Merge branch 'develop' into feature/rlinf-gr00t-n1d6-n1d7-converters

Author: johnnynunez

README.md

@@ -2,7 +2,7 @@
 
 ---
 
-# Isaac Lab
+# Isaac Lab 3.0.0 Beta 2
 
 [![IsaacSim](https://img.shields.io/badge/IsaacSim-6.0.0-silver.svg)](https://docs.isaacsim.omniverse.nvidia.com/latest/index.html)
 [![Python](https://img.shields.io/badge/python-3.12-blue.svg)](https://docs.python.org/3/whatsnew/3.12.html)
@@ -14,19 +14,6 @@
 [![License](https://img.shields.io/badge/license-Apache--2.0-yellow.svg)](https://opensource.org/license/apache-2-0)
 
 
-This branch is a development branch for Isaac Sim 6.0, which is currently only available through the Isaac Sim [GitHub repo](https://github.com/isaac-sim/IsaacSim).
-For installation, please refer to the Isaac Sim GitHub repo to build the latest Isaac Sim branch, and follow the binary installation method in the
-Isaac Lab documentation for Isaac Lab installation.
-
-> [!WARNING]
-> A recent breaking change on the Isaac Lab `develop` branch is not compatible with the `develop` branch of Isaac Sim on GitHub.
-> To run Isaac Lab with Isaac Sim's GitHub `develop` branch, use Isaac Lab commit [`f0234a82e432e2a0b0f0a26ca3c5b59e527ddaaa`](https://github.com/isaac-sim/IsaacLab/commit/f0234a82e432e2a0b0f0a26ca3c5b59e527ddaaa) or an earlier commit.
-> Alternatively, use the Isaac Lab [`v3.0.0-beta`](https://github.com/isaac-sim/IsaacLab/tree/v3.0.0-beta) tag.
-
-Note that this branch is currently under active development and may experience breaking changes or error messages.
-Performance issues and regressions may also be observed in some use cases.
-
-
 **Isaac Lab** is a GPU-accelerated, open-source framework designed to unify and simplify robotics research workflows,
 such as reinforcement learning, imitation learning, and motion planning. Built on [NVIDIA Isaac Sim](https://docs.isaacsim.omniverse.nvidia.com/latest/index.html),
 it combines fast and accurate physics and sensor simulation, making it an ideal choice for sim-to-real
@@ -76,6 +63,7 @@ dependency versions for Isaac Sim.
 
 | Isaac Lab Version             | Isaac Sim Version         |
 | ----------------------------- | ------------------------- |
+| `release/3.0.0-beta` branch   | Isaac Sim 6.0             |
 | `develop` branch              | Isaac Sim 6.0             |
 | `main` branch                 | Isaac Sim 4.5 / 5.0 / 5.1 |
 | `v3.0.0*`                     | Isaac Sim 6.0             |
@@ -84,11 +72,6 @@ dependency versions for Isaac Sim.
 | `v2.1.X`                      | Isaac Sim 4.5             |
 | `v2.0.X`                      | Isaac Sim 4.5             |
 
-## Limitations
-
-The `develop` branch of Isaac Lab 3.0-Beta is currently available on Ubuntu. Windows
-support and Isaac Lab pip wheels will be available soon.
-
 ## Contributing to Isaac Lab
 
 We wholeheartedly welcome contributions from the community to make this framework mature and useful for everyone.
@@ -141,7 +124,10 @@ The Isaac Lab framework is released under [BSD-3 License](LICENSE). The `isaacla
 corresponding standalone scripts are released under [Apache 2.0](LICENSE-mimic). The license files of its
 dependencies and assets are present in the [`docs/licenses`](docs/licenses) directory.
 
-Note that Isaac Lab requires Isaac Sim, which includes components under proprietary licensing terms. Please see the [Isaac Sim license](docs/licenses/dependencies/isaacsim-license.txt) for information on Isaac Sim licensing.
+Note that full-featured workflows (PhysX, RTX rendering, ROS, URDF/MJCF importers) require
+[Isaac Sim](https://docs.isaacsim.omniverse.nvidia.com/latest/index.html), which includes
+components under proprietary licensing terms. Kit-less Newton workflows do not require Isaac Sim.
+Please see the [Isaac Sim license](docs/licenses/dependencies/isaacsim-license.txt) for details.
 
