Remove double call to reset_internal_state() on gym environments (#364)

rl_coach/environments/gym_environment.py

@@ -392,9 +392,6 @@ class GymEnvironment(Environment):
             else:
                 screen.error("Error: Environment {} does not support human control.".format(self.env), crash=True)
 
-        # initialize the state by getting a new state from the environment
-        self.reset_internal_state(True)
-
         # render
         if self.is_rendered:
             image = self.get_rendered_image()
@@ -405,7 +402,6 @@ class GymEnvironment(Environment):
                 self.renderer.create_screen(image.shape[1]*scale, image.shape[0]*scale)
 
         # the info is only updated after the first step
-        self.state = self.step(self.action_space.default_action).next_state
         self.state_space['measurements'] = VectorObservationSpace(shape=len(self.info.keys()))
 
         if self.env.spec and custom_reward_threshold is None:

rl_coach/tests/environments/test_gym_environment.py

@@ -16,6 +16,7 @@ def atari_env():
                          seed=1,
                          frame_skip=4,
                          visualization_parameters=VisualizationParameters())
+    env.reset_internal_state(True)
     return env
 
 
@@ -26,6 +27,7 @@ def continuous_env():
                          seed=1,
                          frame_skip=1,
                          visualization_parameters=VisualizationParameters())
+    env.reset_internal_state(True)
     return env
 
 
@@ -56,7 +58,7 @@ def test_gym_continuous_environment(continuous_env):
     assert np.all(continuous_env.action_space.shape == np.array([1]))
 
     # make sure that the seed is working properly
-    assert np.sum(continuous_env.last_env_response.next_state['observation']) == 1.2661630859028832
+    assert np.sum(continuous_env.last_env_response.next_state['observation']) == 0.6118565010687202
 
 
 @pytest.mark.unit_test