reward ajustments

robot starts balancing without changing phase and farms alive bonus
fix ->external force and bigger height penalty

ml/SimManager.py

@@ -163,4 +163,30 @@ class SimManager:
             self.step()
         heights = self.measure_robot_heights(robot_ids)
         mean_height = float(np.mean(heights)) if heights else fallback_height
-        return mean_height if mean_height > 0.0 else fallback_height
+        return mean_height if mean_height > 0.0 else fallback_height
+
+    def apply_external_force(
+        self,
+        body_id: int,
+        force: list[float] | np.ndarray,
+        link_index: int = -1,
+        position: list[float] | np.ndarray = (0.0, 0.0, 0.0),
+        frame: int = p.WORLD_FRAME,
+    ):
+        """
+        Applies a 3D force vector (in Newtons) to a robot link.
+        
+        :param body_id: PyBullet body ID.
+        :param force: [fx, fy, fz] force vector in Newtons.
+        :param link_index: Target link index (-1 refers to the base/torso).
+        :param position: Offset [x, y, z] relative to link center where force is applied.
+        :param frame: p.WORLD_FRAME (global axes) or p.LINK_FRAME (robot's body axes).
+        """
+        p.applyExternalForce(
+            objectUniqueId=body_id,
+            linkIndex=link_index,
+            forceObj=list(force),
+            posObj=list(position),
+            flags=frame,
+            physicsClientId=self.physics_client,
+        )