[Newton] Drop explicit mujoco/mujoco-warp pins, defer to newton[sim]

[hujc7]

Summary

• mujoco and mujoco-warp are already declared under Newton's [sim] optional-dependency extra. Re-listing them in IsaacLab's setup.py was redundant and pinned them at IsaacLab's chosen versions instead of Newton's.
• The pins also lived in source/isaaclab/setup.py's top-level INSTALL_REQUIRES, so users installing only the PhysX or Kit backends were forced to pull MuJoCo even though nothing in isaaclab core imports it.
• Switch the Newton spec to newton[sim] @ git+... and remove the direct pins from both setup.py files plus the wheel builder manifest. Newton becomes the single source of truth for those versions.

Verification

• grep -rn "import mujoco\|import mujoco_warp\|MjModel\|MjData" across the repo: zero direct usages. The Newton backend uses newton.solvers.SolverMuJoCo (transitive). MJCF asset import goes through Isaac Sim's isaacsim.asset.importer.mjcf, not the mujoco Python package.
• Newton's pyproject.toml [sim] extra: mujoco~=3.8.0, mujoco-warp>=3.8.0.1,~=3.8.0. Newton's compatible-release spec already covers the patch versions IsaacLab was hard-pinning.
• All other install pathways (./isaaclab.sh -i, cli/commands/install.py, docker/Dockerfile.*, environment.yml, root pyproject.toml, per-package pyproject.toml) read deps from these three files; no other version pins exist.
• ./isaaclab.sh -f passes.

Test plan

• Fresh ./isaaclab.sh -i --install newton install resolves Newton, mujoco, and mujoco-warp transitively.
• Smoke-test a Newton-backed task to confirm the MuJoCo Warp solver still loads.
• Wheel build with tools/wheel_builder produces wheels with the same set of MuJoCo packages as before.

[isaaclab-review-bot]

🤖 Isaac Lab Review Bot

Summary

Removes redundant mujoco and mujoco-warp version pins from isaaclab core and isaaclab_newton, deferring to Newton's [sim] extra as the single source of truth for these transitive dependencies. Users installing only PhysX/Kit backends no longer pull in MuJoCo unnecessarily.

Design Assessment

Design is sound. This is the correct architectural decision — isaaclab core doesn't import MuJoCo anywhere (grep -rn "import mujoco" = zero hits), so it shouldn't declare it as a dependency. Newton already pins the compatible versions in its [sim] extra, so newton[sim] @ git+... is the right specifier for isaaclab_newton.

Findings

No issues found. The change is minimal, correct, and well-motivated:

• ✅ isaaclab/setup.py: Removes two lines that forced MuJoCo on PhysX/Kit users
• ✅ isaaclab_newton/setup.py: Switches from direct pins to newton[sim] (single source of truth)
• ✅ tools/wheel_builder/res/python_packages.toml: Same cleanup for the wheel builder
• ✅ Changelogs properly categorized (Removed for isaaclab, Changed for isaaclab_newton)

Test Coverage

No new tests needed — this is a build/dependency configuration change with no runtime code impact. The existing CI Build Wheel check passing confirms the dependency resolution still works.

CI Status

• Build Wheel: ✅ pass
• Check changelog fragments: ✅ pass
• pre-commit: ✅ pass
• Check for Broken Links: ❌ fail (likely unrelated)

Verdict

Ship it ✅

Clean dependency hygiene fix. Single source of truth for MuJoCo versions, no unnecessary transitive deps for PhysX/Kit users.

[greptile-apps]

Greptile Summary

This PR removes redundant mujoco==3.8.0 and mujoco-warp==3.8.0.1 pins from isaaclab core and isaaclab_newton, and upgrades all Newton references from bare newton @ git+... to newton[sim] @ git+... so that MuJoCo packages are resolved transitively through Newton's own [sim] extra.

