Add OVXR camera panels, G1 feedback cameras and desktop mirror

docs/source/_static/css/environment-browser.js

@@ -98,11 +98,11 @@
             ["IsaacContrib-Open-Drawer-Franka-IK-Abs", "rsl_rl", "isaacsim_physx,newton_mjwarp,ovphysx", "", "", "environment-previews/franka_open_drawer.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||", "newton_mjwarp||", "ovphysx||"]],
             ["IsaacContrib-Open-Drawer-Franka-IK-Rel", "rsl_rl", "isaacsim_physx,newton_mjwarp,ovphysx", "", "", "environment-previews/franka_open_drawer.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||", "newton_mjwarp||", "ovphysx||"]],
             ["IsaacContrib-Open-Drawer-OpenArm", "rl_games,rsl_rl", "", "", "", "environment-previews/openarm_uni_open_drawer.jpg"],
-            ["IsaacContrib-PickPlace-FixedBaseUpperBodyIK-G1-Abs", "", "", "isaacsim_rtx,newton_renderer", "", "environment-previews/g1_pick_place_fixed_base.jpg", false, {}, {}, "", null, ["||", "|isaacsim_rtx|", "|newton_renderer|"]],
+            ["IsaacContrib-PickPlace-FixedBaseUpperBodyIK-G1-Abs", "", "isaacsim_physx,newton_mjwarp,ovphysx", "isaacsim_rtx,newton_renderer,ovrtx", "", "environment-previews/g1_pick_place_fixed_base.jpg", false, {}, {}, "", null, ["||", "|isaacsim_rtx|", "|newton_renderer|", "isaacsim_physx||", "isaacsim_physx|isaacsim_rtx|", "isaacsim_physx|newton_renderer|", "newton_mjwarp||", "newton_mjwarp|isaacsim_rtx|", "newton_mjwarp|newton_renderer|", "newton_mjwarp|ovrtx|", "ovphysx|newton_renderer|", "ovphysx|ovrtx|"]],
             ["IsaacContrib-PickPlace-G1-InspireFTP-Abs", "", "", "", "", "environment-previews/g1_pick_place.jpg"],
-            ["IsaacContrib-PickPlace-GR1T2-Abs", "", "", "isaacsim_rtx,newton_renderer", "", "environment-previews/gr_1_pick_place.jpg", false, {}, {}, "", null, ["||", "|isaacsim_rtx|", "|newton_renderer|"]],
-            ["IsaacContrib-PickPlace-GR1T2-WaistEnabled-Abs", "", "", "", "", "environment-previews/gr_1_pick_place_waist.jpg"],
-            ["IsaacContrib-PickPlace-Locomanipulation-G1-Abs", "", "", "isaacsim_rtx,newton_renderer", "", "environment-previews/g1_pick_place_locomanipulation.jpg", false, {}, {}, "", null, ["||", "|isaacsim_rtx|", "|newton_renderer|"]],
+            ["IsaacContrib-PickPlace-GR1T2-Abs", "", "isaacsim_physx,newton_mjwarp,ovphysx", "isaacsim_rtx,newton_renderer,ovrtx", "", "environment-previews/gr_1_pick_place.jpg", false, {}, {}, "", null, ["||", "|isaacsim_rtx|", "|newton_renderer|", "isaacsim_physx||", "isaacsim_physx|isaacsim_rtx|", "isaacsim_physx|newton_renderer|", "newton_mjwarp||", "newton_mjwarp|isaacsim_rtx|", "newton_mjwarp|newton_renderer|", "newton_mjwarp|ovrtx|", "ovphysx|newton_renderer|", "ovphysx|ovrtx|"]],
+            ["IsaacContrib-PickPlace-GR1T2-WaistEnabled-Abs", "", "isaacsim_physx,newton_mjwarp,ovphysx", "", "", "environment-previews/gr_1_pick_place_waist.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||", "newton_mjwarp||", "ovphysx||"]],
+            ["IsaacContrib-PickPlace-Locomanipulation-G1-Abs", "", "isaacsim_physx,newton_mjwarp,ovphysx", "isaacsim_rtx,newton_renderer,ovrtx", "", "environment-previews/g1_pick_place_locomanipulation.jpg", false, {}, {}, "", null, ["||", "|isaacsim_rtx|", "|newton_renderer|", "isaacsim_physx||", "isaacsim_physx|isaacsim_rtx|", "isaacsim_physx|newton_renderer|", "newton_mjwarp||", "newton_mjwarp|isaacsim_rtx|", "newton_mjwarp|newton_renderer|", "newton_mjwarp|ovrtx|", "ovphysx|newton_renderer|", "ovphysx|ovrtx|"]],
             ["IsaacContrib-Place-Mug-Agibot-Left-Arm-RmpFlow", "", "isaacsim_physx", "", "", "environment-previews/agibot_place_mug.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
             ["IsaacContrib-Place-Toy2Box-Agibot-Right-Arm-RmpFlow", "", "isaacsim_physx", "", "", "environment-previews/agibot_place_toy.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
             ["IsaacContrib-Reach-OpenArm", "rl_games,rsl_rl,skrl", "", "", "", "environment-previews/openarm_uni_reach.jpg"],
@@ -129,7 +129,7 @@
             ["IsaacContrib-Stack-Cube-Instance-Randomize-Franka-IK-Rel", "", "", "", "", "environment-previews/franka_instance_randomize_stack.jpg"],
             ["IsaacContrib-Stack-Cube-RedGreen-Franka-IK-Rel", "", "isaacsim_physx", "", "", "environment-previews/franka_stack.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
             ["IsaacContrib-Stack-Cube-RedGreenBlue-Franka-IK-Rel", "", "isaacsim_physx", "", "", "environment-previews/franka_stack.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
-            ["IsaacContrib-Stack-Cube-SO101-IK-Abs-v0", "", "isaacsim_physx", "", "", "environment-previews/so101_stack.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
+            ["IsaacContrib-Stack-Cube-SO101-IK-Abs-v0", "", "isaacsim_physx,newton_mjwarp,ovphysx", "", "", "environment-previews/so101_stack.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||", "newton_mjwarp||", "ovphysx||"]],
             ["IsaacContrib-Stack-Cube-SO101-Joint-Teleop-v0", "", "isaacsim_physx", "", "", "environment-previews/so101_stack.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
             ["IsaacContrib-Stack-Cube-SO101-v0", "", "isaacsim_physx", "", "", "environment-previews/so101_stack.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],
             ["IsaacContrib-Stack-Cube-UR10-Long-Suction-IK-Rel", "", "isaacsim_physx", "", "", "environment-previews/ur10_stack_surface_gripper.jpg", false, {}, {}, "", null, ["||", "isaacsim_physx||"]],

