Merge pull request #59 from Farama-Foundation/doc/improve-reward-bounds

Improve reward bounds on some environments

mo_gymnasium/envs/lunar_lander/lunar_lander.py

@@ -35,7 +35,12 @@ def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
 
         # Result reward, shaping reward, main engine cost, side engine cost
-        self.reward_space = spaces.Box(low=-np.inf, high=np.inf, shape=(4,), dtype=np.float32)
+        self.reward_space = spaces.Box(
+            low=np.array([-100, -np.inf, -1, -1]),
+            high=np.array([100, np.inf, 0, 0]),
+            shape=(4,),
+            dtype=np.float32,
+        )
         self.reward_dim = 4
 
     def step(self, action):

mo_gymnasium/envs/mario/mario.py

@@ -51,7 +51,11 @@ def __init__(
         self.render_mode = render_mode
 
         self.objectives = set(objectives)
-        self.reward_space = gym.spaces.Box(high=np.inf, low=-np.inf, shape=(len(objectives),))
+        self.reward_space = gym.spaces.Box(
+            low=np.array([-np.inf, -np.inf, -25, 0, 0]),
+            high=np.array([np.inf, 0, 0, 100, np.inf]),
+            shape=(len(objectives),),
+        )
         self.reward_dim = len(objectives)
 
         # observation space for the environment is static across all instances

mo_gymnasium/envs/minecart/minecart.py

@@ -192,7 +192,11 @@ def __init__(
             self.observation_space = Box(-np.ones(7), np.ones(7), dtype=np.float32)
 
         self.action_space = Discrete(6)
-        self.reward_space = Box(low=-1, high=self.capacity, shape=(self.ore_cnt + 1,))
+        self.reward_space = Box(
+            low=np.append(np.zeros(self.ore_cnt), -1.0),
+            high=np.append(np.ones(self.ore_cnt) * self.capacity, 0.0),
+            shape=(self.ore_cnt + 1,),
+        )
         self.reward_dim = self.ore_cnt + 1
 
     def convex_coverage_set(self, gamma: float, symmetric: bool = True) -> List[np.ndarray]:

mo_gymnasium/envs/mountain_car/mountain_car.py

@@ -24,7 +24,7 @@ def __init__(self, render_mode: Optional[str] = None, goal_velocity=0):
         super().__init__(render_mode, goal_velocity)
         EzPickle.__init__(self, render_mode, goal_velocity)
 
-        self.reward_space = spaces.Box(low=-1, high=1, shape=(3,), dtype=np.float32)
+        self.reward_space = spaces.Box(low=np.array([-1, -1, -1]), high=np.array([-1, 0, 0]), shape=(3,), dtype=np.float32)
         self.reward_dim = 3
 
     def step(self, action: int):

mo_gymnasium/envs/resource_gathering/resource_gathering.py

@@ -76,7 +76,7 @@ def __init__(self, render_mode: Optional[str] = None):
         # action space specification: 1 dimension, 0 up, 1 down, 2 left, 3 right
         self.action_space = Discrete(4)
         # reward space:
-        self.reward_space = Box(low=-1, high=1, shape=(3,), dtype=np.float32)
+        self.reward_space = Box(low=np.array([0.0, 0.0, -1.0]), high=np.array([1.0, 1.0, 0]), shape=(3,), dtype=np.float32)
         self.reward_dim = 3
 
         # pygame