diff --git a/scripts/benchmark/atomic_action/move_end_effector_benchmark.py b/scripts/benchmark/atomic_action/move_end_effector_benchmark.py
index 8cd4a75c7..b1d1dea97 100644
--- a/scripts/benchmark/atomic_action/move_end_effector_benchmark.py
+++ b/scripts/benchmark/atomic_action/move_end_effector_benchmark.py
@@ -60,6 +60,7 @@ class PoseCase:
 }
 DEFAULT_POSE_CASES = tuple(POSE_CASES.keys())
 MOVE_SAMPLE_INTERVAL = 80
+# Endpoint accuracy of the planned trajectory against the commanded pose.
 SUCCESS_TOLERANCE_M = 0.01
 
 
@@ -142,7 +143,8 @@ def _run_case(
                     binding=binding,
                     motion_policy=MotionPolicy(sample_count=MOVE_SAMPLE_INTERVAL),
                 ),
-            )
+            ),
+            atomic_engine.initial_context(control_dt=sim.sim_config.physics_dt),
         )
     )
     is_success = bool(result.plan_success.all().item())
@@ -321,7 +323,9 @@ def main() -> None:
 
 
 if __name__ == "__main__":
-    main()
+    from scripts.tutorials.atomic_action.tutorial_utils import run_tutorial
+
+    run_tutorial(main)
 
 
 __all__ = ["add_benchmark_args", "run_all_benchmarks"]
diff --git a/scripts/benchmark/atomic_action/move_held_object_benchmark.py b/scripts/benchmark/atomic_action/move_held_object_benchmark.py
index 8f577ed52..c71a7ee93 100644
--- a/scripts/benchmark/atomic_action/move_held_object_benchmark.py
+++ b/scripts/benchmark/atomic_action/move_held_object_benchmark.py
@@ -191,7 +191,7 @@ def _prepare_held_state(
         make_pre_pick_eef_pose,
     )
 
-    hand_open, hand_close = get_hand_open_close_qpos(robot, sim.device)
+    hand_open, hand_close = get_hand_open_close_qpos(robot)
     pickup_args = _make_pickup_args(args, object_preset, profile)
     atomic_engine = AtomicActionEngine(
         motion_generator=motion_gen,
@@ -790,7 +790,9 @@ def main() -> None:
 
 
 if __name__ == "__main__":
-    main()
+    from scripts.tutorials.atomic_action.tutorial_utils import run_tutorial
+
+    run_tutorial(main)
 
 
 __all__ = ["add_benchmark_args", "run_all_benchmarks"]
diff --git a/scripts/benchmark/atomic_action/move_joints_benchmark.py b/scripts/benchmark/atomic_action/move_joints_benchmark.py
index 9d53c20c5..6b0774429 100644
--- a/scripts/benchmark/atomic_action/move_joints_benchmark.py
+++ b/scripts/benchmark/atomic_action/move_joints_benchmark.py
@@ -151,7 +151,10 @@ def _run_case(
     reset_robot(robot, initial_qpos)
     steps = _targets_for_sequence(atomic_engine, case, sim.device)
     elapsed, mem_delta, peak_gpu, result = timed_call(
-        lambda: atomic_engine.compile(steps)
+        lambda: atomic_engine.compile(
+            steps,
+            atomic_engine.initial_context(control_dt=sim.sim_config.physics_dt),
+        )
     )
     is_success = bool(result.plan_success.all().item())
     traj = result.trajectory.positions
@@ -333,7 +336,9 @@ def main() -> None:
 
 
 if __name__ == "__main__":
-    main()
+    from scripts.tutorials.atomic_action.tutorial_utils import run_tutorial
+
+    run_tutorial(main)
 
 
 __all__ = ["add_benchmark_args", "run_all_benchmarks"]
diff --git a/scripts/benchmark/atomic_action/pickup_benchmark.py b/scripts/benchmark/atomic_action/pickup_benchmark.py
index 2bd6a1acb..eaecbe45b 100644
--- a/scripts/benchmark/atomic_action/pickup_benchmark.py
+++ b/scripts/benchmark/atomic_action/pickup_benchmark.py
@@ -150,7 +150,7 @@ def _run_case(
             settle_steps=2,
         )
         initial_obj_position = object_position_tuple(obj)
-        hand_open, hand_close = get_hand_open_close_qpos(robot, sim.device)
+        hand_open, hand_close = get_hand_open_close_qpos(robot)
         initialize_pre_pick_robot_pose(robot, obj, hand_open)
         case_args = _make_pickup_args(args, approach, object_preset, profile)
         approach_direction = resolve_pickup_approach_direction(
@@ -595,7 +595,9 @@ def main() -> None:
 
 
 if __name__ == "__main__":
-    main()
+    from scripts.tutorials.atomic_action.tutorial_utils import run_tutorial
+
+    run_tutorial(main)
 
