diff --git a/environments/factory/base_factory.py b/environments/factory/base_factory.py
index 25e089c..c7d9863 100644
--- a/environments/factory/base_factory.py
+++ b/environments/factory/base_factory.py
@@ -8,6 +8,10 @@ from environments import helpers as h
 
 class BaseFactory:
 
+    @property
+    def movement_actions(self):
+        return (int(self.allow_vertical_movement) + int(self.allow_horizontal_movement)) * 4
+
     def __init__(self, level='simple', n_agents=1, max_steps=1e3):
         self.n_agents = n_agents
         self.max_steps = max_steps
@@ -64,8 +68,7 @@ class BaseFactory:
         return self.state, r, self.done, info
 
     def _is_moving_action(self, action):
-        movement_actions = (int(self.allow_vertical_movement) + int(self.allow_horizontal_movement)) * 4
-        if action < movement_actions:
+        if action < self.movement_actions:
             return True
         else:
             return False
@@ -106,6 +109,9 @@ class BaseFactory:
             # Agent seems to be trying to collide in this step
             return old_pos, valid
 
+    def agent_i_position(self, agent_i):
+        return np.argwhere(self.state[h.AGENT_START_IDX+agent_i] == h.IS_OCCUPIED_CELL)
+
     @property
     def free_cells(self) -> np.ndarray:
         free_cells = self.state.sum(0)
diff --git a/environments/factory/simple_factory_getting_dirty.py b/environments/factory/simple_factory_getting_dirty.py
index c3a8b01..739a480 100644
--- a/environments/factory/simple_factory_getting_dirty.py
+++ b/environments/factory/simple_factory_getting_dirty.py
@@ -1,6 +1,8 @@
 import numpy as np
 from environments.factory.base_factory import BaseFactory
 from collections import namedtuple
+from typing import Iterable
+from environments import helpers as h
 
 DIRT_INDEX = -1
 DirtProperties = namedtuple('DirtProperties', ['clean_amount', 'max_spawn_ratio', 'gain_amount'],
@@ -9,29 +11,49 @@ DirtProperties = namedtuple('DirtProperties', ['clean_amount', 'max_spawn_ratio'
 
 class GettingDirty(BaseFactory):
 
+    @property
+    def _clean_up_action(self):
+        return self.movement_actions + 1
+
     def __init__(self, *args, dirt_properties:DirtProperties, **kwargs):
         super(GettingDirty, self).__init__(*args, **kwargs)
         self._dirt_properties = dirt_properties
         self.slice_strings.update({self.state.shape[0]-1: 'dirt'})
 
-    def spawn_dirt(self):
+    def spawn_dirt(self) -> None:
         free_for_dirt = self.free_cells
         # randomly distribute dirt across the grid
         n_dirt_tiles = self._dirt_properties.max_spawn_ratio * len(free_for_dirt)
         for x, y in free_for_dirt[:n_dirt_tiles]:
             self.state[DIRT_INDEX, x, y] += self._dirt_properties.gain_amount
 
-    def additional_actions(self, agent_i, action) -> ((int, int), bool):
-        if action ==
+    def clean_up(self, pos: (int, int)) -> ((int, int), bool):
+        new_dirt_amount = self.state[DIRT_INDEX][pos] - self._dirt_properties.clean_amount
+        cleanup_was_sucessfull: bool
+        if self.state[DIRT_INDEX][pos] == h.IS_FREE_CELL:
+            cleanup_was_sucessfull = False
+            return pos, cleanup_was_sucessfull
+        else:
+            cleanup_was_sucessfull = True
+            self.state[DIRT_INDEX][pos] = max(new_dirt_amount, h.IS_FREE_CELL)
+            return pos, cleanup_was_sucessfull
 
-    def reset(self):
+
+    def additional_actions(self, agent_i, action) -> ((int, int), bool):
+        if not action == self._is_moving_action(action):
+            if action == self._clean_up_action:
+                self.clean_up()
+        else:
+            raise RuntimeError('This should not happen!!!')
+
+    def reset(self) -> None:
         # ToDo: When self.reset returns the new states and stuff, use it here!
         super().reset()  # state, agents, ... =
         dirt_slice = np.zeros((1, *self.state.shape[1:]))
         self.state = np.concatenate((self.state, dirt_slice))  # dirt is now the last slice
         self.spawn_dirt()
 
-    def calculate_reward(self, collisions_vecs, actions, r):
+    def calculate_reward(self, collisions_vec: np.ndarray, actions: Iterable[int], r: int) -> (int, dict):
         for agent_i, cols in enumerate(collisions_vecs):
             cols = np.argwhere(cols != 0).flatten()
             print(f't = {self.steps}\tAgent {agent_i} has collisions with '