diff --git a/isaac_sim/policies/m20.py b/isaac_sim/policies/m20.py
index c8f4a98..8ee7964 100644
--- a/isaac_sim/policies/m20.py
+++ b/isaac_sim/policies/m20.py
@@ -48,6 +48,8 @@
 )
 WHEEL_ACTION_SCALE = 5.0  # env.yaml joint_vel action scale
 
+ZERO_CMD_EPS = 1e-3
+
 
 class M20VelocityPolicy(PolicyController):
     """DeepRobotics M20 wheeled quadruped running a velocity-tracking policy."""
@@ -205,6 +207,9 @@ def forward(self, dt: float, command: np.ndarray) -> None:
         vel_cmd = np.zeros(len(pos_cmd), dtype=np.float32)
         vel_cmd[self._wheel_ids] = self._wheel_vel_target
 
+        if float(np.abs(command).max()) < ZERO_CMD_EPS:
+            vel_cmd[self._wheel_ids] = 0.0
+
         self.robot.apply_action(
             ArticulationAction(joint_positions=pos_cmd, joint_velocities=vel_cmd)
         )