Bump Newton to include torque routing

[AntoineRichard]

Description

Pins every active newton[sim] dependency to Newton commit
2064e3b79807dcc1679d1eb86ef7efd9ef0f28ee, the squash-merge commit for
Newton PR #3134. The
update also changes the core IsaacLab Warp requirement from 1.14.0 to the
prerelease build 1.15.0.dev20260626.

This branch is based on the reviewed origin/develop snapshot
e5478465baf58f735ba308e40faf6b4eb2993d29.

Upstream comparison from the prior pin to the new pin:
newton-physics/newton@49ca73b...2064e3b

Compatibility updates

• Reinterprets seven PhysX/OVPhysX COM staging sources as wp.float32 views
  before copying, matching Warp 1.15 element-size validation.
• Forwards model.shape_margin through the MJWarp/VBD and Featherstone/VBD
  reaction paths, with regression coverage for both solvers.
• Migrates Kamino reset to SolverKamino.ResetConfig.from_joints() while
  keeping the reset migration minimal.

IsaacLab adds no custom Kamino torque-routing path. Newton PR #3134 consumes
the generic Control.joint_f effort input upstream. IsaacLab retains only its
Kamino lifecycle, reset, forward-kinematics, collision, and graph
specialization.

Local verification

Exact local environment: Newton
2064e3b79807dcc1679d1eb86ef7efd9ef0f28ee, Warp
1.15.0.dev20260626.

• Wheel metadata: 3 passed.
• Newton manager abstraction: 72 passed across both Kamino modes.
• Exact Kamino Cartpole integration smoke: 16 environments, 5 iterations,
  exit code 0, no traceback, and Training time: 2.83 seconds. This proves
  runtime/integration compatibility of the generic effort path; it does not
  establish numerical torque effectiveness.
• Two-way coupling: MJWarp 1 passed; Featherstone 1 passed.
• Soft/cloth: 4 passed, 122 deselected.
• Full ./isaaclab.sh -f: passed before and after the final commit.

Local COM coverage gap

The focused kitless COM fallback run produced 76 passed, 520 deselected,
28 setup errors, and 0 skipped. The 28 setup errors are a pre-existing
OVPhysX fixture/API mismatch: mock_bindings.bindings is a dictionary where
the fixture expects an object exposing .view. They are not a failure caused
by this pin, but the affected COM assertions remain a disclosed local coverage
gap. The final-head CI matrix will be authoritative once terminal.

CI result

Current head 6d7b40682a2117a28ff4fb2dff7a8aaadca75b2e, based on reviewed
origin/develop snapshot e5478465baf58f735ba308e40faf6b4eb2993d29:
CI pending for this rebased head. Historical results below are from
pre-rebase head 5b004bd04e79d735324ae64c9d4210f79b77c519 and are retained as
diagnostic context; they are not the final-head result.

On pre-rebase head 5b004bd04e79d735324ae64c9d4210f79b77c519, the
compatibility-focused jobs passed, including isaaclab_newton,
isaaclab_physx, isaaclab_contrib, all core and task shards, both
installation jobs, and isaaclab_ov on unchanged rerun.

Original and rerun evidence

• isaaclab_ov original
  failed the GPU test_body_root_state_properties[True-cuda:0-1] numerical
  velocity comparison after its three built-in attempts: the final maximum
  absolute difference was 0.000190109 for a 0.0001 tolerance. The
  unchanged rerun
  passed, confirming this as known intermittent and non-blocking.
• rendering-correctness-kitless original
  exhausted its built-in attempts for an OVPhysX Cartpole diffuse comparison
  (2.38% different pixels versus a 1.00% limit) and a Newton Dexsuite
  full-MDL comparison (10.37% versus 8.00%). Both original cases passed
  on the unchanged rerun,
  while a different Newton Cartpole normals comparison exhausted its attempts
  (2.39% versus 1.00%, final SSIM 0.9566). An independently triggered
  unchanged attempt 3
  passed. The shifting failures followed by a clean rerun confirm this as a
  known intermittent, non-blocking rendering flake.
• test-curobo original
  was cancelled when its runner received a shutdown signal during OCI export,
  followed by an RPC Unavailable/EOF. The
  unchanged rerun
  passed, confirming an infrastructure-only transient.

The final branch delta from reviewed base
e5478465baf58f735ba308e40faf6b4eb2993d29 touches the OV/PhysX COM staging
sources listed above but does not modify rendering test or baseline files.
Because the reviewed base includes Cartpole lighting and rendering test/golden
updates, final-head rendering CI is an important integration check.

The CI matrix does not make the locally blocked COM assertions pass and does
not close the clean-wheel [newton] installation coverage gap; both remain
documented separately.

Remaining installation coverage gap

Current installation CI does not clean-install a built wheel with the
[newton] extra. End-to-end resolution of the generated Git Newton URL plus
the prerelease Warp build from the NVIDIA index remains untested by that path.

[greptile-apps]

Greptile Summary

This PR bumps the newton[sim] pin from commit 79e95bf to 49ca73b across all four active source packages and the wheel builder, and simultaneously upgrades the Warp pin from warp-lang==1.14.0 (a stable PyPI release) to warp-lang==1.15.0.dev20260626 (a dev prerelease hosted exclusively on the NVIDIA package index). Per the PR description, this is an intentional compatibility probe — no runtime API fixes are included — designed to surface CI failures against the latest Newton main branch.

