[Update atari_preprocessing.py]: Remove FireReset (#1661)

* Update atari_preprocessing.py

* Update test_atari_preprocessing.py

* Update test_atari_preprocessing.py

gym/wrappers/atari_preprocessing.py

@@ -15,7 +15,6 @@ class AtariPreprocessing(gym.Wrapper):
     Specifically:
 
     * NoopReset: obtain initial state by taking random number of no-ops on reset. 
-    * FireReset: take action on reset for environments that are fixed until firing. 
     * Frame skipping: 4 by default
     * Max-pooling: most recent two observations
     * Termination signal when a life is lost: turned off by default. Not recommended by Machado et al. (2018).
@@ -105,12 +104,6 @@ def reset(self, **kwargs):
             if done:
                 self.env.reset(**kwargs)
 
-        # FireReset
-        action_meanings = self.env.unwrapped.get_action_meanings()
-        if action_meanings[1] == 'FIRE' and len(action_meanings) >= 3:
-            self.env.step(1)
-            self.env.step(2)
-
         self.lives = self.ale.lives()
         if self.grayscale_obs:
             self.ale.getScreenGrayscale(self.obs_buffer[0])

gym/wrappers/test_atari_preprocessing.py

@@ -16,19 +16,21 @@ def test_atari_preprocessing_grayscale(env_fn):
     env1 = env_fn()
     env2 = AtariPreprocessing(env_fn(), screen_size=84, grayscale_obs=True, frame_skip=1, noop_max=0)
     env3 = AtariPreprocessing(env_fn(), screen_size=84, grayscale_obs=False, frame_skip=1, noop_max=0)
-    env1.reset()
-    # take these steps to imitate actions of FireReset logic
-    env1.step(1)
-    obs1 = env1.step(2)[0]
+    env1.seed(0)
+    env2.seed(0)
+    env3.seed(0)
+    obs1 = env1.reset()
     obs2 = env2.reset()
     obs3 = env3.reset()
     assert obs1.shape == (210, 160, 3)
     assert obs2.shape == (84, 84)
     assert obs3.shape == (84, 84, 3)
-    np.testing.assert_allclose(obs3, cv2.resize(obs1, (84, 84), interpolation=cv2.INTER_AREA))
+    assert np.allclose(obs3, cv2.resize(obs1, (84, 84), interpolation=cv2.INTER_AREA))
     obs3_gray = cv2.cvtColor(obs3, cv2.COLOR_RGB2GRAY)
     # the edges of the numbers do not render quite the same in the grayscale, so we ignore them
-    np.testing.assert_allclose(obs2[10:], obs3_gray[10:])
+    assert np.allclose(obs2[10:38], obs3_gray[10:38])
+    # the paddle also do not render quite the same
+    assert np.allclose(obs2[44:], obs3_gray[44:])
 
     env1.close()
     env2.close()
@@ -53,4 +55,4 @@ def test_atari_preprocessing_scale(env_fn):
                 assert (0 <= obs).all() and (obs <= max_obs).all(), 'Obs. must be in range [0,{}]'.format(max_obs)
                 step_i += 1
 
-            env.close()
+            env.close()