 Note that the `isaaclab_mimic` extension requires cuRobo, which has proprietary licensing terms that can be found in [`docs/licenses/dependencies/cuRobo-license.txt`](docs/licenses/dependencies/cuRobo-license.txt).
 

docs/index.rst

@@ -100,7 +100,6 @@ Table of Contents
    :titlesonly:
 
    source/setup/quickstart
-   source/setup/quick_installation
    source/overview/own-project/index
    source/setup/walkthrough/index
    source/tutorials/index

docs/source/experimental-features/rlinf_vla_posttraining.rst

@@ -73,6 +73,8 @@ From the Isaac Lab root directory:
    export OMNI_KIT_ACCEPT_EULA=yes
 
    # Step 1: Install safe dependencies via the isaaclab_contrib[rlinf] extra
+   # NOTE: On DGX Spark / aarch64 systems, build decord from source first
+   # (see "Building decord on DGX Spark / aarch64" below), then run this step.
    uv pip install -e "source/isaaclab_contrib[rlinf]"
 
    # Step 2: Install packages with conflicting constraints (--no-deps to bypass resolver)
@@ -100,9 +102,48 @@ If Step 4 fails, skip installation of flash-attn and apply this patch instead:
    cd Isaac-GR00T
    git apply /path/to/IsaacLab/scripts/imitation_learning/locomanipulation_sdg/gr00t/no_flash_attn.patch
 
+.. note::
+
+   **Windows 11**: If ``git apply`` fails with ``error: corrupt patch at line 41``,
+   use ``patch.exe`` (bundled with Git for Windows) instead:
+
+   .. code-block:: bash
+
+      cd Isaac-GR00T
+      "C:\Program Files\Git\usr\bin\patch.exe" -p1 < \path\to\IsaacLab\scripts\imitation_learning\locomanipulation_sdg\gr00t\no_flash_attn.patch
+
 The patch switches GR00T to PyTorch SDPA, so flash-attn is no longer required.
 The training and evaluation commands below work unchanged.
 
+.. _rlinf-decord-aarch64:
+
+Building decord on DGX Spark / aarch64
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The ``decord`` package only publishes pre-built wheels for ``manylinux2010_x86_64``
+and ``win_amd64``, so installation fails on aarch64 hosts (e.g. DGX Spark / Grace).
+Build decord from source **before** Step 1:
+
+.. code-block:: bash
+
+   git clone --recursive https://github.com/jasontitus/decord
+   cd decord && mkdir -p build && cd build
+   cmake .. -DUSE_CUDA=0 -DCMAKE_BUILD_TYPE=Release
+   make -j$(nproc)
+   cd ../python && pip install -e .
+
+Then preload the OpenMP library so it can be loaded into the Python process
+(see the IsaacLab `pip installation guide
+<https://isaac-sim.github.io/IsaacLab/develop/source/setup/installation/pip_installation.html#installing-dependencies>`_):
+
+.. code-block:: bash
+
+   unset LD_PRELOAD
+   export LD_PRELOAD=/lib/aarch64-linux-gnu/libgomp.so.1
+
+Now re-run Step 1; the resolver will see the locally-installed decord and stop trying to fetch a wheel.
+
+
 Quick Start
 -----------
 
@@ -114,20 +155,64 @@ Quick Start
        --config_name isaaclab_ppo_gr00t_assemble_trocar \
        --model_path /path/to/checkpoint
 
-**Evaluation** — Evaluate a trained checkpoint with video recording:
+**Evaluation** — Evaluate a pretrained (base) model with video recording:
+
+.. code-block:: bash
+
+   python scripts/reinforcement_learning/rlinf/play.py \
+       --config_name isaaclab_ppo_gr00t_assemble_trocar \
+       --model_path /path/to/base_model \
+       --video
+
+**Evaluation** — Evaluate an RL-finetuned checkpoint with video recording:
 
