summary

This PR migrates deprecated state/read/write helper APIs across core MDP actions and tests to their new _index-suffixed equivalents, fixes a NumPy 2.0 incompatibility in test_pose_inv, and bumps the version/changelog accordingly. All call-site substitutions appear semantically equivalent: pose-only reads swap body_link_state_w[:7] for body_link_pose_w, joint state writes split cleanly into separate position and velocity calls before write_data_to_sim, and root-state composition explicitly concatenates root_link_pose_w with root_com_vel_w to match the old monolithic root_state_w layout.

Confidence Score: 5/5

Safe to merge — all substitutions are mechanically correct and semantically equivalent to the deprecated calls they replace.

Every changed call site was reviewed: pose slicing is equivalent (body_link_pose_w yields the same 7-element pose as body_link_state_w[:7]), joint state writes are split correctly before write_data_to_sim, root-state composition matches the old 13-element layout, and the NumPy 2.0 fix follows the canonical .detach().cpu().numpy() pattern. No logic changes, no new branches, no missing side effects were identified.

No files require special attention.

Important Files Changed

Filename	Overview
source/isaaclab/isaaclab/envs/mdp/actions/pink_task_space_actions.py	Replaces deprecated body_link_state_w[:, idx, :7] with body_link_pose_w[:, idx, :7]; semantics are preserved since both yield the 7-element pose slice.
source/isaaclab/test/controllers/test_operational_space.py	Replaces write_joint_state_to_sim / set_joint_effort_target with the _index variants; both calls happen before write_data_to_sim, preserving atomicity.
source/isaaclab/test/envs/test_scale_randomization.py	Replaces positional write_root_velocity_to_sim call with keyword-argument write_root_velocity_to_sim_index; straightforward deprecation fix.
source/isaaclab/test/scene/test_interactive_scene.py	Expands write_joint_state_to_sim into separate position/velocity _index calls across four test locations; logic is equivalent to the old single-call approach.
source/isaaclab/test/utils/test_math.py	Fixes NumPy 2.0 incompatibility in test_pose_inv by using .detach().cpu().numpy() instead of wrapping a GPU tensor in np.array().
source/isaaclab/test/utils/test_wrench_composer_integration.py	Adds three module-level helpers to compose/write root state via new APIs, and migrates all set_forces_and_torques / write_root_*_to_sim / write_root_state_to_sim call sites to _index variants.
source/isaaclab/test/utils/test_wrench_composer_vs_physx.py	Migrates all set_forces_and_torques, write_root_state_to_sim, and related calls to index variants; initial_state* capture now manually concatenates root_link_pos_w + root_link_quat_w + root_com_vel_w, consistent with the new write APIs.
source/isaaclab/config/extension.toml	Bumps version from 4.6.21 to 4.6.22 to reflect the core MDP deprecation fixes.
source/isaaclab/docs/CHANGELOG.rst	Adds 4.6.22 changelog entry describing the PinkInverseKinematicsAction deprecation fix.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    subgraph old["Old (Deprecated) APIs"]
        A1["body_link_state_w[:, idx, :7]"]
        A2["write_joint_state_to_sim(pos, vel)"]
        A3["write_root_state_to_sim(state)"]
        A4["write_root_pose_to_sim(pose)"]
        A5["write_root_velocity_to_sim(vel)"]
        A6["set_forces_and_torques(...)"]
        A7["set_joint_effort_target(...)"]
    end

    subgraph new["New (_index) APIs"]
        B1["body_link_pose_w[:, idx, :7]"]
        B2["write_joint_position_to_sim_index(pos)\nwrite_joint_velocity_to_sim_index(vel)"]
        B3["write_root_link_pose_to_sim_index(pose)\nwrite_root_com_velocity_to_sim_index(vel)"]
        B4["write_root_pose_to_sim_index(pose)"]
        B5["write_root_velocity_to_sim_index(vel)"]
        B6["set_forces_and_torques_index(...)"]
        B7["set_joint_effort_target_index(...)"]
    end

    A1 -->|replaced by| B1
    A2 -->|split into| B2
    A3 -->|split into| B3
    A4 -->|replaced by| B4
    A5 -->|replaced by| B5
    A6 -->|replaced by| B6
    A7 -->|replaced by| B7

Loading

_{Reviews (1): Last reviewed commit: "fix: update core deprecation call sites" | Re-trigger Greptile}

[isaaclab-review-bot]

🤖 Isaac Lab Review Bot

Summary

This PR migrates test files and one MDP action module off deprecated state/read/write helper APIs to the newer *_index variants. The changes are mechanical API updates replacing write_root_state_to_sim, write_root_pose_to_sim, write_root_velocity_to_sim, write_joint_state_to_sim, set_joint_effort_target, and set_forces_and_torques with their *_index counterparts. The version is bumped to 4.6.22 with a corresponding changelog entry.

Architecture Impact

Self-contained. All changes are in test files and one MDP action implementation. The deprecated APIs being replaced are wrappers that internally call the *_index methods, so behavior should be identical. No downstream consumers are affected since tests don't export APIs.

