[Fix] Force numpy>=2.4.1 in --install and Dockerfile.curobo

The setup.py constraint "numpy>=2,!=2.3.5" landed in isaac-sim#5642 is silently
overridden during isaaclab.sh --install: each pip install -e <submodule>
runs an independent resolve, and the final pin-pink force-reinstall in
_ensure_pink_ik_dependencies_installed lands on numpy 2.3.5 because pip
sees only pin-pink's own deps (numpy>=1.19) plus cmeel-boost's numpy<2.4
cap.  numpy 2.3.5 ships a vendored OpenBLAS
(libscipy_openblas64_-fdde5778.so) whose pthread_atfork handler crashes
Kit's libomni.platforminfo fork() during SimulationApp startup.

Two changes, both restating an explicit "pip install --upgrade
numpy>=2.4.1" as the *last* pip invocation in each install path:

1. _ensure_numpy_above_openblas_atfork_bug() in install.py — runs
   unconditionally at the end of --install (not gated by the pink-ik
   probe outcome), so upgrades on an already-functioning env also
   pull numpy forward.
2. Dockerfile.curobo — apply the same upgrade after its post-install
   steps (nvidia-curobo + isaaclab_teleop editable install), which
   otherwise drag numpy back to 2.3.5 via dex-retargeting -> pin ->
   cmeel-boost.

pip prints a resolver warning about cmeel-boost's cap then installs
numpy 2.4.5 anyway.  numpy 2.4.1+ ships the upstream OpenBLAS atfork
fix, so the entire 2.3.x risk class is bypassed.  numpy's stable C
ABI keeps cmeel's compiled extensions (libpinocchio, libcoal, ...)
working at runtime.

Validated:
- env_isaaclab_test smoke test (numpy 2.4.5 + cmeel pinocchio + pink +
  daqp + qpsolvers all import; toy IK solve OK).
- IsaacLab Pink IK unit tests: 54/54 pass against numpy 2.4.5
  (test_pink_ik_components 21/21, test_local_frame_task 24/24,
   test_null_space_posture_task 9/9).
- PR isaac-sim#5655 (validation): every base-image test job reports numpy 2.4.5
  + openblas -32a4b2a6 (clean, not the broken -fdde5778).  Worst-case
  import order (numpy imported before pytest spawns Kit) also passes —
  confirming the upstream atfork fix is real, not just dodge-by-order.

Related: numpy/numpy#30092, OpenMathLib/OpenBLAS#5520

Author: hujc7

docker/Dockerfile.curobo

@@ -178,6 +178,14 @@ RUN ${ISAACLAB_PATH}/isaaclab.sh -p -m pip install --no-build-isolation \
 # Install isaaclab_teleop (needed by Pink IK tests and related env configs)
 RUN ${ISAACLAB_PATH}/isaaclab.sh -p -m pip install --editable ${ISAACLAB_PATH}/source/isaaclab_teleop
 
+# Force numpy >= 2.4.1 after the curobo + isaaclab_teleop installs above, because both
+# pull pin (via cmeel) whose numpy<2.4 cap drags numpy back to 2.3.5 — the release whose
+# vendored OpenBLAS atfork handler SIGSEGVs inside Kit's libomni.platforminfo fork().
+# isaaclab.sh --install does the same upgrade at the base layer; this restates it here
+# so the cuRobo image picks numpy with the upstream OpenBLAS fix. See numpy/numpy#30092
+# and OpenMathLib/OpenBLAS#5520.
+RUN ${ISAACLAB_PATH}/isaaclab.sh -p -m pip install --upgrade "numpy>=2.4.1"
+
 # aliasing isaaclab.sh and python for convenience
 RUN echo "export ISAACLAB_PATH=${ISAACLAB_PATH}" >> ${DOCKER_USER_HOME}/.bashrc && \
     echo "alias isaaclab=${ISAACLAB_PATH}/isaaclab.sh" >> ${DOCKER_USER_HOME}/.bashrc && \

