diff --git a/lerobot/common/teleoperators/gamepad/teleop_gamepad.py b/lerobot/common/teleoperators/gamepad/teleop_gamepad.py
index 0fcd4167..61c2b4b9 100644
--- a/lerobot/common/teleoperators/gamepad/teleop_gamepad.py
+++ b/lerobot/common/teleoperators/gamepad/teleop_gamepad.py
@@ -89,9 +89,9 @@ class GamepadTeleop(Teleoperator):
         # use HidApi for macos
         if sys.platform == "darwin":
             # NOTE: On macOS, pygame doesn’t reliably detect input from some controllers so we fall back to hidapi
-            from lerobot.scripts.server.end_effector_control_utils import GamepadControllerHID as Gamepad
+            from lerobot.common.utils.end_effector_control import GamepadControllerHID as Gamepad
         else:
-            from lerobot.scripts.server.end_effector_control_utils import GamepadController as Gamepad
+            from lerobot.common.utils.end_effector_control import GamepadController as Gamepad
 
         self.gamepad = Gamepad(x_step_size=1.0, y_step_size=1.0, z_step_size=1.0)
         self.gamepad.start()
diff --git a/lerobot/common/utils/end_effector_control.py b/lerobot/common/utils/end_effector_control.py
index 74003809..3bb9385a 100644
--- a/lerobot/common/utils/end_effector_control.py
+++ b/lerobot/common/utils/end_effector_control.py
@@ -24,7 +24,7 @@ import torch
 
 from lerobot.common.utils.robot_utils import busy_wait
 from lerobot.common.utils.utils import init_logging
-from lerobot.scripts.server.kinematics import RobotKinematics
+from lerobot.common.model.kinematics import RobotKinematics
 
 
 class InputController:
diff --git a/lerobot/scripts/rl/gym_manipulator.py b/lerobot/scripts/rl/gym_manipulator.py
index b7b9780b..d41a18fd 100644
--- a/lerobot/scripts/rl/gym_manipulator.py
+++ b/lerobot/scripts/rl/gym_manipulator.py
@@ -2225,7 +2225,7 @@ def main(cfg: EnvConfig):
     while num_episode < 10:
         start_loop_s = time.perf_counter()
         # Sample a new random action from the robot's action space.
-        new_random_action = env.action_space.sample()
+        new_random_action = env.action_space.sample() * 0
         # Update the smoothed action using an exponential moving average.
         smoothed_action = alpha * new_random_action + (1 - alpha) * smoothed_action