Add 5 more kitchen envs and droid stand xy scaling supports (#1149)

## Summary
Short description of the change (max 50 chars)

## Detailed description
- Added stand xy scaling into droid params, allowing fitting kitchens
with narrower stand footprint
- Added LW kitchen envs: 1x multi-level PnP, 1 open-cabinet(hard), 2x
fridge (left door) opening, 1x freezer(right door) opening.
- Tested with PI, only cabinet env has 0 SR but small joint (0.25)
movement

---------

Signed-off-by: Xinjie Yao <xyao@nvidia.com>

Author: xyao-nv

docs/images/kitchen_bench/kitchen_bench_lightwheel_open_cabinet_u_shaped_with_island_farmhouse2.gif

@@ -35,7 +35,7 @@ Kitchen benchmark environment graph YAMLs live under
 ``isaaclab_arena_environments/kitchen_bench/``. They define DROID manipulation
 tasks across Lightwheel RoboCasa and Replicator kitchen layouts.
 
-See :doc:`kitchen_bench_catalog` for all 21 environment specs and their Pi
+See :doc:`kitchen_bench_catalog` for all 26 environment specs and their Pi
 policy executions.
 
 Python Environment Catalog

docs/images/kitchen_bench/kitchen_bench_lightwheel_open_freezer_g_shaped_large_scandinavian.gif

@@ -1,7 +1,7 @@
 Kitchen Benchmark Catalog
 =========================
 
-The kitchen benchmark contains 21 DROID tasks defined as environment graph
+The kitchen benchmark contains 26 DROID tasks defined as environment graph
 YAML specs under ``isaaclab_arena_environments/kitchen_bench/``. Pass a spec
 to ``policy_runner.py`` with ``--env_spec`` to run that environment.
 Each row links to the source YAML and shows a Pi policy execution recorded from
@@ -14,10 +14,26 @@ the maintained DROID external camera.
 
    * - Environment graph YAML
      - Pi policy execution
+   * - `kitchen_bench_lightwheel_open_cabinet_u_shaped_with_island_farmhouse2.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_open_cabinet_u_shaped_with_island_farmhouse2.yaml>`_
+     - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_open_cabinet_u_shaped_with_island_farmhouse2.gif
+          :alt: Pi policy opening a cabinet in the U-shaped farmhouse kitchen
+          :width: 100%
+   * - `kitchen_bench_lightwheel_open_freezer_g_shaped_large_scandinavian.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_open_freezer_g_shaped_large_scandinavian.yaml>`_
+     - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_open_freezer_g_shaped_large_scandinavian.gif
+          :alt: Pi policy opening the freezer in the G-shaped Scandinavian kitchen
+          :width: 100%
    * - `kitchen_bench_lightwheel_open_fridge.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_open_fridge.yaml>`_
      - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_open_fridge_pi.gif
           :alt: Pi policy opening the refrigerator in the Lightwheel kitchen
           :width: 100%
+   * - `kitchen_bench_lightwheel_open_fridge_g_shaped_large_scandinavian.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_open_fridge_g_shaped_large_scandinavian.yaml>`_
+     - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_open_fridge_g_shaped_large_scandinavian.gif
+          :alt: Pi policy opening the refrigerator in the G-shaped Scandinavian kitchen
+          :width: 100%
+   * - `kitchen_bench_lightwheel_open_fridge_u_shaped_with_island_farmhouse2.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_open_fridge_u_shaped_with_island_farmhouse2.yaml>`_
+     - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_open_fridge_u_shaped_with_island_farmhouse2.gif
+          :alt: Pi policy opening the refrigerator in the U-shaped farmhouse kitchen
+          :width: 100%
    * - `kitchen_bench_lightwheel_open_microwave.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_open_microwave.yaml>`_
      - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_open_microwave_pi.gif
           :alt: Pi policy opening the microwave in the Lightwheel kitchen
@@ -26,6 +42,10 @@ the maintained DROID external camera.
      - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_pick_and_place_pi.gif
           :alt: Pi policy placing mustard in a bowl in the Lightwheel kitchen
           :width: 100%
+   * - `kitchen_bench_lightwheel_pick_lemon_shelf_on_plate_u_shaped_with_island_farmhouse2.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_pick_lemon_shelf_on_plate_u_shaped_with_island_farmhouse2.yaml>`_
+     - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_pick_lemon_shelf_on_plate_lightwheel_kitchen_u_shaped_with_island_farmhouse2.gif
+          :alt: Pi policy picking a lemon from a shelf onto a plate in the U-shaped farmhouse kitchen
+          :width: 100%
    * - `kitchen_bench_lightwheel_place_bananas_in_wooden_bowl_u_shaped_with_island_farmhouse2.yaml <https://github.com/isaac-sim/IsaacLab-Arena/blob/main/isaaclab_arena_environments/kitchen_bench/kitchen_bench_lightwheel_place_bananas_in_wooden_bowl_u_shaped_with_island_farmhouse2.yaml>`_
      - .. image:: ../../images/kitchen_bench/kitchen_bench_lightwheel_place_bananas_in_wooden_bowl_u_shaped_with_island_farmhouse2_pi.gif
           :alt: Pi policy placing bananas in a wooden bowl in the U-shaped farmhouse kitchen

