diff --git a/isaaclab_arena/assets/background_library.py b/isaaclab_arena/assets/background_library.py
index 4e5639bcfb..7fcaf01c9a 100644
--- a/isaaclab_arena/assets/background_library.py
+++ b/isaaclab_arena/assets/background_library.py
@@ -173,7 +173,12 @@ class LightwheelKitchenBackground(LibraryBackground):
     initial_pose = Pose.identity()
     object_min_z = -0.2
 
-    def __init__(self, layout_id: int = 1, style_id: int = 1):
+    def __init__(
+        self,
+        layout_id: int = 1,
+        style_id: int = 1,
+        **kwargs,
+    ):
         from lightwheel_sdk.loader import floorplan_loader
 
         # Lazily download the USD
@@ -188,7 +193,7 @@ def __init__(self, layout_id: int = 1, style_id: int = 1):
                 backend="robocasa",
             )[0]
         )
-        super().__init__()
+        super().__init__(**kwargs)
 
     def get_viewer_cfg(self) -> ViewerCfg:
         # Looking in through the open front.
diff --git a/isaaclab_arena/environments/relation_solver_interface.py b/isaaclab_arena/environments/relation_solver_interface.py
index c8730b60ce..a0dc93974d 100644
--- a/isaaclab_arena/environments/relation_solver_interface.py
+++ b/isaaclab_arena/environments/relation_solver_interface.py
@@ -11,7 +11,7 @@
 
 from isaaclab_arena.relations.collision_mode import CollisionMode, get_object_collision_mode
 from isaaclab_arena.relations.object_placer_params import ObjectPlacerParams
-from isaaclab_arena.relations.placement_events import get_pose_from_layout, solve_and_place_objects
+from isaaclab_arena.relations.placement_events import PlacementPoolHandle, get_pose_from_layout, solve_and_place_objects
 from isaaclab_arena.relations.pooled_object_placer import PooledObjectPlacer
 from isaaclab_arena.relations.relations import get_anchor_objects
 from isaaclab_arena.utils.pose import PosePerEnv
@@ -184,8 +184,7 @@ def _apply_dynamic_spawn_pose(
         func=solve_and_place_objects,
         mode="reset",
         params={
-            "assets": assets,
-            "placement_pool": placement_pool,
+            "placement_pool": PlacementPoolHandle(placement_pool),
         },
     )
 
diff --git a/isaaclab_arena/relations/placement_asset.py b/isaaclab_arena/relations/placement_asset.py
index c659f54979..02a191ab88 100644
--- a/isaaclab_arena/relations/placement_asset.py
+++ b/isaaclab_arena/relations/placement_asset.py
@@ -27,16 +27,25 @@
 class PlaceableAsset(Asset, ABC):
     """Asset whose root pose can be constrained by spatial relations."""
 
-    def __init__(self, name: str, tags: list[str] | None = None, **kwargs) -> None:
+    def __init__(
+        self,
+        name: str,
+        tags: list[str] | None = None,
+        collision_mode: CollisionMode | str | None = None,
+        repair_collision_mesh_non_watertight: bool = True,
+        **kwargs,
+    ) -> None:
         super().__init__(name=name, tags=tags, **kwargs)
         self.initial_pose: Pose | PoseRange | PosePerEnv | None = None
         self._pose_event_cfg: EventTermCfg | None = None
         """Reset event restoring this asset's root pose; ``None`` until a pose with a reset event is set."""
         self.relations: list[RelationBase] = []
         # None delegates collision-mode selection to the solver.
-        self.collision_mode: CollisionMode | None = None
+        if collision_mode is not None:
+            collision_mode = CollisionMode(collision_mode)
+        self.collision_mode = collision_mode
         # Whether to replace a non-watertight collision mesh with its convex hull.
-        self.repair_collision_mesh_non_watertight = True
+        self.repair_collision_mesh_non_watertight = repair_collision_mesh_non_watertight
 
     def add_relation(self, relation: RelationBase) -> None:
         """Attach a relation to the asset."""
diff --git a/isaaclab_arena/relations/placement_events.py b/isaaclab_arena/relations/placement_events.py
index bed1ca3d7e..49cf6315e4 100644
--- a/isaaclab_arena/relations/placement_events.py
+++ b/isaaclab_arena/relations/placement_events.py
@@ -26,6 +26,32 @@
 PLACEMENT_RESET_EVENT_NAME = "placement_reset"
 