 .. code-block:: bash
 
    python scripts/reinforcement_learning/rlinf/play.py \
        --config_name isaaclab_ppo_gr00t_assemble_trocar \
-       --model_path /path/to/checkpoint \
+       --model_path /path/to/base_model \
+       --rl_model_path /path/to/checkpoints/global_step_N \
        --video
 
+Here ``--model_path`` points to the HuggingFace-format base model (with
+``config.json``), and ``--rl_model_path`` points to the RLinf checkpoint
+directory (the ``global_step_<N>`` folder). The script loads the model
+architecture from the base model and overlays the RL-finetuned weights
+(``full_weights.pt``) from the checkpoint.
+
 .. note::
 
    The ``--config_path`` flag is optional. When omitted, the scripts automatically
    search the ``isaaclab_tasks`` package for the matching YAML configuration file.
 
+Checkpoints
+-----------
+
+Checkpoints are saved every ``save_interval`` epochs (default: ``2``) to::
+
+   scripts/reinforcement_learning/rlinf/logs/rlinf/<timestamp>-Isaac-Assemble-Trocar-G129-Dex3-v0/<experiment_name>/checkpoints/global_step_<N>/
+
+The placeholders are configurable in the task YAML
+(``source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/assemble_trocar/config/isaaclab_ppo_gr00t_assemble_trocar.yaml``):
+
+- ``<experiment_name>`` — ``runner.logger.experiment_name`` (default: ``test_gr00t``)
+- ``<N>`` — increments every ``runner.save_interval`` epochs
+
+The exact path is printed at startup as ``[INFO] Logging to: ...``. To resume,
+pass the ``global_step_<N>`` directory via ``--resume_dir``.
+
+.. tip::
+
+   Training throughput scales with the number of parallel environments. If your
+   GPU has spare memory, increase ``env.train.total_num_envs`` (default: ``4``)
+   in the task YAML.
+
+.. tip::
+
+   Each checkpoint can be several gigabytes. To avoid filling up disk space,
+   increase ``save_interval`` in the task YAML so that fewer
+   intermediate checkpoints are saved during training.
+
 Configuration
 -------------
 

docs/source/how-to/record_video.rst

@@ -81,10 +81,10 @@ the Rerun visualizer rather than producing ``--video`` clips, and Viser does not
 ``--video`` recording backend.
 
 Set ``VideoRecorderCfg.backend_source = "renderer"`` to ignore active visualizers and choose from the
-physics/renderer stack instead. In that mode, PhysX physics (``presets=physx,...``) or Isaac RTX
-(``presets=isaac_rtx_renderer,...``) selects the Kit path. Newton physics (``presets=newton_mjwarp,...``) or
-the Newton Warp renderer (``presets=newton_renderer,...``) selects the Newton GL path when no Kit
-signal is present. OVRTX (``presets=ovrtx_renderer,...`` from ``isaaclab_ov``) can pair with IsaacSim
+physics/renderer stack instead. In that mode, PhysX physics (``physics=physx``) or Isaac RTX
+(``renderer=isaacsim_rtx_renderer``) selects the Kit path. Newton physics (``physics=newton_mjwarp``) or
+the Newton Warp renderer (``renderer=newton_renderer``) selects the Newton GL path when no Kit
+signal is present. OVRTX (``renderer=ovrtx_renderer`` from ``isaaclab_ov``) can pair with IsaacSim
 or Newton physics; in that case the video backend is selected via the physics preset. If both Kit and
 Newton GL signals are present, the Kit path is chosen.
 