docs/images/kitchen_bench/kitchen_bench_lightwheel_open_fridge_g_shaped_large_scandinavian.gif

@@ -57,7 +57,7 @@
     ref_prim_path="/World/franka_table",
     payload_child_name="franka_table",
     footprint_translate_xyz=(-0.05, 0.0, 0.0),
-    footprint_scale_xy=(1.2, 1.2),
+    stand_default_footprint_xy_m=(1.08, 0.91032),
     stand_default_height=1.35,
 )
 _DROID_JOINT_NAMES = (
@@ -86,6 +86,7 @@ class DroidEmbodimentBase(EmbodimentBase, ABC):
     ``initial_pose`` / ``set_initial_pose`` set the base of the robot in world frame.
     ``stand_height_m`` sets the height of the stand mesh under the robot base link,
     which changes how far the stand extends below the root link.
+    ``stand_footprint_xy_m`` sets the stand footprint dimensions in the robot-base X/Y plane.
     When manually placing the robot on floor, ``set_initial_pose`` z value and
     ``stand_height_m`` should be adjusted together to keep the bottom of stand fixed.
     ``placement_bbox_stand_only`` uses the stand footprint for ``On`` / ``NextTo`` placement
@@ -103,6 +104,7 @@ def __init__(
         concatenate_observation_terms: bool = False,
         arm_mode: ArmMode | None = None,
         stand_height_m: float = _DROID_STAND_PRIM.stand_default_height,
+        stand_footprint_xy_m: tuple[float, float] | list[float] = _DROID_STAND_PRIM.stand_default_footprint_xy_m,
         placement_bbox_stand_only: bool = False,
         collision_mode: CollisionMode | str | None = None,
     ):
@@ -114,12 +116,18 @@ def __init__(
             collision_mode=collision_mode,
         )
         self.stand_height_m = stand_height_m
+        assert len(stand_footprint_xy_m) == 2, f"stand_footprint_xy_m must have 2 values, got {stand_footprint_xy_m!r}"
+        assert all(
+            value > 0.0 for value in stand_footprint_xy_m
+        ), f"stand_footprint_xy_m must be positive, got {stand_footprint_xy_m}"
+        self.stand_footprint_xy_m = tuple(stand_footprint_xy_m)
         self.placement_bbox_stand_only = placement_bbox_stand_only
         self.scene_config = DroidSceneCfg()
         self.scene_config.robot.spawn.usd_path = compose_on_stand_usd(
             _DROID_ROBOT_PRIM,
             _DROID_STAND_PRIM,
             stand_height_m=stand_height_m,
+            stand_footprint_xy_m=self.stand_footprint_xy_m,
             output_basename="droid_franka_robotiq_on_stand",
         )
         self.action_config = None
@@ -181,6 +189,7 @@ def __init__(
         concatenate_observation_terms: bool = False,
         arm_mode: ArmMode | None = None,
         stand_height_m: float = _DROID_STAND_PRIM.stand_default_height,
+        stand_footprint_xy_m: tuple[float, float] | list[float] = _DROID_STAND_PRIM.stand_default_footprint_xy_m,
         placement_bbox_stand_only: bool = False,
         collision_mode: CollisionMode | str | None = None,
     ):
@@ -191,6 +200,7 @@ def __init__(
             concatenate_observation_terms=concatenate_observation_terms,
             arm_mode=arm_mode,
             stand_height_m=stand_height_m,