 
+class PlacementPoolHandle:
+    """Opaque holder for a runtime placement pool to bypass EventTermCfg param deepcopy/validation errors.
+
+    PooledObjectPlacer is used as an EventTermCfg param to set the initial spawn pose. Isaac Lab deep-copies
+    and validates the configclass param, leading to two crashes: deepcopy fails for the Warp GPU cache
+    wp.Mesh BVHs ("ctypes objects containing pointers cannot be pickled"); validation hits RecursionError
+    when recursively walking all dicts and reaches placement assets (including embodiments with cyclic
+    scene configs).
+
+    This handle wraps PooledObjectPlacer with overrides for deepcopy and validation, while
+    PooledObjectPlacer itself stays a normal class. EventTermCfg params use this handle.
+    """
+
+    __slots__ = ("pool",)
+    """Store pool in a slot instead of an instance dictionary; ``hasattr(handle, "__dict__")`` is false,
+    so ``_validate(handle)`` stops traversing into PooledObjectPlacer ."""
+
+    def __init__(self, pool: PooledObjectPlacer) -> None:
+        self.pool = pool
+
+    def __deepcopy__(self, memo: dict[int, object]) -> PlacementPoolHandle:
+        """Share the live pool across ``copy.deepcopy`` to avoid deep-copying the Warp cache BVHs."""
+        memo[id(self)] = self
+        return self
+
+
 def get_placement_pool(env) -> PooledObjectPlacer | None:
     """Return the pooled placer stored on the env reset event, or ``None`` when absent.
 
@@ -39,7 +65,9 @@ def get_placement_pool(env) -> PooledObjectPlacer | None:
         term_cfg = env.unwrapped.event_manager.get_term_cfg(PLACEMENT_RESET_EVENT_NAME)
     except ValueError:
         return None
-    return term_cfg.params.get("placement_pool")
+    handle = term_cfg.params.get("placement_pool")
+    assert handle is not None, f"'{PLACEMENT_RESET_EVENT_NAME}' event is missing its placement_pool parameter."
+    return handle.pool
 
 
 def get_rotation_xyzw(asset: PlaceableAsset) -> tuple[float, float, float, float]:
@@ -113,8 +141,7 @@ def write_layout_to_sim(
 def solve_and_place_objects(
     env: ManagerBasedEnv,
     env_ids: torch.Tensor | None,
-    assets: list[PlaceableAsset],
-    placement_pool: PooledObjectPlacer,
+    placement_pool: PlacementPoolHandle,
 ) -> None:
     """Coordinated reset event that draws layouts from the pool and writes poses.
 
@@ -125,17 +152,19 @@ def solve_and_place_objects(
     Args:
         env: The Isaac Lab environment.
         env_ids: 1-D tensor of environment indices being reset.
-        assets: Assets participating in relation solving.
-        placement_pool: Runtime pool of solved placement layouts.
+        placement_pool: Opaque handle to the runtime pool of solved placement layouts.
+            Layout assets come from ``placement_pool.pool.objects``.
     """
+    pool = placement_pool.pool
     if env_ids is None or len(env_ids) == 0:
         return
+    assets = pool.objects
     reset_env_ids = env_ids.tolist()
     num_scene_envs = env.scene.env_origins.shape[0]
     assert (
-        placement_pool.num_envs == num_scene_envs
-    ), f"Placement pool has {placement_pool.num_envs} envs, but scene has {num_scene_envs} env origins."
-    results_by_env = placement_pool.sample_for_envs(reset_env_ids)
+        pool.num_envs == num_scene_envs
+    ), f"Placement pool has {pool.num_envs} envs, but scene has {num_scene_envs} env origins."
+    results_by_env = pool.sample_for_envs(reset_env_ids)
     anchor_assets = set(get_anchor_objects(assets))
     base_rotations = get_base_rotation_per_asset(assets)
 
diff --git a/isaaclab_arena/tests/test_placement_events.py b/isaaclab_arena/tests/test_placement_events.py
index 29b7dfbeda..8ccd23051d 100644
--- a/isaaclab_arena/tests/test_placement_events.py
+++ b/isaaclab_arena/tests/test_placement_events.py
@@ -149,16 +149,11 @@ def scene_getitem(self, name: str) -> MagicMock:
     return env
 
 
-def _solve_and_place_with_pool(env, env_ids, objects, pool):
+def _solve_and_place_with_pool(env, env_ids, pool):
     """Call the reset event with the same runtime params EventTermCfg stores."""
-    from isaaclab_arena.relations.placement_events import solve_and_place_objects
+    from isaaclab_arena.relations.placement_events import PlacementPoolHandle, solve_and_place_objects
 
