[#144] Fix issue with failure penalty in multiagent env

Author: Mark2000

docs/source/release_notes.rst

@@ -33,6 +33,8 @@ Development - |version|
 * Add a maximum range checking dynamics model in :class:`MaxRangeDynModel`. Useful for keeping an agent
   in the vicinity of a target early in training.
 * Add properties in spacecraft dynamics for orbital element observations.
+* Fix an issue with failure penalties in the PettingZoo environment when the rewarder
+  does not return a reward for a satellite.
 
 
 Version 1.1.0

src/bsk_rl/gym.py

@@ -671,7 +671,10 @@ def _get_reward(self) -> dict[AgentID, float]:
         reward = deepcopy(self.reward_dict)
         for agent, satellite in zip(self.possible_agents, self.satellites):
             if not satellite.is_alive():
-                reward[agent] += self.failure_penalty
+                if agent in reward:
+                    reward[agent] += self.failure_penalty
+                else:
+                    reward[agent] = self.failure_penalty
 
         reward_keys = list(reward.keys())
         for agent in reward_keys:

tests/unittest/test_gym_env.py

@@ -538,6 +538,30 @@ def test_get_reward(self):
             sat.name: -10.0 for i, sat in enumerate(env.unwrapped.satellites)
         }
 
+    @patch(
+        "bsk_rl.GeneralSatelliteTasking._get_truncated",
+        MagicMock(return_value=False),
+    )
+    def test_get_reward_missing_sat(self):
+        env = ConstellationTasking(
+            satellites=[
+                MagicMock(is_alive=MagicMock(return_value=False)) for i in range(3)
+            ],
+            world_type=MagicMock(),
+            scenario=MagicMock(),
+            rewarder=MagicMock(),
+            failure_penalty=-20.0,
+        )
+        env._agents_last_compute_time = None
+        env.simulator = MagicMock(sim_time=0.0)
+        env.newly_dead = [sat.name for sat in env.unwrapped.satellites]
+        env.reward_dict = {env.unwrapped.satellites[0].name: 10.0}
+        assert env._get_reward() == {
+            env.unwrapped.satellites[0].name: -10.0,
+            env.unwrapped.satellites[1].name: -20.0,
+            env.unwrapped.satellites[2].name: -20.0,
+        }
+
     @pytest.mark.parametrize("timeout", [False, True])
     @pytest.mark.parametrize("terminate_on_time_limit", [False, True])
     def test_get_terminated(self, timeout, terminate_on_time_limit):