+            stand_footprint_xy_m=stand_footprint_xy_m,
             placement_bbox_stand_only=placement_bbox_stand_only,
             collision_mode=collision_mode,
         )
@@ -212,6 +222,7 @@ def __init__(
         concatenate_observation_terms: bool = False,
         arm_mode: ArmMode | None = None,
         stand_height_m: float = _DROID_STAND_PRIM.stand_default_height,
+        stand_footprint_xy_m: tuple[float, float] | list[float] = _DROID_STAND_PRIM.stand_default_footprint_xy_m,
         placement_bbox_stand_only: bool = False,
         collision_mode: CollisionMode | str | None = None,
     ):
@@ -222,6 +233,7 @@ def __init__(
             concatenate_observation_terms=concatenate_observation_terms,
             arm_mode=arm_mode,
             stand_height_m=stand_height_m,
+            stand_footprint_xy_m=stand_footprint_xy_m,
             placement_bbox_stand_only=placement_bbox_stand_only,
             collision_mode=collision_mode,
         )
@@ -244,6 +256,7 @@ def __init__(
         concatenate_observation_terms: bool = False,
         arm_mode: ArmMode | None = None,
         stand_height_m: float = _DROID_STAND_PRIM.stand_default_height,
+        stand_footprint_xy_m: tuple[float, float] | list[float] = _DROID_STAND_PRIM.stand_default_footprint_xy_m,
         placement_bbox_stand_only: bool = False,
         collision_mode: CollisionMode | str | None = None,
     ):
@@ -254,6 +267,7 @@ def __init__(
             concatenate_observation_terms=concatenate_observation_terms,
             arm_mode=arm_mode,
             stand_height_m=stand_height_m,
+            stand_footprint_xy_m=stand_footprint_xy_m,
             placement_bbox_stand_only=placement_bbox_stand_only,
             collision_mode=collision_mode,
         )

docs/images/kitchen_bench/kitchen_bench_lightwheel_open_fridge_u_shaped_with_island_farmhouse2.gif

@@ -60,7 +60,7 @@
     ref_prim_path="/Stand",
     payload_child_name="Stand",
     footprint_translate_xyz=(-0.05, 0.0, 0.0),
-    footprint_scale_xy=(1.2, 1.2),
+    stand_default_footprint_xy_m=(0.3888, 0.3888),
     stand_default_height=0.8755,
 )
 _FRANKA_JOINT_NAMES = (

docs/images/kitchen_bench/kitchen_bench_lightwheel_pick_lemon_shelf_on_plate_lightwheel_kitchen_u_shaped_with_island_farmhouse2.gif

@@ -21,6 +21,7 @@
 _ROBOT_ON_STAND_USD_CACHE_DIR = "robot_on_stand"
 
 _HEIGHT_ATOL = 1e-3
+_FOOTPRINT_ATOL = 1e-3
 _ALIGN_ATOL = 5e-2
 
 
@@ -52,17 +53,17 @@ class StandPrimSpec:
     ref_prim_path: str
     payload_child_name: str
     footprint_translate_xyz: tuple[float, float, float]
-    footprint_scale_xy: tuple[float, float]
+    stand_default_footprint_xy_m: tuple[float, float]
     stand_default_height: float
 
 
-@functools.cache
 def compose_on_stand_usd(
     robot: RobotPrimSpec,
     stand: StandPrimSpec,
     *,
     stand_height_m: float,
     output_basename: str,
+    stand_footprint_xy_m: tuple[float, float] | None = None,
 ) -> str:
     """Build a robot+stand USD with stand under the robot base link.
 