@@ -143,16 +143,16 @@ Summary
    :widths: 40 22 38
    :header-rows: 1
 
-   * - Stack example (``presets=...``)
+   * - Stack example (``physics=`` / ``renderer=``)
      - Video backend
      - Capture mechanism
-   * - ``physx,...`` or ``isaac_rtx_renderer,...``
+   * - ``physics=physx`` or ``renderer=isaacsim_rtx_renderer``
      - Kit (``"kit"``)
      - ``/OmniverseKit_Persp`` + Replicator RGB
-   * - ``newton_mjwarp,...`` or ``newton_renderer,...`` (no Kit signals)
+   * - ``physics=newton_mjwarp`` or ``renderer=newton_renderer`` (no Kit signals)
      - Newton GL (``"newton_gl"``)
      - ``newton.viewer.ViewerGL`` on the SDP Newton model
-   * - ``newton_mjwarp,...,ovrtx_renderer,...`` (OVRTX + Newton physics)
+   * - ``physics=newton_mjwarp`` + ``renderer=ovrtx_renderer`` (OVRTX + Newton physics)
      - Newton GL (``"newton_gl"``)
      - ``newton.viewer.ViewerGL`` on the SDP Newton model
    * - ``--visualizer kit`` with default ``backend_source``

docs/source/migration/migrating_to_isaaclab_3-0.rst

@@ -599,18 +599,19 @@ when no CLI override is given. Other fields are named presets selectable with
    @configclass
    class MyPhysicsCfg(PresetCfg):
        default: PhysxCfg = PhysxCfg(...)   # used when no override is given
-       physx:   PhysxCfg = PhysxCfg(...)   # selected by presets=physx
-       newton_mjwarp:  NewtonCfg = NewtonCfg(...)  # selected by presets=newton_mjwarp
+       physx:   PhysxCfg = PhysxCfg(...)   # selected by physics=physx
+       newton_mjwarp:  NewtonCfg = NewtonCfg(...)  # selected by physics=newton_mjwarp
 
 Selecting a preset at launch
 -----------------------------
 
-Pass ``presets=newton_mjwarp`` (or ``presets=physx``) on the CLI to swap the entire config section:
+Pass ``physics=newton_mjwarp`` (or ``physics=physx``) on the CLI to swap the entire config section.
+The legacy ``presets=NAME`` form still works for the same values.
 
 .. code-block:: bash
 
    # Run with Newton backend
-   python train.py task=Isaac-Franka-Cabinet-v0 presets=newton_mjwarp
+   python train.py task=Isaac-Franka-Cabinet-v0 physics=newton_mjwarp
 
    # Run with default (PhysX) backend
    python train.py task=Isaac-Franka-Cabinet-v0

docs/source/overview/core-concepts/physical-backends/newton/installation.rst

@@ -75,4 +75,4 @@ To verify that the installation was successful, run the following command from t
 
 .. code-block:: bash
 
-    ./isaaclab.sh -p scripts/environments/zero_agent.py --task Isaac-Cartpole-Direct-v0 --num_envs 128 --viz newton presets=newton_mjwarp
+    ./isaaclab.sh -p scripts/environments/zero_agent.py --task Isaac-Cartpole-Direct-v0 --num_envs 128 --viz newton physics=newton_mjwarp

docs/source/overview/core-concepts/physical-backends/newton/kamino-solver.rst

@@ -20,17 +20,17 @@ Before adding Kamino, first make sure the task runs with the Newton backend:
 
 .. code-block:: bash
 
-    ./isaaclab.sh -p scripts/environments/zero_agent.py --task Isaac-Cartpole-Direct-v0 --num_envs 128 --viz newton presets=newton_mjwarp
+    ./isaaclab.sh -p scripts/environments/zero_agent.py --task Isaac-Cartpole-Direct-v0 --num_envs 128 --viz newton physics=newton_mjwarp
 
 Then run the same task with the Kamino preset if it is available:
 
 .. code-block:: bash
 
-    ./isaaclab.sh -p scripts/environments/zero_agent.py --task Isaac-Cartpole-Direct-v0 --num_envs 128 --viz newton presets=newton_kamino
+    ./isaaclab.sh -p scripts/environments/zero_agent.py --task Isaac-Cartpole-Direct-v0 --num_envs 128 --viz newton physics=newton_kamino
 
 At the time of writing, the ``newton_kamino`` preset is defined for
 ``Isaac-Cartpole-Direct-v0``, ``Isaac-Ant-Direct-v0``, ``Isaac-Cartpole-v0``,
-and ``Isaac-Ant-v0``. Passing ``presets=newton_kamino`` to another task does not
+and ``Isaac-Ant-v0``. Passing ``physics=newton_kamino`` to another task does not
 automatically enable Kamino; the task must define and validate its own ``newton_kamino``
 preset.
 