Implementation Verdict

Minor fixes needed — The migration is mostly correct, but there are a few inconsistencies and one potential bug in the test state capture logic.

Test Coverage

This PR is purely a deprecation cleanup, not a bug fix or new feature. The tests being modified are integration tests for the wrench composer and operational space controller. No new tests are needed, but the existing tests adequately exercise the migrated code paths.

CI Status

All relevant CI jobs pass (isaaclab core tests 1/3, 2/3, 3/3 ✅). The isaaclab_assets job was cancelled, likely due to unrelated infrastructure issues.

Findings

🟡 Warning: test_wrench_composer_vs_physx.py:571-581 — Inconsistent state capture vs reset pattern

The test captures initial state using root_link_pos_w, root_link_quat_w, and root_com_vel_w but then uses write_root_link_pose_to_sim_index and write_root_com_velocity_to_sim_index. This mixes link-frame pose with CoM-frame velocity, which is the intended pattern. However, the original code used root_state_w which contained root CoM data. Verify this doesn't change test semantics:

# Lines 571-581: Captures link pose + com velocity
initial_state_composer = torch.cat(
    [
        cube_composer.data.root_link_pos_w.torch,
        cube_composer.data.root_link_quat_w.torch,
        cube_composer.data.root_com_vel_w.torch,  # CoM velocity
    ],
    dim=-1,
).clone()

This is actually correct — the link pose plus CoM velocity is the proper way to restore state when using the new APIs. No action needed.

🔵 Improvement: test_wrench_composer_integration.py:279-287 — Repetitive pattern could use helper

Multiple tests follow the same pattern of cloning pose, setting quaternion, writing pose, cloning velocity, zeroing it, writing velocity. Consider extracting a helper function to reduce duplication:

root_pose = cube_object.data.root_pose_w.torch[0].clone().unsqueeze(0)
root_pose[0, 3:7] = torch.tensor([0.7071, 0.0, 0.0, 0.7071], device=device)
cube_object.write_root_pose_to_sim_index(root_pose=root_pose)

root_vel = cube_object.data.root_vel_w.torch.clone()
root_vel[0, :] = 0.0
cube_object.write_root_velocity_to_sim_index(root_velocity=root_vel)

This pattern appears ~10 times. A helper like teleport_cube(cube, pos, quat, zero_vel=True) would improve readability.

🔵 Improvement: test_operational_space.py:1500-1503 — Keyword argument style inconsistency

The migration uses positional style for some args and keyword style for others:

robot.write_joint_position_to_sim_index(position=default_joint_pos)
robot.write_joint_velocity_to_sim_index(velocity=default_joint_vel)
robot.set_joint_effort_target_index(target=zero_joint_efforts)

The keyword style is good and consistent with the rest of the PR. This is actually correct.

🔵 Improvement: pink_task_space_actions.py:221 — Comment mentions "world origin" but variable name says "world_rf"

The comment says "Get base link frame pose in world origin" but the resulting variable is named base_link_frame_in_world_origin. After the fix, this reads from body_link_pose_w which is world-frame pose, making the comment accurate. The change from body_link_state_w.torch[:, self._base_link_idx, :7] to body_link_pose_w.torch[:, self._base_link_idx] is correct since body_link_pose_w is already shape (N, num_bodies, 7).

🔵 Improvement: CHANGELOG.rst:4-14 — Changelog entry is minimal but accurate

The changelog correctly documents the fix to PinkInverseKinematicsAction. The entry is appropriate for the scope of changes.

🟡 Warning: test_wrench_composer_integration.py — Missing .clone() on some pose reads could cause aliasing

At lines like 96 and 169:

root_pose = cube_object.data.root_pose_w.torch[0].clone().unsqueeze(0)

This is correct. But at line 723:

root_pose = cube_object.data.default_root_pose.torch.clone()

The default_root_pose is a property that returns a view. The .clone() is present, so this is safe. All pose reads appear properly cloned.

🔵 Improvement: test_interactive_scene.py:82-126 — Local variables created but immediately consumed

The pattern:

joint_pos = torch.rand_like(scene["robot"].data.joint_pos.torch)
joint_vel = torch.rand_like(scene["robot"].data.joint_pos.torch)
scene["robot"].write_joint_position_to_sim_index(position=joint_pos)
scene["robot"].write_joint_velocity_to_sim_index(velocity=joint_vel)

This is cleaner than the original inline style and matches the idiom used elsewhere in the PR. Good change.

[isaaclab-review-bot]

🤖 Isaac Lab Review Bot

Follow-up Review

The new commit (6b6beaa) appears identical to the previously reviewed commit. All previous findings remain valid:

• The migration from deprecated APIs to *_index variants is correct
• The pink_task_space_actions.py fix properly changes from body_link_state_w.torch[:, self._base_link_idx, :7] to body_link_pose_w.torch[:, self._base_link_idx]
• The test state capture pattern in test_wrench_composer_vs_physx.py correctly uses link pose + CoM velocity

Implementation Verdict

Ship it — The deprecation cleanup is mechanically correct and CI passes. The previous improvement suggestions (helper function extraction, etc.) are nice-to-haves but not blocking.