@@ -74,15 +75,39 @@ def compose_on_stand_usd(
         stand: Stand reference and footprint parameters.
         stand_height_m: Target absolute stand height after align.
         output_basename: Stable basename for the composed USD (embodiment-specific).
+        stand_footprint_xy_m: Target footprint ``(x, y)`` dimensions in meters. ``None`` uses the default.
 
     Returns:
         Local path to the composed on-stand USD.
     """
     assert stand_height_m > 0.0, f"stand_height_m must be positive, got {stand_height_m}"
+    if stand_footprint_xy_m is None:
+        stand_footprint_xy_m = stand.stand_default_footprint_xy_m
+    assert len(stand_footprint_xy_m) == 2, f"stand_footprint_xy_m must have 2 values, got {stand_footprint_xy_m!r}"
+    assert all(
+        value > 0.0 for value in stand_footprint_xy_m
+    ), f"stand_footprint_xy_m must be positive, got {stand_footprint_xy_m}"
+    return _compose_on_stand_usd_cached(
+        robot,
+        stand,
+        stand_height_m,
+        output_basename,
+        stand_footprint_xy_m,
+    )
+
 
+@functools.cache
+def _compose_on_stand_usd_cached(
+    robot: RobotPrimSpec,
+    stand: StandPrimSpec,
+    stand_height_m: float,
+    output_basename: str,
+    stand_footprint_xy_m: tuple[float, float],
+) -> str:
     cache_root = get_arena_asset_cache_dir().parent / "usd" / _ROBOT_ON_STAND_USD_CACHE_DIR
     cache_root.mkdir(parents=True, exist_ok=True)
-    out_path = cache_root / f"{output_basename}_{stand_height_m:.3f}.usd"
+    footprint_x_m, footprint_y_m = stand_footprint_xy_m
+    out_path = cache_root / f"{output_basename}_{stand_height_m}_{footprint_x_m}x{footprint_y_m}.usd"
 
     with tempfile.NamedTemporaryFile(suffix=".usd", dir=cache_root, delete=False) as tmp_file:
         tmp_path = Path(tmp_file.name)
@@ -94,7 +119,7 @@ def compose_on_stand_usd(
         root.GetReferences().AddReference(robot_resolved, robot.root_prim_path)
         stage.SetDefaultPrim(root)
 
-        _mount_stand_normalized(stage, robot, stand, stand_height_m)
+        _add_stand(stage, robot, stand, stand_height_m, stand_footprint_xy_m)
         assert stage.GetRootLayer().Save(), f"failed to save composed on-stand USD to {tmp_path}"
         os.replace(tmp_path, out_path)
     except Exception:
@@ -104,19 +129,20 @@ def compose_on_stand_usd(
     return str(out_path)
 
 
-def _mount_stand_normalized(
+def _add_stand(
     stage: Usd.Stage,
     robot: RobotPrimSpec,
     stand: StandPrimSpec,
     stand_height_m: float,
+    stand_footprint_xy_m: tuple[float, float],
 ) -> None:
-    """Parent a stand payload under the robot base link, scale, and align to the robot base.
+    """Add a stand under the robot base link with the requested size and alignment.
 