@@ -66,7 +66,7 @@ You can select the preset globally:
 
 .. code-block:: bash
 
-    ./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Cartpole-v0 presets=newton_kamino
+    ./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Cartpole-v0 physics=newton_kamino
 
 or select the physics field directly:
 
@@ -75,7 +75,7 @@ or select the physics field directly:
     ./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Cartpole-v0 env.sim.physics=newton_kamino
 
 Use the direct path override when only one task field should use the Kamino preset.
-Use ``presets=newton_kamino`` when you want every matching preset field in the task config
+Use ``physics=newton_kamino`` when you want every matching preset field in the task config
 to resolve to ``newton_kamino``.
 Isaac Lab training scripts accept these Hydra overrides after the regular command
 line flags; no separator is needed for the examples above.
@@ -87,7 +87,7 @@ Check Task and Asset Compatibility
 Kamino uses the Newton model built from the task assets. When adding Kamino to a
 new task, validate the following before tuning solver parameters:
 
-* The task must already be compatible with the Newton backend. If ``presets=newton_mjwarp``
+* The task must already be compatible with the Newton backend. If ``physics=newton_mjwarp``
   fails during model construction, fix the asset or task configuration first.
 * The assets should use Newton-supported rigid bodies, articulations, and collision
   geometry. PhysX-only features, unsupported schemas, or missing collision shapes
@@ -226,7 +226,7 @@ Tuning Workflow
 
 Use the following sequence when bringing up a new Kamino task:
 
-1. Run the task with ``presets=newton_mjwarp`` and fix Newton model construction or task
+1. Run the task with ``physics=newton_mjwarp`` and fix Newton model construction or task
    compatibility issues first.
 2. Add a ``newton_kamino`` preset with conservative values copied from the closest
    validated task.

docs/source/overview/core-concepts/physical-backends/newton/supported-features.rst

@@ -17,7 +17,7 @@ currently supports Newton:
 
     grep -rln "newton_mjwarp" source/isaaclab_tasks/
 
-Passing ``presets=newton_mjwarp`` to a task without that preset will raise an
+Passing ``physics=newton_mjwarp`` to a task without that preset will raise an
 error at launch. The :doc:`mjwarp-solver` page covers how to add a Newton
 preset to your own task.
 

docs/source/overview/core-concepts/physical-backends/newton/using-vbd-solver.rst

@@ -91,7 +91,7 @@ You can select the deformable Newton preset globally:
 
 .. code-block:: bash
 
-    ./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Lift-Soft-Franka-v0 presets=newton_mjwarp_vbd
+    ./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Lift-Soft-Franka-v0 physics=newton_mjwarp_vbd
 
 or select the physics field directly:
 
@@ -100,7 +100,7 @@ or select the physics field directly:
     ./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Lift-Soft-Franka-v0 env.sim.physics=newton_mjwarp_vbd
 
 Use the direct path override when only one task field should use the VBD preset.
-Use ``presets=newton_mjwarp_vbd`` when you want every matching preset field in
+Use ``physics=newton_mjwarp_vbd`` when you want every matching preset field in
 the task config to resolve to that preset. Isaac Lab training scripts accept
 these Hydra overrides after the regular command line flags; no separator is
 needed for the examples above.

docs/source/overview/core-concepts/physical-backends/newton/warp-environments.rst

@@ -228,7 +228,7 @@ specific to warp envs; for Newton physics limitations see :doc:`supported-featur
   ``class_type`` fields resolve to ``isaaclab_physx.*`` classes that depend on
   ``omni.physics.tensors`` (a Kit module the warp runtime does not initialise), and several
   warp APIs (env-mask reset, CUDA graph capture) require the Newton articulation. Configure
-  the cfg with a Newton physics block (or ``presets=newton_mjwarp``).
+  the cfg with a Newton physics block (or ``physics=newton_mjwarp``).
 
 **MDP coverage**