source/isaaclab/changelog.d/jichuanh-force-numpy-install.rst

@@ -0,0 +1,17 @@
+Fixed
+^^^^^
+
+* Fixed ``isaaclab.sh --install`` so a final
+  :meth:`isaaclab.cli.commands.install._ensure_numpy_above_openblas_atfork_bug` step
+  unconditionally force-upgrades ``numpy`` to ``>=2.4.1``. Without it, pip resolves the
+  per-submodule installs and the pin-pink force-reinstall independently, so
+  cmeel-boost's ``numpy<2.4`` cap (transitive via pin-pink) keeps landing on numpy
+  2.3.5 — the release whose vendored OpenBLAS
+  ``libscipy_openblas64_-fdde5778.so`` registers a buggy ``pthread_atfork`` handler
+  that crashes Kit's ``libomni.platforminfo`` ``fork()`` during ``SimulationApp``
+  startup. numpy 2.4.1+ ships the upstream OpenBLAS atfork fix, so the entire 2.3.x
+  risk class is bypassed.
+* Fixed ``docker/Dockerfile.curobo`` to apply the same ``numpy>=2.4.1`` force-upgrade
+  after its post-``--install`` steps (``nvidia-curobo`` and ``isaaclab_teleop`` editable
+  install), which would otherwise drag numpy back to 2.3.5 via ``dex-retargeting`` →
+  ``pin`` → cmeel-boost.

source/isaaclab/isaaclab/cli/commands/install.py

@@ -225,6 +225,39 @@ def _ensure_pink_ik_dependencies_installed(python_exe: str, pip_cmd: list[str],
         )
 
 
+def _ensure_numpy_above_openblas_atfork_bug(python_exe: str, pip_cmd: list[str]) -> None:
+    """Force numpy >= 2.4.1 so the OpenBLAS atfork SIGSEGV in numpy 2.3.5 is skipped.
+
+    numpy 2.3.5 ships a vendored OpenBLAS (``libscipy_openblas64_-fdde5778.so``) whose
+    ``pthread_atfork`` handler crashes inside Kit's ``libomni.platforminfo`` ``fork()``
+    during ``SimulationApp`` startup. The fix landed upstream in numpy 2.4.1+.
+
+    The ``--install`` pipeline runs many independent ``pip install`` invocations and
+    pip's resolver does not retroactively check installed packages' constraints. In
+    particular, the kit-bundled and force-reinstalled ``pin-pink`` stack pulls
+    ``cmeel-boost`` which caps numpy ``<2.4``, so every other install path resolves
+    back to numpy 2.3.5. This step runs *last* with an explicit ``--upgrade
+    numpy>=2.4.1`` to override the cap; pip prints a resolver warning about
+    cmeel-boost but installs numpy 2.4.5 anyway. numpy's stable C ABI keeps cmeel's
+    compiled extensions (``libpinocchio``, ``libcoal``, ...) functional at runtime.
+
+    Always runs unconditionally — independent of
+    :func:`_ensure_pink_ik_dependencies_installed`'s probe — so that an upgrade on
+    an already-functioning environment also pulls numpy forward. See numpy/numpy#30092,
+    OpenMathLib/OpenBLAS#5520, OMPE-92261.
+    """
+    print_info("Forcing numpy >= 2.4.1 to skip the OpenBLAS atfork SIGSEGV in numpy 2.3.5.")
+    numpy_upgrade = run_command(
+        pip_cmd + ["install", "--upgrade", "numpy>=2.4.1"],
+        check=False,
+    )
+    if numpy_upgrade.returncode != 0:
+        print_warning(
+            f"Forcing numpy >= 2.4.1 failed (returncode {numpy_upgrade.returncode}). "
+            "Kit fork() may SIGSEGV on numpy 2.3.5's vendored OpenBLAS."
+        )
+
+
 def _ensure_cuda_torch() -> None:
     """Ensure correct PyTorch and CUDA versions are installed."""
     python_exe = extract_python_exe()
@@ -857,6 +890,11 @@ def command_install(install_type: str = "all") -> None:
         # ``pip install -e source/isaaclab`` to be silently skipped.
         _ensure_pink_ik_dependencies_installed(python_exe, pip_cmd, probe_env=probe_env)
 
+        # Override the numpy<2.4 cap that cmeel-boost (transitive dep of pin-pink) imposes
+        # on every prior pip resolution, so the test image lands on numpy with the upstream
+        # OpenBLAS atfork fix instead of the broken 2.3.5 release.
+        _ensure_numpy_above_openblas_atfork_bug(python_exe, pip_cmd)
+
         # Repoint prebundled packages in Isaac Sim to the environment's copies so
         # the active venv/conda versions are always loaded regardless of PYTHONPATH
         # ordering (e.g. torch+cu130 in venv vs torch+cu128 in prebundle on aarch64).

