[hujc7]

Summary

Id and mask selections now resolve through two shared helpers, resolve_ids and resolve_1d_mask, instead of
about 30 hand-written copies across the Newton, OVPhysX and PhysX assets, sensors and the wrench composer
(−1086 / +612 source lines). Newton mask writers that converted masks to ids now write through the mask: 4-11x faster.

Description

1. Changes

1. Shared helpers (isaaclab/utils/warp/utils.py)
   • New resolve_ids: ids or a boolean mask → ids on a device.
   • resolve_1d_mask: scratch_mask is optional for callers that never pass ids; a Warp mask on another device is
     copied; a boolean torch mask already on the device is accepted inside a CUDA graph capture.
   • isaaclab_experimental.utils.warp.resolve_1d_mask is now this function instead of a copy.
2. Callers routed through the helpers: the Newton, OVPhysX and PhysX assets, SensorBase, Camera and
   WrenchComposer. Every selection now accepts the same inputs: any integer sequence, slices with a negative step,
   ids on another device (moved to the owner's device). A ProxyArray raises TypeError.
3. Faster Newton write paths (see 2)
   • The RigidObjectCollection mask writers and the fixed-tendon mask setters write through the masks instead of
     converting them to ids, which waited for the GPU on every call.
   • The joint-coordinate writes reuse a scratch mask instead of allocating one per call.
   • One generic masked 2-D write kernel, with a cached overload per dtype, replaces 2 copies.
4. Fixes
   • The deformable write_nodal_*_to_sim_mask writers raised a RuntimeError whenever env_mask was given
     (BaseDeformableObject, so OVPhysX, and PhysX).
   • The Newton fixed-tendon mask setters accepted data whose shape does not match the articulation.

2. Performance

Desk RTX PRO 6000, Newton, main vs this PR. Micro: contract-test Newton assets, 4096 envs, host µs per call
including the GPU wait, median of 7x300 calls, 2 alternated runs per side.

Write path	Selection	main (µs)	this PR (µs)	Speed-up
Fixed-tendon stiffness mask setter	half envs, 2 tendons	201-208	18.5-19.0	11x
Fixed-tendon position limit mask setter	half envs, 2 tendons / all	185-194 / 80-86	18.3 / 17.5	10x / 4.7x
write_fixed_tendon_properties_to_sim_mask	half envs, 2 tendons / all	263-272 / 147-154	26.0-26.8 / 25.2-25.8	10x / 5.9x
Collection link / com pose mask writers	half envs, 2 bodies	165-175	32.6-38.9	4.5-5x
Collection link / com velocity mask writers	half envs, 2 bodies	168-188	34.8-47.8	3.9-4.7x
write_joint_position_to_sim_index / write_joint_state_to_sim_index	half envs	70-72	47.9-51.8	1.4-1.5x

End to end, isaaclab benchmark runtime ... presets=newton_mjwarp, 3 alternated runs per side, total FPS:

Task	main	this PR	Change
Isaac-Cartpole-Direct (3000 steps)	2841-2851K	2985-3011K	+5.4%
Isaac-Reorient-Cube-Shadow-Direct (1000 steps, 8192 envs)	495.7-495.8K	498.0-501.4K	+0.9%
Isaac-Velocity-Flat-UnitreeGo2 (1000 steps)			-0.6%, within noise

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)

Checklist

Docker and GPU tests run on demand. Push the commits you want tested, then
comment run-ci on the pull request.

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with uv run isaaclab --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have validated the changed behavior with focused coverage and applied the test-audit criteria to any test changes
• I have added a changelog fragment under source/<pkg>/changelog.d/ for every touched package (do not edit CHANGELOG.rst or bump package versions — CI handles that)
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[hujc7]

run-ci

[hujc7]

run-ci

[hujc7]

run-ci