-    return solve_and_place_objects(
-        env,
-        env_ids,
-        assets=objects,
-        placement_pool=pool,
-    )
+    return solve_and_place_objects(env, env_ids, placement_pool=PlacementPoolHandle(pool))
 
 
 def test_solve_and_place_objects_writes_poses_to_sim():
@@ -176,7 +171,7 @@ def test_solve_and_place_objects_writes_poses_to_sim():
     placer_params = ObjectPlacerParams(solver_params=solver_params)
     pool = PooledObjectPlacer(objects=objects, placer_params=placer_params, pool_size=10)
 
-    _solve_and_place_with_pool(env, env_ids, objects, pool)
+    _solve_and_place_with_pool(env, env_ids, pool)
 
     # Anchor (desk) should NOT have been written.
     assert "desk" not in env._assets, "Anchor pose should not be written to sim"
@@ -191,7 +186,7 @@ def test_solve_and_place_objects_writes_poses_to_sim():
 
 
 def test_solve_and_place_objects_uses_runtime_pool():
-    from isaaclab_arena.relations.placement_events import solve_and_place_objects
+    from isaaclab_arena.relations.placement_events import PlacementPoolHandle, solve_and_place_objects
     from isaaclab_arena.relations.placement_result import PlacementResult
     from isaaclab_arena.tests.dummy_embodiment import DummyEmbodiment
     from isaaclab_arena.utils.bounding_box import AxisAlignedBoundingBox
@@ -206,6 +201,7 @@ def test_solve_and_place_objects_uses_runtime_pool():
 
     class Pool:
         num_envs = 1
+        objects = [desk, robot]
 
         def sample_for_envs(self, env_ids: list[int]) -> dict[int, PlacementResult]:
             assert env_ids == [0]
@@ -221,8 +217,7 @@ def sample_for_envs(self, env_ids: list[int]) -> dict[int, PlacementResult]:
     solve_and_place_objects(
         env,
         torch.tensor([0]),
-        assets=[desk, robot],
-        placement_pool=Pool(),
+        placement_pool=PlacementPoolHandle(Pool()),
     )
 
     assert "desk" not in env._assets
@@ -296,14 +291,14 @@ def test_reset_placement_asset_pose_per_env_requires_full_env_coverage():
 
 
 def test_get_placement_pool_returns_runtime_pool():
-    from isaaclab_arena.relations.placement_events import get_placement_pool
+    from isaaclab_arena.relations.placement_events import PlacementPoolHandle, get_placement_pool
 
     class Pool:
         pass
 
     pool = Pool()
     env = MagicMock()
-    env.unwrapped.event_manager.get_term_cfg.return_value.params = {"placement_pool": pool}
+    env.unwrapped.event_manager.get_term_cfg.return_value.params = {"placement_pool": PlacementPoolHandle(pool)}
     assert get_placement_pool(env) is pool
 
 
@@ -328,7 +323,7 @@ def test_solve_and_place_objects_applies_random_yaw():
     )
     pool = PooledObjectPlacer(objects=objects, placer_params=placer_params, pool_size=10)
 
-    _solve_and_place_with_pool(env, env_ids, objects, pool)
+    _solve_and_place_with_pool(env, env_ids, pool)
 
     # Anchor (desk) is never rotated or written, even with random yaw enabled.
     assert "desk" not in env._assets, "Anchor pose should not be written to sim"
@@ -356,7 +351,7 @@ def test_solve_and_place_objects_skips_empty_env_ids():
     placer_params = ObjectPlacerParams(solver_params=solver_params)
     pool = PooledObjectPlacer(objects=[desk, box1, box2], placer_params=placer_params, pool_size=10)
 
-    _solve_and_place_with_pool(env, torch.tensor([], dtype=torch.int64), [desk, box1, box2], pool)
+    _solve_and_place_with_pool(env, torch.tensor([], dtype=torch.int64), pool)
 
     assert len(env._assets) == 0, "No writes should occur for empty env_ids"
 
@@ -373,7 +368,7 @@ def test_solve_and_place_objects_skips_none_env_ids():
     placer_params = ObjectPlacerParams(solver_params=solver_params)
     pool = PooledObjectPlacer(objects=[desk, box1, box2], placer_params=placer_params, pool_size=10)
 
-    _solve_and_place_with_pool(env, None, [desk, box1, box2], pool)
+    _solve_and_place_with_pool(env, None, pool)
 
     assert len(env._assets) == 0, "No writes should occur for None env_ids"
 
@@ -394,7 +389,7 @@ def test_solve_and_place_objects_handles_multiple_env_ids():
     placer_params = ObjectPlacerParams(solver_params=solver_params)
     pool = PooledObjectPlacer(objects=objects, placer_params=placer_params, pool_size=12, num_envs=num_envs)
 
