Merge pull request #488 from LLNL/abmarl-456-sim-unwrapped-in-stage

External wrappers have unwrapped property

abmarl/external/gym_env_wrapper.py

@@ -34,6 +34,16 @@ def observation_space(self):
         """
         return self.agent.observation_space
 
+    @property
+    def unwrapped(self):
+        """
+        Fall through all the wrappers and obtain the original, completely unwrapped simulation.
+        """
+        try:
+            return self.sim.unwrapped
+        except AttributeError:
+            return self.sim
+
     def reset(self, **kwargs):
         """
         Return the observation from the single agent.
@@ -58,13 +68,3 @@ def render(self, **kwargs):
         Forward render calls to the composed simulation.
         """
         self.sim.render(**kwargs)
-
-    @property
-    def unwrapped(self):
-        """
-        Fall through all the wrappers and obtain the original, completely unwrapped simulation.
-        """
-        try:
-            return self.sim.unwrapped
-        except AttributeError:
-            return self.sim

abmarl/external/open_spiel_env_wrapper.py

@@ -121,6 +121,16 @@ def current_player(self, value):
                 "Current player must be an agent in the simulation."
             self._current_player = value
 
+        @property
+        def unwrapped(self):
+            """
+            Fall through all the wrappers and obtain the original, completely unwrapped simulation.
+            """
+            try:
+                return self.sim.unwrapped
+            except AttributeError:
+                return self.sim
+
         def reset(self, **kwargs):
             """
             Reset the simulation.

abmarl/external/rllib_multiagentenv_wrapper.py

@@ -38,6 +38,16 @@ def __init__(self, sim):
             })
             self._spaces_in_preferred_format = True
 
+        @property
+        def unwrapped(self):
+            """
+            Fall through all the wrappers and obtain the original, completely unwrapped simulation.
+            """
+            try:
+                return self.sim.unwrapped
+            except AttributeError:
+                return self.sim
+
         def reset(self):
             """See SimulationManager."""
             return self.sim.reset()

tests/test_workflow.py

@@ -121,6 +121,8 @@ def tally_rewards(sim, trainer):
         trainer:
             Trainer that computes actions using the trained policies.
     """
+    from abmarl.managers import SimulationManager
+    assert isinstance(sim, SimulationManager), "sim must be a SimulationManager."
     # Run the simulation with actions chosen from the trained policies
     policy_agent_mapping = trainer.config['multiagent']['policy_mapping_fn']
     for episode in range(5):