docs/source/features/isaac_teleop.rst

@@ -1344,6 +1344,34 @@ through this variable -- later runs no longer prompt. The variable applies to ev
 that launches the runtime, including the process-scoped launcher in
 ``teleop_replay_agent.py``.
 
+.. _isaac-teleop-ovxr-mcap:
+
+Replay MCAP with a Kitless OVXR Viewer
+--------------------------------------
+
+The MCAP replay entrypoint can present recorded input through an independent
+OVXR viewer. Follow the :ref:`kitless XR guide <kitless-xr>` to install Isaac
+Capture with its OVXR extra and select the compatible rendering dependencies.
+
+.. code-block:: bash
+
+   uv run --extra ovxr python \
+       scripts/environments/teleoperation/teleop_replay_agent.py \
+       --task IsaacContrib-PickPlace-GR1T2-Abs --physics ovphysx \
+       --device cpu --num_envs 1 --ovxr \
+       --replay_file /absolute/path/to/session.mcap
+
+Use ``--physics newton_mjwarp`` for Newton. MCAP supplies tracker and control
+input; the connected client supplies the view. A recorded STOP, task success
+or failure, or the replay duration limit ends the run. The separate
+``isaaclab teleop replay`` command replays HDF5 demonstrations through Kit.
+
+Add ``--ovxr_pip_camera robot_pov_cam --ovxr_pip_head_locked`` for a camera
+panel. OVXR live and MCAP replay open a desktop mirror by default; add
+``--headless`` to disable that window while keeping XR and PiP rendering.
+See the kitless XR guide for installation, camera selection and display requirements.
+
+
 .. _isaac-teleop-xr-anchor:
 