     Composed USD hierarchy::
 
         /panda/panda_link0/stand_instanceable     # outer mount
           translate: (footprint_xy, align_z)
-          scale:    (footprint_xy, stand_height_m / native_height)
+          scale:    (stand_footprint_xy_m / native_footprint_xy_m, stand_height_m / native_height)
           /<payload_child_name>/                   # inner payload
             reference: stand USD @ ref_prim_path
 
@@ -125,6 +151,7 @@ def _mount_stand_normalized(
         robot: Robot prim layout under the composed default prim.
         stand: Stand reference and footprint parameters.
         stand_height_m: Target stand height after align.
+        stand_footprint_xy_m: Target footprint ``(x, y)`` dimensions in meters.
     """
     # Robot-only stage: bottom of the base link is the align target (before stand exists).
     robot_base = stage.GetPrimAtPath(robot.robot_base_prim_path)
@@ -136,36 +163,41 @@ def _mount_stand_normalized(
 
     stand_resolved = retrieve_file_path(stand.stand_usd_path)
     tx, ty, tz = stand.footprint_translate_xyz
-    sx, sy = stand.footprint_scale_xy
+    footprint_x_m, footprint_y_m = stand_footprint_xy_m
     stand_prim_path = robot.stand_prim_path
 
-    # Outer mount under link0; Z scale stays at 1 until native height is measured.
+    # Outer mount under link0 stays at unit scale until the source stand dimensions are measured.
     stand_xf = UsdGeom.Xform.Define(stage, stand_prim_path)
     translate_op = stand_xf.AddTranslateOp()
     translate_op.Set(Gf.Vec3d(tx, ty, tz))
     scale_op = stand_xf.AddScaleOp()
-    scale_op.Set(Gf.Vec3d(sx, sy, 1.0))
+    scale_op.Set(Gf.Vec3d(1.0, 1.0, 1.0))
 
     payload_prim = stage.DefinePrim(f"{stand_prim_path}/{stand.payload_child_name}")
     payload_prim.GetReferences().AddReference(stand_resolved, stand.ref_prim_path)
 
     stand_prim = stand_xf.GetPrim()
     bbox_cache = UsdGeom.BBoxCache(Usd.TimeCode.Default(), [UsdGeom.Tokens.default_])
 
-    # Native height at footprint XY scale; then scale Z to the requested stand height.
+    # Scale the source stand to the requested physical dimensions.
     stand_range = bbox_cache.ComputeWorldBound(stand_prim).ComputeAlignedRange()
     assert not stand_range.IsEmpty(), f"empty stand bounds at {stand_prim.GetPath()}"
-    native_height = float(stand_range.GetSize()[2])
+    native_size = stand_range.GetSize()
+    native_x_m, native_y_m, native_height = (float(value) for value in native_size)
+    assert native_x_m > 0.0 and native_y_m > 0.0, f"non-positive stand footprint at {stand_prim.GetPath()}"
     assert native_height > 0.0, f"non-positive stand height at {stand_prim.GetPath()}"
-    scale_op.Set(Gf.Vec3d(sx, sy, stand_height_m / native_height))
+    scale_op.Set(Gf.Vec3d(footprint_x_m / native_x_m, footprint_y_m / native_y_m, stand_height_m / native_height))
 
     # Raise/lower outer translate so stand top meets robot_min_z.
     translate_op.Set(Gf.Vec3d(tx, ty, robot_min_z))
 
     # Verify stand/robot alignment and stand height on a fresh BBoxCache.
     verify_cache = UsdGeom.BBoxCache(Usd.TimeCode.Default(), [UsdGeom.Tokens.default_])
     stand_range = verify_cache.ComputeWorldBound(stand_prim).ComputeAlignedRange()
-    stand_height = float(stand_range.GetSize()[2])
+    stand_size = stand_range.GetSize()
+    stand_x_m, stand_y_m, stand_height = (float(value) for value in stand_size)
     stand_max_z = float(stand_range.GetMax()[2])
+    assert abs(stand_x_m - footprint_x_m) < _FOOTPRINT_ATOL, stand_x_m
+    assert abs(stand_y_m - footprint_y_m) < _FOOTPRINT_ATOL, stand_y_m
     assert abs(stand_height - stand_height_m) < _HEIGHT_ATOL, stand_height
     assert abs(stand_max_z - robot_min_z) < _ALIGN_ATOL, (stand_max_z, robot_min_z)