-    _solve_and_place_with_pool(env, env_ids, objects, pool)
+    _solve_and_place_with_pool(env, env_ids, pool)
 
     assert "desk" not in env._assets, "Anchor pose should not be written to sim"
 
@@ -424,7 +419,7 @@ def test_solve_and_place_objects_partial_reset_homogeneous_pool_consumes_only_re
     pool = PooledObjectPlacer(objects=objects, placer_params=placer_params, pool_size=12, num_envs=num_envs)
 
     available_before = pool.total_remaining
-    _solve_and_place_with_pool(env, env_ids, objects, pool)
+    _solve_and_place_with_pool(env, env_ids, pool)
     available_after = pool.total_remaining
 
     assert available_before - available_after == len(env_ids)
@@ -436,11 +431,11 @@ def test_solve_and_place_objects_writes_invalid_fallback_layout(capsys):
     from isaaclab_arena.relations.placement_result import PlacementResult
 
     desk, box1, box2 = _create_test_objects()
-    objects = [desk, box1, box2]
     env = _make_mock_env(num_envs=1)
 
     class InvalidPool:
         num_envs = 1
+        objects = [desk, box1, box2]
 
         def sample_for_envs(self, env_ids: list[int]) -> dict[int, PlacementResult]:
             assert env_ids == [0]
@@ -453,7 +448,7 @@ def sample_for_envs(self, env_ids: list[int]) -> dict[int, PlacementResult]:
                 )
             }
 
-    _solve_and_place_with_pool(env, torch.tensor([0]), objects, InvalidPool())
+    _solve_and_place_with_pool(env, torch.tensor([0]), InvalidPool())
     captured = capsys.readouterr()
 
     assert set(env._assets) == {box1.name, box2.name}
@@ -466,12 +461,12 @@ def test_solve_and_place_objects_partial_reset_applies_absolute_env_origin():
     from isaaclab_arena.relations.placement_result import PlacementResult
 
     desk, box1, box2 = _create_test_objects()
-    objects = [desk, box1, box2]
     env = _make_mock_env(num_envs=4)
     env.scene.env_origins[2] = torch.tensor([10.0, 0.0, 0.0])
 
     class EnvIndexedPool:
         num_envs = 4
+        objects = [desk, box1, box2]
         requested_env_ids = None
 
         def sample_without_replacement(self, count: int) -> list[PlacementResult]:
@@ -493,7 +488,7 @@ def sample_for_envs(self, env_ids: list[int]) -> dict[int, PlacementResult]:
             }
 
     pool = EnvIndexedPool()
-    _solve_and_place_with_pool(env, torch.tensor([2]), objects, pool)
+    _solve_and_place_with_pool(env, torch.tensor([2]), pool)
 
     box1_pose = env._assets[box1.name].write_root_pose_to_sim.call_args[0][0]
     box2_pose = env._assets[box2.name].write_root_pose_to_sim.call_args[0][0]
@@ -510,14 +505,14 @@ def test_solve_and_place_objects_asserts_env_indexed_pool_size_matches_scene():
     """Env-indexed pool slots must line up with absolute Isaac Lab env ids."""
 
     desk, box1, box2 = _create_test_objects()
-    objects = [desk, box1, box2]
     env = _make_mock_env(num_envs=2)
 
     class MismatchedEnvIndexedPool:
         num_envs = 1
+        objects = [desk, box1, box2]
 
     with pytest.raises(AssertionError, match="scene has 2 env origins"):
-        _solve_and_place_with_pool(env, torch.tensor([0]), objects, MismatchedEnvIndexedPool())
+        _solve_and_place_with_pool(env, torch.tensor([0]), MismatchedEnvIndexedPool())
 
 
 def test_pooled_placer_sample_without_replacement_returns_different_layouts():
@@ -982,6 +977,13 @@ def test_solve_and_apply_relation_placement_drops_embodiment_from_event_params()
     assert params.reachability_config.embodiment is embodiment
     # ...while the pool the reset event captured no longer references the embodiment -- on the placer params
     # and on every built validator alike -- so configclass never deep-copies or recurses into it.
-    pool = event.params["placement_pool"]
+    from isaaclab.utils.configclass import _validate
+
+    from isaaclab_arena.relations.placement_events import PlacementPoolHandle
+
+    pool_handle = event.params["placement_pool"]
+    assert isinstance(pool_handle, PlacementPoolHandle)
+    _validate(event, prefix="")
+    pool = pool_handle.pool
     assert pool._placer.params.reachability_config.embodiment is None
     assert all(v._params.reachability_config.embodiment is None for v in pool._placer._validators)
diff --git a/isaaclab_arena/tests/test_relation_solver_interface.py b/isaaclab_arena/tests/test_relation_solver_interface.py
index 561a446770..a0af67fc02 100644
--- a/isaaclab_arena/tests/test_relation_solver_interface.py
+++ b/isaaclab_arena/tests/test_relation_solver_interface.py
@@ -34,8 +34,13 @@ def _make_box(name: str = "box"):
 