 Configure the XR Anchor
@@ -1485,7 +1513,7 @@ unchanged. For either G1 task, set ``env.isaac_teleop.xr_camera_feeds=[]`` to di
 the camera sensor and any configured camera observations.
 
 The reference feeds request render-product-local DLSS Ray Reconstruction and ``quality`` execution
-mode through :class:`~isaaclab_teleop.XrCameraFeedCfg`. The private PiP adapter authors those two
+mode through :class:`~isaaclab_teleop.XrCameraFeedCfg`. The Kit Scene UI presenter authors those two
 attributes only on the selected camera render product. Ray Reconstruction also requires the
 process-global responsive-denoising setting, which the teleoperation and recording scripts
 enable before environment construction when a selected feed requests Ray Reconstruction.
@@ -1596,8 +1624,10 @@ Kit Scene UI presentation
 PiP presentation uses Kit Scene UI and ``SpatialSource`` placement. Kit imports are deferred until
 an enabled feed is requested. If the Scene UI modules cannot be imported, the scripts log a warning
 and continue without PiP; task-owned cameras and recording observations remain unchanged. This keeps
-the camera selection configuration usable when a future kitless entry point no longer provides
-Scene UI. Render-product tuning is also best-effort: backends without a compatible render product
+the camera selection configuration usable without Scene UI. The
+:ref:`kitless XR guide <kitless-xr>` uses separate OpenXR quad
+layers and does not import Scene UI; those layers are visible in the headset/client, not in the
+passive eye-image desktop mirror. Render-product tuning is also best-effort: backends without a compatible render product
 keep the Camera-buffer path, and schema-authoring failures warn while PiP continues. Other
 configuration, camera-buffer, and panel-initialization errors still fail during startup.
 
@@ -2031,8 +2061,8 @@ Optimize XR Performance
 
    .. note::
 
-      The ``--headless`` CLI flag was removed in Isaac Lab 3.0. Headless is now the **default**
-      (omit ``--visualizer``), or set ``HEADLESS=1`` in the environment.
+      For Kit XR, omit ``--visualizer`` to run without the local viewport.
+      For kitless ``--ovxr``, pass ``--headless`` to disable its default desktop mirror.
 
 .. dropdown:: Configure retargeting execution
    :open:

docs/source/how-to/cloudxr_teleoperation.rst

@@ -146,9 +146,10 @@ to capture demonstrations, and ``replay`` to play a dataset back. :ref:`The next
 
 .. note::
 
-   ``teleop`` cannot be combined with ``ov`` or ``ovphysx`` in a single ``uv run``: the
-   bundled Isaac Sim pins ``packaging==26.0`` while those runtimes require ``<24``. Install
-   the OV runtimes separately when you need them.
+   For XR teleoperation without Isaac Sim, follow the
+   :ref:`kitless XR guide <kitless-xr>`. It explains how to select a public
+   Isaac Capture wheel and its optional OVXR dependencies. The Kit setup below
+   uses the published ``teleop`` extra.
 
 Complete these steps first:
 

docs/source/how-to/index.rst

@@ -510,6 +510,12 @@ How-to Guides
 
             :guide-tag:`PhysX` :guide-tag:`Isaac Sim`
 
+      .. container:: guide-entry
+
+         :doc:`Kitless XR with Isaac Capture </source/how-to/kitless_xr>`
+
+         Install Isaac Capture and OVXR packages for live teleoperation and MCAP replay.
+
       .. container:: guide-entry
 
          :doc:`Setting up Haply teleoperation </source/how-to/haply_teleoperation>`
@@ -556,3 +562,9 @@ How-to Guides
 
