Fixed the reward hacking bug

See oxwhirl#72 for more details.
benellis3 · Apr 19, 2022 · 8ab0b54 · 8ab0b54
1 parent 2ff475a
commit 8ab0b54
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/smac/env/starcraft2/starcraft2.py b/smac/env/starcraft2/starcraft2.py
@@ -979,7 +979,7 @@ def reward_battle(self):
                     delta_enemy += prev_health - e_unit.health - e_unit.shield
 
         if self.reward_only_positive:
-            reward = abs(delta_enemy + delta_deaths)  # shield regeneration
+            reward = max(delta_enemy + delta_deaths, 0)  # shield regeneration
         else:
             reward = delta_enemy + delta_deaths - delta_ally