• source/isaaclab/setup.py: Removes mujoco and mujoco-warp from top-level INSTALL_REQUIRES; neither package is directly imported by isaaclab core, eliminating a spurious install-time dep for PhysX/Kit-only users.
• source/isaaclab_newton/setup.py, source/isaaclab_physx/setup.py, source/isaaclab_visualizers/setup.py: All Newton extras switch to newton[sim] to prevent pip's URL-based deduplication from stripping the [sim] extras when multiple packages in the same environment declare the same VCS URL.
• tools/wheel_builder/res/python_packages.toml: Explicit mujoco/mujoco-warp entries dropped from the newton group; the generated pyproject.toml will carry only newton[sim] @ git+... and rely on pip for transitive resolution at install time.

Confidence Score: 4/5

Safe to merge; the change is a straightforward dependency-management cleanup with no functional code changes and a clear rationale for every decision.

The core removal of mujoco from isaaclab's top-level deps is clean and correct. The switch to newton[sim] everywhere is logically sound given pip's URL-deduplication behavior. Two minor concerns are worth a second look: the version constraint on mujoco loosens from an exact pin to a compatible-release range, so future patch releases could be picked up; and the rerun/viser extras now carry a full mujoco install that those visualization backends don't actually use.

source/isaaclab_visualizers/setup.py — the rerun and viser extras now transitively install mujoco/mujoco-warp; worth confirming this is intentional for all user-facing install scenarios.

Important Files Changed

Filename	Overview
source/isaaclab/setup.py	Removes mujoco==3.8.0 and mujoco-warp==3.8.0.1 from top-level INSTALL_REQUIRES; no isaaclab core code imports these packages directly.
source/isaaclab_newton/setup.py	Removes explicit mujoco/mujoco-warp pins from the "all" extra and upgrades newton to newton[sim]; mujoco resolution is now deferred to Newton's compatible-release constraint (~=3.8.0) rather than the former exact pin (==3.8.0).
source/isaaclab_physx/setup.py	Upgrades the newton extra from bare newton to newton[sim]; PhysX users who enable the newton add-on will now transitively install mujoco and mujoco-warp.
source/isaaclab_visualizers/setup.py	All three newton-bearing extras (newton, rerun, viser) now use newton[sim]; rerun- and viser-only installs will now pull in mujoco and mujoco-warp even though those backends don't use the MuJoCo solver.
tools/wheel_builder/res/python_packages.toml	Explicit mujoco and mujoco-warp entries removed from the newton optional group; the wheel's generated pyproject.toml will list only newton[sim] @ git+..., relying on pip to resolve mujoco transitively at install time.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["pip install isaaclab_newton[all]"] --> B["newton[sim] @ git+...@v1.2.0rc2"]
    B --> C["Newton package\n(physics engine)"]
    B --> D["[sim] extras\nmujoco~=3.8.0\nmujoco-warp>=3.8.0.1,~=3.8.0"]
    D --> E["mujoco (latest 3.8.x)"]
    D --> F["mujoco-warp (latest 3.8.x)"]
    G["pip install isaaclab_physx[newton]"] --> B
    H["pip install isaaclab_visualizers[newton/rerun/viser]"] --> B
    I["pip install isaaclab (core only)"] --> J["No mujoco/mujoco-warp Installed"]
    style D fill:#f9f,stroke:#333
    style I fill:#9f9,stroke:#333
    style J fill:#9f9,stroke:#333

Loading

_{Reviews (1): Last reviewed commit: "Use newton[sim] in all Newton declaratio..." | Re-trigger Greptile}

[greptile-apps]

Loosened mujoco/mujoco-warp version constraints

The previous explicit pins (mujoco==3.8.0, mujoco-warp==3.8.0.1) guaranteed exact reproducibility. Newton's [sim] extra specifies mujoco~=3.8.0 (compatible-release, i.e. >=3.8.0,<3.9.0) and mujoco-warp>=3.8.0.1,~=3.8.0, so a fresh install after a new 3.8.x release could resolve a different patch version than was tested. Since Newton is pinned to an immutable tag (v1.2.0rc2) the constraint itself won't change, but pip will still pick the latest matching patch from PyPI at install time. Worth confirming the team is comfortable trading bit-for-bit reproducibility for deferred version management.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[hujc7]

By design — the goal of this change is for Newton to be the single source of truth for these versions, so the loosening from ==3.8.0 to ~=3.8.0 is the trade-off we accept (called out in the PR description). If a specific 3.8.x patch turns out to be incompatible, the right fix is to tighten Newton's own [sim] spec rather than override it here.