• All seven Newton commit pins and three Warp pins are updated consistently, and the existing metadata test (test_wheel_builder_metadata.py) verifies this consistency.
• The warp-lang==1.15.0.dev20260626 upgrade moves a core isaaclab dependency from the default PyPI index to the NVIDIA-only index, meaning any install environment without the NVIDIA extra index will fail to resolve the core package.
• A design document is added under the newly created docs/superpowers/specs/ directory.

Confidence Score: 3/5

Not safe to merge as-is: the upgrade from a stable PyPI Warp release to a NVIDIA-index-only dev prerelease breaks standard installs, and the PR explicitly states Kamino runtime calls will fail.

The core isaaclab package now depends on warp-lang==1.15.0.dev20260626, which exists only on the NVIDIA package index. Any environment resolving from PyPI alone fails before reaching runtime. This is different in kind from the previous warp-lang==1.14.0 pin, which was a stable PyPI package. Additionally, known Kamino reset API breakages will cause runtime test failures — the PR description is explicit that CI will find blockers that are not yet fixed.

source/isaaclab/pyproject.toml needs attention for the Warp prerelease pin; source/isaaclab_newton/pyproject.toml would benefit from an explicit Warp pin in its [all] extra.

Important Files Changed

Filename	Overview
source/isaaclab/pyproject.toml	Core warp-lang pin bumped from stable 1.14.0 (on PyPI) to prerelease 1.15.0.dev20260626 (NVIDIA-only index), making isaaclab unresolvable from PyPI without extra index configuration.
tools/wheel_builder/res/python_packages.toml	Both the core Warp pin and the newton-extra Warp/Newton pins updated consistently; matches the source package declarations as validated by the metadata test.
source/isaaclab_newton/pyproject.toml	Newton commit updated in the all extra; no explicit warp-lang pin here (pre-existing pattern), relying on the core isaaclab package to constrain Warp.
source/isaaclab_physx/pyproject.toml	Newton commit updated in the newton extra; single-line change, consistent with other packages.
source/isaaclab_visualizers/pyproject.toml	All four Newton-bearing extras (newton, rerun, viser, all) updated to the new commit; changes are consistent.
docs/superpowers/specs/2026-06-29-newton-latest-pin-design.md	Design document added to the newly created docs/superpowers/specs/ directory — this path has no prior precedent in the repo.
source/isaaclab/changelog.d/pin-newton-49ca73b.rst	Changelog fragment for isaaclab; accurately describes both the Newton commit and the Warp version bump.
source/isaaclab_newton/changelog.d/pin-newton-49ca73b.rst	Changelog fragment; references the correct Newton commit and recommends the matching Warp version.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[pip install isaaclab] --> B{Index configured?}
    B -- PyPI only default --> C[FAIL: warp-lang dev build not on PyPI]
    B -- PyPI + NVIDIA index --> D[Resolves warp-lang==1.15.0.dev20260626]
    D --> E[Install newton at 49ca73b from GitHub]
    E --> F{Runtime tests}
    F -- Kamino solver --> G[FAIL: joint_q/joint_u removed by Newton upstream]
    F -- Other paths --> H[CI reveals remaining blockers]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[pip install isaaclab] --> B{Index configured?}
    B -- PyPI only default --> C[FAIL: warp-lang dev build not on PyPI]
    B -- PyPI + NVIDIA index --> D[Resolves warp-lang==1.15.0.dev20260626]
    D --> E[Install newton at 49ca73b from GitHub]
    E --> F{Runtime tests}
    F -- Kamino solver --> G[FAIL: joint_q/joint_u removed by Newton upstream]
    F -- Other paths --> H[CI reveals remaining blockers]

Loading

Comments Outside Diff (1)

1. source/isaaclab_newton/pyproject.toml, line 25-31 (link)

   warp-lang not explicitly pinned in [all] extra

   isaaclab_newton[all] lists newton[sim] at the new commit but carries no explicit warp-lang==1.15.0.dev20260626 pin. When this package is resolved standalone — without isaaclab pulling in the exact Warp version — pip satisfies Newton's own transitive Warp requirement, which may allow a different dev build. The wheel-builder's newton extra does carry the pin, so the wheel path is covered, but a direct pip install "isaaclab_newton[all]" from source is not constrained. Consider mirroring the explicit Warp pin here.

_{Reviews (1): Last reviewed commit: "Bump Newton to latest main commit" | Re-trigger Greptile}

[greptile-apps]

Dev prerelease blocks standard PyPI installs

warp-lang==1.15.0.dev20260626 is not on PyPI — it is only available from https://pypi.nvidia.com. Unlike the previous warp-lang==1.14.0 (a stable PyPI release), this version requires an explicit --extra-index-url https://pypi.nvidia.com when resolving isaaclab's core dependencies. Any CI environment, Docker image, or downstream user that does pip install isaaclab without the NVIDIA extra index will fail at dependency resolution, even before reaching the known Kamino runtime failures.

[greptile-apps]

Non-standard documentation directory

docs/superpowers/specs/ is created for the first time by this PR and contains only this one file. Standard Isaac Lab design and spec content lives under docs/source/. An isolated new top-level subtree with no cross-references from the rendered docs site will be hard to discover and may accumulate stale specs over time. If this is intentional, a README or index explaining the superpowers/ convention would help future contributors understand its purpose.

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!