       This collection is a work in progress. If a question is not answered here, open an issue on the
       `Isaac Lab GitHub repository <https://github.com/isaac-sim/IsaacLab>`_.
+
+.. toctree::
+   :hidden:
+
+   kitless_xr
+   kitless_xr_source

docs/source/how-to/kitless_xr.md

@@ -0,0 +1,15 @@
+---
+myst:
+  heading_anchors: 2
+---
+
+<!--
+Copyright (c) 2022-2026, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
+All rights reserved.
+SPDX-License-Identifier: BSD-3-Clause
+-->
+
+(kitless-xr)=
+
+```{include} ../../../source/isaaclab_teleop/docs/kitless_xr.md
+```

docs/source/how-to/kitless_xr_source.md

@@ -0,0 +1,8 @@
+<!--
+Copyright (c) 2022-2026, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
+All rights reserved.
+SPDX-License-Identifier: BSD-3-Clause
+-->
+
+```{include} ../../../source/isaaclab_teleop/docs/kitless_xr_source.md
+```

source/isaaclab/changelog.d/kitless-ovxr-lifecycle.fixed.rst

@@ -0,0 +1,2 @@
+* Allowed asynchronous XR owners to drain before releasing the simulation context.
+* Supported empty CPU Warp arrays in Torch conversion.

source/isaaclab/changelog.d/pink-hand-joint-order.fixed.rst

@@ -0,0 +1 @@
+* Preserved configured hand joint ordering when applying Pink action targets.

source/isaaclab/isaaclab/app/sim_launcher.py

@@ -412,12 +412,19 @@ def validate_launch(  # noqa: C901
 def launch_simulation(
     cfg,
     launcher_args: argparse.Namespace | dict | None = None,
+    *,
+    cleanup_context: bool = True,
 ) -> Generator[PhysicsCfg | None, None, None]:
     """Context manager that launches the appropriate simulation runtime for *cfg*.
 
     Resolves configured consumers before starting their required runtimes. Each runtime starts
     once and closes on exit. Cameras are auto-enabled for Kit-renderer sensors.
 
+    On exit, stops and releases a simulation context created inside the scope before closing
+    its runtimes. An existing context owned by the caller is preserved. Applications
+    with asynchronous resource owners can set ``cleanup_context=False`` and take
+    responsibility for context cleanup, including failed construction and drain.
+
     Resolve viewer and physics choices on the same config used to construct the simulation::
 
         sim_cfg = SimulationCfg()
@@ -437,6 +444,10 @@ def launch_simulation(
               a tool that reaches a Kit-only extension API. This is additive -- it can only turn a
               kitless launch into a Kit one, never the reverse, so a config that already needs Kit
               still launches it when the key is absent or ``False``.
+        cleanup_context: Automatically release a context created inside this scope.
+            Disable only when the application owns its cleanup and must retain it
+            if asynchronous consumers cannot prove they have drained. This controls
+            context release only; configured runtime launchers still close on exit.
     """
     launcher_args = {} if launcher_args is None else launcher_args
     launcher_args = vars(launcher_args) if isinstance(launcher_args, argparse.Namespace) else launcher_args
@@ -500,7 +511,7 @@ def launch_simulation(
         finally:
             # Release runtime consumers before a launcher can terminate the process.
             context = SimulationContext.instance()
-            if context is not None and context is not previous_context:
+            if cleanup_context and context is not None and context is not previous_context:
                 SimulationContext.clear_instance()
     except KeyboardInterrupt:
         exit_code = 130

source/isaaclab/isaaclab/envs/mdp/actions/pink_task_space_actions.py

@@ -72,8 +72,12 @@ def _initialize_joint_info(self) -> None:
         self._isaaclab_all_joint_ids = list(range(len(self._asset.data.joint_names)))
         self.cfg.controller.all_joint_names = self._asset.data.joint_names
 