[isaaclab-review-bot]

Makes sense — single source of truth in Newton with ~= as the compatibility range is a clean approach. Agreed that tightening upstream is the right lever if a patch breaks things.

[greptile-apps]

rerun and viser extras now force-install MuJoCo

Before this PR the rerun and viser extras pulled in bare newton (no MuJoCo). After, they request newton[sim], so users who install only isaaclab_visualizers[rerun] or isaaclab_visualizers[viser] will now also download mujoco and mujoco-warp — packages those visualization backends do not use directly. The PR description gives a valid pip-deduplication rationale (a bare newton @ git+URL elsewhere could shadow the [sim] extras), but it is worth documenting in the changelog or a comment why visualization-only extras now carry the full MuJoCo install.

[hujc7]

Added a block comment above EXTRAS_REQUIRE explaining why even the rerun/viser extras pin newton[sim]. tl;dr: pip resolves a git-URL requirement once per URL, so a bare newton @ git+... here would shadow the [sim] extra requested by isaaclab_newton and silently break installs. Forcing [sim] everywhere is what keeps the resolution consistent — the cost is that rerun/viser-only installs carry a mujoco install they don't use, which we accept.

[isaaclab-review-bot]

TIL about pip deduplicating git-URL requirements — that explains the necessity well. The added comment should save future readers the same question. 👍

[isaaclab-review-bot]

Update (commit 1c394ff): Reviewed incremental changes since de295cc.

New changes:

• ✅ Dexsuite config refactoring: Newton/mesh-based shapes, consolidated events, lifted multi-asset restrictions
• ✅ Carb null-client fix in _fabric_notices.py and _cubric.py
• ✅ CI image pinned with SHA256 digest
• ✅ Changelog entries added
• ✅ Test timeout increase for test_articulation.py

Previous suggestions: Both were acknowledged (version constraints rationale and newton[sim] extras for pip deduplication).

Result: No new issues. LGTM 👍

[isaaclab-review-bot]

Re-review (1c394ff): New commits look good ✅

• Carb null-client fixes address a real issue (properly documented in changelogs)
• Mesh-based shape refactoring enables Newton heterogeneous object spawning
• CI image tag pinned with sha256 digest
• Previous concerns about version constraints were acknowledged — design decision accepted

LGTM 👍

[isaaclab-review-bot]

📋 Re-review after merge from develop (e35394a)

New commits: Merge commit brings in develop changes:

• Changelog compilation
• Newton contact sensor fix ([Newton] Fixes contact sensor metadata access #5588)
• OVPhysX RigidObject asset ([OVPHYSX] RigidObject + RigidObjectData asset #5426)

These are unrelated to this PR's scope.

PR changes (unchanged): Removes redundant mujoco/mujoco-warp pins from IsaacLab setup.py files, deferring to newton[sim] extra.

✅ No new concerns from the merge.

[isaaclab-review-bot]

🔄 Re-Review Update (e35394a)

Reviewed incremental changes from 1c394ff to e35394a8.

The new commits are merge-from-develop bringing upstream changes:

• b8fead49: CI auto version bump
• 5ea75124: Newton contact sensor fix ([Newton] Fixes contact sensor metadata access #5588)
• 68a651f7: OVPhysX RigidObject implementation ([OVPHYSX] RigidObject + RigidObjectData asset #5426)
• e35394a8: Merge commit

These are unrelated to the PR's stated goal (dropping mujoco pins). The actual PR changes remain as previously reviewed.

Previous concerns status:

• ✅ Loosened mujoco version constraints: Acknowledged — Newton's ~= range is intentional; tightening upstream is the right lever if issues arise.
• ✅ Visualization extras pulling MuJoCo: Acknowledged — pip deduplication rationale documented.

No new issues introduced by the merge.