 
 __all__ = ["add_benchmark_args", "run_all_benchmarks"]
diff --git a/scripts/benchmark/atomic_action/place_benchmark.py b/scripts/benchmark/atomic_action/place_benchmark.py
index 4ad32f1d6..37419eaa9 100644
--- a/scripts/benchmark/atomic_action/place_benchmark.py
+++ b/scripts/benchmark/atomic_action/place_benchmark.py
@@ -189,7 +189,7 @@ def _prepare_held_state(
         initialize_pre_pick_robot_pose,
     )
 
-    hand_open, hand_close = get_hand_open_close_qpos(robot, sim.device)
+    hand_open, hand_close = get_hand_open_close_qpos(robot)
     initialize_pre_pick_robot_pose(robot, obj, hand_open)
     pickup_args = _make_pickup_args(args, object_preset, profile)
     atomic_engine = AtomicActionEngine(
@@ -769,7 +769,9 @@ def main() -> None:
 
 
 if __name__ == "__main__":
-    main()
+    from scripts.tutorials.atomic_action.tutorial_utils import run_tutorial
+
+    run_tutorial(main)
 
 
 __all__ = ["add_benchmark_args", "run_all_benchmarks"]
diff --git a/scripts/tutorials/atomic_action/move_end_effector.py b/scripts/tutorials/atomic_action/move_end_effector.py
index 6aa4bfe1c..65114dc37 100644
--- a/scripts/tutorials/atomic_action/move_end_effector.py
+++ b/scripts/tutorials/atomic_action/move_end_effector.py
@@ -46,6 +46,9 @@
     replay_trajectory,
     run_tutorial,
 )
+from scripts.tutorials.atomic_action.tutorial_utils import (
+    initialize_benchmark_simulation,
+)
 
 MOVE_SAMPLE_INTERVAL = 80
 POST_TRAJECTORY_STEPS = 120
@@ -128,3 +131,16 @@ def main() -> None:
 
 if __name__ == "__main__":
     run_tutorial(main)
+
+
+def initialize_simulation(args) -> "SimulationManager":
+    """Create the tutorial simulation for interactive or benchmark runs."""
+    return initialize_benchmark_simulation(args)
+
+
+def create_robot(sim: "SimulationManager") -> "Robot":
+    """Add the default MoveEndEffector tutorial robot."""
+    robot = add_tutorial_robot(sim, "ur5")
+    # DexSim binds body_data (and therefore get_qpos) only after prepare().
+    sim.prepare()
+    return robot
diff --git a/scripts/tutorials/atomic_action/move_held_object.py b/scripts/tutorials/atomic_action/move_held_object.py
index 4341110e0..8b98d1a19 100644
--- a/scripts/tutorials/atomic_action/move_held_object.py
+++ b/scripts/tutorials/atomic_action/move_held_object.py
@@ -60,6 +60,11 @@
     replay_trajectory,
     run_tutorial,
 )
+from scripts.tutorials.atomic_action.tutorial_utils import (
+    compute_pick_close_end_step,
+    initialize_benchmark_simulation,
+    make_eef_pose_at,
+)
 
 OBJECT_MESH_PATH = "PaperCup/paper_cup.ply"
 OBJECT_XY = (-0.42, -0.08)
@@ -226,3 +231,21 @@ def main() -> None:
 
 if __name__ == "__main__":
     run_tutorial(main)
+
+
+def initialize_simulation(args) -> "SimulationManager":
+    """Create the tutorial simulation for interactive or benchmark runs."""
+    return initialize_benchmark_simulation(args)
+
+
+def create_robot(sim: "SimulationManager") -> "Robot":
+    """Add the default MoveHeldObject tutorial robot."""
+    robot = add_tutorial_robot(sim, "ur5")
+    # DexSim binds body_data (and therefore get_qpos) only after prepare().
+    sim.prepare()
+    return robot
+
+
+def make_pre_pick_eef_pose(robot: "Robot", position) -> "torch.Tensor":
+    """Build the canonical top-down EEF pose at ``position``."""
+    return make_eef_pose_at(robot, position)
diff --git a/scripts/tutorials/atomic_action/move_joints.py b/scripts/tutorials/atomic_action/move_joints.py
index c1a315dc9..b0bd44336 100644
--- a/scripts/tutorials/atomic_action/move_joints.py
+++ b/scripts/tutorials/atomic_action/move_joints.py
@@ -45,6 +45,9 @@
     replay_trajectory,
     run_tutorial,
 )
+from scripts.tutorials.atomic_action.tutorial_utils import (
+    initialize_benchmark_simulation,
+)
 
