[StafaH]

Description

The reset_hand kernel in the experimental Warp reorientation env sampled hand joint reset positions around the lower joint limit instead of between the lower and upper limits:

dof_pos_noise = wp.randf(rng_state[env_id], -1.0, 1.0)
rand_delta = delta_min + (delta_max - delta_min) * 0.5 * dof_pos_noise  # centred on delta_min

As a result, every hand joint reset at or below its default position. With a large enough reset_dof_pos_noise it could also end up outside the joint limits. For example, with limits [-1, 1], default 0.1 and noise scale 0.2, resets fell in [-0.32, 0.08] instead of [-0.12, 0.28].

The torch reorient and handover tasks were already fixed in #6413 via sample_joint_positions_within_limits. This PR applies the same [-1, 1] -> [0, 1] mapping in the Warp kernel, so both versions of the task reset joints over the same range. This was the last copy of the formula from the issue on develop.

Fixes #4404

Type of change

• Bug fix (non-breaking change which fixes an issue)

Test

I added test_warp_reset_hand_samples_the_same_range_as_the_torch_helper to the existing source/isaaclab_tasks/test/core/test_reorient_utils.py. It launches reset_hand on CPU for 512 envs and checks two things:

• every sample stays inside the range that sample_joint_positions_within_limits produces;
• each joint gets samples on both sides of its default.

The test fails without the fix and passes with it (uv run python -m pytest source/isaaclab_tasks/test/core/test_reorient_utils.py: 12 passed).

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 added a changelog fragment under source/<pkg>/changelog.d/ for every touched package (do not edit CHANGELOG.rst or bump extension.toml — CI handles that)

• I have added my name to the CONTRIBUTORS.md or my name already exists there

• Backport this pull request to the active release branch after it merges into develop

[StafaH]

run-ci

[isaaclab-review-bot]

Isaac Lab Review Bot

The patch minimally corrects Warp hand-reset sampling by mapping the random value from [-1, 1] to [0, 1] before interpolation, restoring the intended range between lower and upper joint-limit deltas. Package-specific changelog fragments and focused regression coverage are included.

• Design and architecture: The Warp and torch reset paths are now behaviorally aligned without introducing a new abstraction or dependency. The regression test in the stable task suite imports the experimental Warp environment, which slightly couples those test surfaces but does not warrant pre-merge action for this cross-implementation parity check.
• API: No public signature, kernel argument order, dtype, or export changes. The only externally observable change is the intended correction to the sampled joint-position distribution.
• Implementation: The revised expression maps dof_pos_noise to [0, 1], so rand_delta spans [delta_min, delta_max] and the configured noise scale interpolates from the default position toward either joint limit. RNG advancement, velocity sampling, masking, and target initialization remain unchanged. The test checks both range containment and sampling on each side of the default.

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.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge, with the corrected interpolation matching the established Torch behavior and adequate regression coverage.

Summary

The PR corrects Warp hand-reset sampling by mapping random noise from [-1, 1] to the full interpolation interval between lower and upper joint-limit deltas.

• Aligns the Warp reset formula with the existing Torch helper.
• Adds a CPU regression test covering limit bounds and samples on both sides of each default position.
• Adds changelog fragments for the affected packages.

_{Reviews (1) · Last reviewed commit: "Fix Warp reorient reset sampling around ..."}

[ooctipus]

run-ci

[StafaH]

run-ci

[isaaclab-bot]

Backported to release/3.0.0 as 3d44066.