 
 class _FakePlacementPool:
-    def __init__(self, layouts) -> None:
+    def __init__(self, layouts, objects=None) -> None:
         self._layouts = layouts
+        self._objects = objects or []
+
+    @property
+    def objects(self):
+        return self._objects
 
     def sample_with_replacement(self, count: int):
         return self._layouts[:count]
@@ -142,7 +147,10 @@ def test_dynamic_spawn_pose_event_params_use_runtime_assets():
 
     desk = _make_desk()
     box = _make_box()
-    placement_pool = _FakePlacementPool([_fallback_layout(positions={box: (0.1, 0.2, 0.3)})])
+    placement_pool = _FakePlacementPool(
+        [_fallback_layout(positions={box: (0.1, 0.2, 0.3)})],
+        objects=[desk, box],
+    )
 
     event_cfg = _apply_dynamic_spawn_pose(
         assets=[desk, box],
@@ -150,8 +158,48 @@ def test_dynamic_spawn_pose_event_params_use_runtime_assets():
         anchor_assets={desk},
     )
 
-    assert [asset.name for asset in event_cfg.params["assets"]] == ["desk", "box"]
     assert "placement_pool" in event_cfg.params
+    from isaaclab_arena.relations.placement_events import PlacementPoolHandle
+
+    assert isinstance(event_cfg.params["placement_pool"], PlacementPoolHandle)
+    assert [asset.name for asset in event_cfg.params["placement_pool"].pool.objects] == [
+        "desk",
+        "box",
+    ]
+
+
+def test_dynamic_spawn_pose_event_cfg_deepcopy_after_mesh_solve():
+    """EventTermCfg deep-copies params; handle shares the pool across config copies."""
+    import copy
+    import trimesh
+
+    from isaaclab.managers import EventTermCfg
+
+    from isaaclab_arena.environments.relation_solver_interface import solve_and_apply_relation_placement
+    from isaaclab_arena.relations.object_placer_params import ObjectPlacerParams
+    from isaaclab_arena.relations.relation_solver_params import CollisionMode, RelationSolverParams
+    from isaaclab_arena.relations.relations import On
+
+    desk = _make_desk()
+    box = _make_box()
+    box.add_relation(On(desk, clearance_m=0.01))
+    box.collision_mode = CollisionMode.MESH
+    box._collision_mesh = trimesh.creation.box(extents=(0.2, 0.2, 0.2))
+
+    params = ObjectPlacerParams(
+        placement_seed=17,
+        resolve_on_reset=True,
+        min_unique_layouts_per_env=1,
+        solver_params=RelationSolverParams(collision_mode=CollisionMode.MESH, max_iters=50),
+    )
+    event_cfg = solve_and_apply_relation_placement([desk, box], num_envs=1, placer_params=params)
+
+    assert event_cfg is not None
+    assert isinstance(event_cfg, EventTermCfg)
+    copy.deepcopy(event_cfg)
+    from isaaclab.utils.configclass import _validate
+
+    _validate(event_cfg, prefix="")
 
 
 def test_static_embodiment_placement_stores_per_env_poses():
diff --git a/isaaclab_arena_environments/kitchen_bench/droid_pick_and_place_lightwheel_kitchen.yaml b/isaaclab_arena_environments/kitchen_bench/droid_pick_and_place_lightwheel_kitchen.yaml
index 81827a2156..15d0836214 100644
--- a/isaaclab_arena_environments/kitchen_bench/droid_pick_and_place_lightwheel_kitchen.yaml
+++ b/isaaclab_arena_environments/kitchen_bench/droid_pick_and_place_lightwheel_kitchen.yaml
@@ -13,7 +13,9 @@ embodiment:
 background:
   id: kitchen
   registry_name: lightwheel_robocasa_kitchen
-  params: {}
+  params:
+    collision_mode: mesh
+    repair_collision_mesh_non_watertight: false
 objects:
 - id: mustard_bottle
   registry_name: mustard_bottle_hope_robolab