Added shared piles option to dirt_quadrant eval + Changed dirt_quadrant layout and adapted configs

marl_factory_grid/algorithms/marl/a2c_dirt.py

@@ -383,6 +383,14 @@ class A2C:
             obs[0][1][x][y] = 1
             print("Missing agent position")
 
+    def get_all_cleaned_dirt_piles(self, dirt_piles_positions, cleaned_dirt_piles):
+        meta_cleaned_dirt_piles = {pos: False for pos in dirt_piles_positions}
+        for agent_idx in range(self.n_agents):
+            for (pos, cleaned) in cleaned_dirt_piles[agent_idx].items():
+                if cleaned:
+                    meta_cleaned_dirt_piles[pos] = True
+        return meta_cleaned_dirt_piles
+
     def handle_dirt(self, env, cleaned_dirt_piles, ordered_dirt_piles, target_pile, indices, reward, done):
         # Check if agent moved on field with dirt. If that is the case collect dirt automatically
         agent_positions = [env.state.moving_entites[agent_idx].pos for agent_idx in range(self.n_agents)]
@@ -427,12 +435,7 @@ class A2C:
                     done = True
             elif self.cfg[nms.ALGORITHM]["pile_all_done"] == "shared":
                 # End episode if both agents together have cleaned all dirt piles
-                meta_cleaned_dirt_piles = {pos: False for pos in dirt_piles_positions}
-                for agent_idx in range(self.n_agents):
-                    for (pos, cleaned) in cleaned_dirt_piles[agent_idx].items():
-                        if cleaned:
-                            meta_cleaned_dirt_piles[pos] = True
-                if all(meta_cleaned_dirt_piles.values()):
+                if all(self.get_all_cleaned_dirt_piles(dirt_piles_positions, cleaned_dirt_piles).values()):
                     done = True
 
         return reward, done

marl_factory_grid/algorithms/marl/configs/dirt_quadrant_config.yaml

@@ -24,7 +24,7 @@ algorithm:
   entropy_coef:       0.01
   vf_coef:            0.05
   n_steps:            0 # How much experience should be sampled at most (n-TD) until the next value and policy update is performed. Default 0: MC
-  max_steps:          270000
+  max_steps:          240000
   advantage:          "Advantage-AC" # Options: "Advantage-AC", "TD-Advantage-AC", "Reinforce"
   pile-order:         "fixed" # Options: "fixed", "random", "none", "agents", "dynamic", "smart" (Use "fixed", "random" and "none" for single agent training and the other for multi agent inference)
   pile-observability: "single" # Options: "single", "all"

marl_factory_grid/configs/custom/MultiAgentConfigs/dirt_quadrant_eval_config.yaml

@@ -37,7 +37,7 @@ Agents:
 
 Entities:
   DirtPiles:
-    coords_or_quantity: (9,9), (4,5), (1,1) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
+    coords_or_quantity: (9,9), (7,9), (4,7), (2,4), (1, 1) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
     initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
     clean_amount: 1
     dirt_spawn_r_var: 0

marl_factory_grid/configs/custom/MultiAgentConfigs/two_rooms_one_door_modified_eval_config_emergent.yaml

@@ -59,4 +59,4 @@ Rules:
   # Done Conditions
   #DoneOnAllDirtCleaned:
   DoneAtMaxStepsReached:
-    max_steps: 50
+    max_steps: 30

marl_factory_grid/configs/custom/dirt_quadrant_eval_config.yaml

@@ -25,8 +25,15 @@ Agents:
       #- Self
     #Positions:
       #- (9,1)
+      #- (1,1)
+      #- (2,4)
+      #- (4,7)
+      #- (7,9)
+      #- (2,4)
+      #- (4,7)
+      #- (7,9)
       #- (9,9)
-      #- (4,5)
+      #- (9,1)
   Wolfgang:
     Actions:
       - Move4
@@ -35,12 +42,19 @@ Agents:
       - Self
     Positions:
       - (9,5)
-      - (9,9)
-      - (4,5)
+      #- (1,1)
+      #- (2,4)
+      #- (4,7)
+      #- (7,9)
+      #- (2,4)
+      #- (4,7)
+      #- (7,9)
+      #- (9,9)
+      #- (9,5)
 
 Entities:
   DirtPiles:
-    coords_or_quantity: (9,9), (4,5), (1,1) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
+    coords_or_quantity: (1, 1), (2,4), (4,7), (7,9), (9,9) #(9,9), (7,9), (4,7), (2,4), (1, 1) #(1, 1), (2,4), (4,7), (7,9), (9,9) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
     initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
     clean_amount: 1
     dirt_spawn_r_var: 0

marl_factory_grid/configs/custom/dirt_quadrant_train_config.yaml

@@ -24,11 +24,17 @@ Agents:
       #- Self
     #Positions:
       #- (9,1)
-      #- (4,5)
       #- (1,1)
-      #- (4,5)
-      #- (9,1)
+      #- (2,4)
+      #- (4,7)
+      #- (6,8)
+      #- (7,9)
+      #- (2,4)
+      #- (4,7)
+      #- (6,8)
+      #- (7,9)
       #- (9,9)
+      #- (9,1)
   Wolfgang:
     Actions:
       - Move4
@@ -37,16 +43,22 @@ Agents:
       - Self
     Positions:
       - (9,5)
-      - (4,5)
       - (1,1)
-      - (4,5)
-      - (9,5)
+      - (2,4)
+      - (4,7)
+      - (6,8)
+      - (7,9)
+      - (2,4)
+      - (4,7)
+      - (6,8)
+      - (7,9)
       - (9,9)
+      - (9,5)
 
 
 Entities:
   DirtPiles:
-    coords_or_quantity: (9,9), (1,1), (4,5)  # (4,7), (2,4), (1, 1) #(1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
+    coords_or_quantity: (1, 1), (2,4), (4,7), (6,8), (7,9), (9,9)  # (4,7), (2,4), (1, 1) #(1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
     initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
     clean_amount: 1
     dirt_spawn_r_var: 0

marl_factory_grid/configs/dirt_quadrant.yaml

@@ -27,7 +27,7 @@ Agents:
       - DirtPiles
       - Self
     Positions:
-      - (9,2)
+      - (9,1)
   Reiner:
     Actions:
       - Move4
@@ -43,7 +43,7 @@ Agents:
 
 Entities:
   DirtPiles:
-    coords_or_quantity: (1, 1), (4,5), (9,9)
+    coords_or_quantity: (1, 1), (2,4), (4,7), (7,9), (9,9)
     initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
     clean_amount: 1
     dirt_spawn_r_var: 0

marl_factory_grid/levels/quadrant.txt

@@ -1,10 +1,10 @@
 ###########
-#---#######
-#-----#####
-#------####
-#-------###
-#--------##
-#--------##
+#---------#
+#---------#
+#---------#
+#---------#
+#---------#
+#---------#
 #---------#
 #---------#
 #---------#