source/isaaclab/setup.py

@@ -18,7 +18,13 @@
 # Minimum dependencies required prior to installation
 INSTALL_REQUIRES = [
     # generic
-    "numpy>=2",
+    # numpy 2.3.5 ships a vendored OpenBLAS whose pthread_atfork handler crashes
+    # Kit's libomni.platforminfo fork() during SimulationApp startup. See
+    # numpy/numpy#30092 and OMPE-92261. ``isaaclab.sh --install`` also
+    # force-upgrades numpy >= 2.4.1 at the end of the install pipeline as the
+    # authoritative defense, but the per-package exclusion here keeps the broken
+    # release out of intermediate resolutions too.
+    "numpy>=2,!=2.3.5",
     "torch>=2.10",
     "onnx>=1.18.0",  # 1.16.2 throws access violation on Windows
     "prettytable==3.3.0",

source/isaaclab_rl/changelog.d/jichuanh-force-numpy-install.rst

@@ -0,0 +1,8 @@
+Fixed
+^^^^^
+
+* Excluded the broken ``numpy 2.3.5`` release from the package's install requirements.
+  ``numpy 2.3.5``'s vendored OpenBLAS registers a ``pthread_atfork`` handler that
+  crashes Kit's ``libomni.platforminfo`` ``fork()`` during ``SimulationApp`` startup.
+  See ``source/isaaclab/setup.py`` and ``isaaclab.cli.commands.install`` for the
+  authoritative install-time defense.

source/isaaclab_rl/setup.py

@@ -19,7 +19,8 @@
 # Minimum dependencies required prior to installation
 INSTALL_REQUIRES = [
     # generic
-    "numpy",
+    # See ``source/isaaclab/setup.py`` for the rationale on ``!=2.3.5``.
+    "numpy>=2,!=2.3.5",
     "torch>=2.10",
     "torchvision>=0.25.0",  # ensure compatibility with torch 2.10.0
     "protobuf>=4.25.8,!=5.26.0",

source/isaaclab_tasks/changelog.d/jichuanh-force-numpy-install.rst

@@ -0,0 +1,8 @@
+Fixed
+^^^^^
+
+* Excluded the broken ``numpy 2.3.5`` release from the package's install requirements.
+  ``numpy 2.3.5``'s vendored OpenBLAS registers a ``pthread_atfork`` handler that
+  crashes Kit's ``libomni.platforminfo`` ``fork()`` during ``SimulationApp`` startup.
+  See ``source/isaaclab/setup.py`` and ``isaaclab.cli.commands.install`` for the
+  authoritative install-time defense.

source/isaaclab_tasks/setup.py

@@ -18,7 +18,8 @@
 # Minimum dependencies required prior to installation
 INSTALL_REQUIRES = [
     # generic
-    "numpy>=2",
+    # See ``source/isaaclab/setup.py`` for the rationale on ``!=2.3.5``.
+    "numpy>=2,!=2.3.5",
     "torch>=2.10",
     "torchvision>=0.25.0",  # ensure compatibility with torch 2.10.0
     "protobuf>=4.25.8,!=5.26.0",

source/isaaclab_visualizers/changelog.d/jichuanh-force-numpy-install.rst

@@ -0,0 +1,8 @@
+Fixed
+^^^^^
+
+* Excluded the broken ``numpy 2.3.5`` release from the package's install requirements.
+  ``numpy 2.3.5``'s vendored OpenBLAS registers a ``pthread_atfork`` handler that
+  crashes Kit's ``libomni.platforminfo`` ``fork()`` during ``SimulationApp`` startup.
+  See ``source/isaaclab/setup.py`` and ``isaaclab.cli.commands.install`` for the
+  authoritative install-time defense.

source/isaaclab_visualizers/setup.py

@@ -10,7 +10,8 @@
 # Base requirements shared across visualizer backends.
 INSTALL_REQUIRES = [
     "isaaclab",
-    "numpy",
+    # See ``source/isaaclab/setup.py`` for the rationale on ``!=2.3.5``.
+    "numpy>=2,!=2.3.5",
 ]
 
 # Every Newton declaration in the repo must use the SAME extra spec (`newton[sim]`).