-        # Resolve hand joints
-        self._hand_joint_ids, self._hand_joint_names = self._asset.find_joints(self.cfg.hand_joint_names)
+        # Hand targets occupy the action tensor in configuration order. Newton's
+        # articulation order groups these joints differently, so preserve the
+        # declared order when resolving ids instead of silently permuting targets.
+        self._hand_joint_ids, self._hand_joint_names = self._asset.find_joints(
+            self.cfg.hand_joint_names, preserve_order=True
+        )
 
         # Combine all joint information
         self._controlled_joint_ids = self._isaaclab_controlled_joint_ids + self._hand_joint_ids

source/isaaclab/isaaclab/utils/warp/__init__.py

@@ -37,8 +37,9 @@ def _wp_to_torch_with_proxyarray(a, requires_grad=None):
 
     The shim routes :class:`ProxyArray` arguments to their cached ``.torch``
     view (a zero-copy :class:`torch.Tensor` of the same underlying memory) and
-    emits a one-shot :class:`DeprecationWarning`. For any other input type,
-    the original :func:`warp.to_torch` handles the call as before.
+    emits a one-shot :class:`DeprecationWarning`. Empty CPU arrays use DLPack
+    because NumPy rejects their null-pointer array interface. Other native
+    arrays keep the original :func:`warp.to_torch` conversion.
     """
     global _WP_TO_TORCH_WARNED
     if isinstance(a, ProxyArray):
@@ -51,6 +52,19 @@ def _wp_to_torch_with_proxyarray(a, requires_grad=None):
                 stacklevel=2,
             )
         return a.torch
+    if isinstance(a, wp.array) and a.size == 0 and a.device is not None and a.device.is_cpu:
+        # Preserve Warp's structured-array rejection before the DLPack fallback.
+        if isinstance(a.dtype, wp._src.codegen.Struct):  # noqa: SLF001
+            return _WP_TO_TORCH_ORIGINAL(a, requires_grad=requires_grad)
+        import torch
+
+        if requires_grad is None:
+            requires_grad = a.requires_grad
+        tensor = torch.from_dlpack(wp.to_dlpack(a))
+        tensor.requires_grad = requires_grad
+        if requires_grad and a.requires_grad:
+            tensor.grad = torch.from_dlpack(wp.to_dlpack(a.grad))
+        return tensor
     return _WP_TO_TORCH_ORIGINAL(a, requires_grad=requires_grad)
 
 

source/isaaclab/test/app/test_launch_simulation_require_kit.py

@@ -256,3 +256,40 @@ def test_legacy_recording_reuses_the_display_producer(kit_branch_taken, visualiz
         assert cfg.sim.visualizer_cfgs[0].headless == (not visualizers)
         assert args.get("enable_cameras", False) == (source == "kit")
     assert kit_branch_taken == [True] * (source == "kit")
+
+
+@pytest.mark.parametrize(
+    "application_owns_cleanup,drain_failed", [(False, False), (False, True), (True, False), (True, True)]
+)
+def test_context_cleanup_respects_application_drain_ownership(monkeypatch, application_owns_cleanup, drain_failed):
+    """An outer launcher must not release a context retained by a failed native drain."""
+    from isaaclab.sim import SimulationContext
+
+    state = {"context": None}
+    released = []
+
+    def clear():
+        released.append(state["context"])
+        state["context"] = None
+
+    monkeypatch.setattr(SimulationContext, "instance", lambda: state["context"])
+    monkeypatch.setattr(SimulationContext, "clear_instance", clear)
+    context = object()
+
+    def run():
+        with launch_simulation(PhysicsCfg(), cleanup_context=not application_owns_cleanup):
+            state["context"] = context
+            if drain_failed:
+                raise RuntimeError("consumer drain unproved")
+            if application_owns_cleanup:
+                # The already-owning environment/worker performs normal release.
+                SimulationContext.clear_instance()
+
+    if drain_failed:
+        with pytest.raises(RuntimeError, match="consumer drain unproved"):
+            run()
+    else:
+        run()
+    retained = application_owns_cleanup and drain_failed
+    assert released == ([] if retained else [context])
+    assert state["context"] is (context if retained else None)