 MOVE_JOINTS_SAMPLE_INTERVAL = 80
 POST_TRAJECTORY_STEPS = 120
@@ -144,3 +147,16 @@ def offset_from_home(offsets: tuple[float, ...]) -> torch.Tensor:
 
 if __name__ == "__main__":
     run_tutorial(main)
+
+
+def initialize_simulation(args) -> "SimulationManager":
+    """Create the tutorial simulation for interactive or benchmark runs."""
+    return initialize_benchmark_simulation(args)
+
+
+def create_robot(sim: "SimulationManager") -> "Robot":
+    """Add the default MoveJoints tutorial robot."""
+    robot = add_tutorial_robot(sim, "ur5")
+    # DexSim binds body_data (and therefore get_qpos) only after prepare().
+    sim.prepare()
+    return robot
diff --git a/scripts/tutorials/atomic_action/pickup.py b/scripts/tutorials/atomic_action/pickup.py
index e35bc902a..c85bf7760 100644
--- a/scripts/tutorials/atomic_action/pickup.py
+++ b/scripts/tutorials/atomic_action/pickup.py
@@ -56,6 +56,10 @@
     replay_trajectory,
     run_tutorial,
 )
+from scripts.tutorials.atomic_action.tutorial_utils import (
+    compute_pick_close_end_step,
+    initialize_benchmark_simulation,
+)
 
 OBJECT_SIZE = (0.05, 0.05, 0.05)
 OBJECT_XY = (-0.42, -0.08)
@@ -205,3 +209,16 @@ def main() -> None:
 
 if __name__ == "__main__":
     run_tutorial(main)
+
+
+def initialize_simulation(args) -> "SimulationManager":
+    """Create the tutorial simulation for interactive or benchmark runs."""
+    return initialize_benchmark_simulation(args)
+
+
+def create_robot(sim: "SimulationManager") -> "Robot":
+    """Add the default PickUp tutorial robot."""
+    robot = add_tutorial_robot(sim, "ur5", tcp_z=0.15)
+    # DexSim binds body_data (and therefore get_qpos) only after prepare().
+    sim.prepare()
+    return robot
diff --git a/scripts/tutorials/atomic_action/place.py b/scripts/tutorials/atomic_action/place.py
index f943bd2e1..d67f2adb7 100644
--- a/scripts/tutorials/atomic_action/place.py
+++ b/scripts/tutorials/atomic_action/place.py
@@ -60,6 +60,10 @@
     replay_trajectory,
     run_tutorial,
 )
+from scripts.tutorials.atomic_action.tutorial_utils import (
+    compute_pick_close_end_step,
+    initialize_benchmark_simulation,
+)
 
 OBJECT_SIZE = (0.05, 0.05, 0.05)
 OBJECT_XY = (-0.42, -0.08)
@@ -226,3 +230,16 @@ def main() -> None:
 
 if __name__ == "__main__":
     run_tutorial(main)
+
+
+def initialize_simulation(args) -> "SimulationManager":
+    """Create the tutorial simulation for interactive or benchmark runs."""
+    return initialize_benchmark_simulation(args)
+
+
+def create_robot(sim: "SimulationManager") -> "Robot":
+    """Add the default Place tutorial robot."""
+    robot = add_tutorial_robot(sim, "ur5", tcp_z=0.15)
+    # DexSim binds body_data (and therefore get_qpos) only after prepare().
+    sim.prepare()
+    return robot
diff --git a/scripts/tutorials/atomic_action/tutorial_utils.py b/scripts/tutorials/atomic_action/tutorial_utils.py
index 1167510b1..5f7307d76 100644
--- a/scripts/tutorials/atomic_action/tutorial_utils.py
+++ b/scripts/tutorials/atomic_action/tutorial_utils.py
@@ -1600,3 +1600,43 @@ def create_tutorial_robot_cfg(
     "stop_auto_play_recording",
     "draw_axis_marker",
 ]
+
+
+def initialize_benchmark_simulation(args) -> "SimulationManager":
+    """Create the tutorial simulation from a benchmark-style namespace.
+
+    Benchmark argument namespaces carry only ``device``/``renderer``; fill the
+    remaining launcher fields with tutorial defaults so
+    :func:`create_tutorial_simulation` accepts them unchanged.
+
+    Args:
+        args: Namespace with optional ``num_envs``/``device``/``renderer``.
+
+    Returns:
+        The shared tutorial simulation.
+    """
+    namespace = argparse.Namespace(
+        num_envs=getattr(args, "num_envs", 1),
+        device=getattr(args, "device", "cpu"),
+        renderer=getattr(args, "renderer", "auto"),
+        headless=True,
+    )
+    return create_tutorial_simulation(namespace)
+
+
+def compute_pick_close_end_step(compiled=None, invocation_index: int = 0) -> int:
+    """Trajectory step where PickUp's hand-close segment ends (lift start).
+
+    Args:
+        compiled: Optional compiled engine result; when given, the exact
+            ``lift`` segment start of the selected invocation is returned.
+        invocation_index: Invocation to inspect within ``compiled``.
+
+    Returns:
+        Step index separating the grasp phase from the lift phase. Without a
+        compiled result this uses the tutorial defaults
+        (``sample_count=120`` + ``hand_interp_steps=12`` + ``settle=0``).
+    """
+    if compiled is not None:
+        return int(compiled.segment(invocation_index, "lift").start)
+    return 120 + 12