Check Action Type

docs/misc/changelog.rst

@@ -85,6 +85,7 @@ Others:
 - Removed redundant return value from ``a2c.utils::total_episode_reward_logger``. (@shwang)
 - Cleanup and refactoring in ``common/identity_env.py`` (@shwang)
 - Added a Makefile to simplify common development tasks (build the doc, type check, run the tests)
+- Action Type Check: Assertion added to ``VecEnv` to check if action is of type list or np.ndarray. Otherwise a developer friendly message is displayed on how to fix the issue. (@mentalgear)
 
 
 Documentation:

stable_baselines/common/vec_env/base_vec_env.py

@@ -146,6 +146,8 @@ def step(self, actions):
         :param actions: ([int] or [float]) the action
         :return: ([int] or [float], [float], [bool], dict) observation, reward, done, information
         """
+        if not isinstance( actions,( list, np.ndarray ) ):
+            raise TypeError( "Action must be of type list or np.ndarray. Try wrapping your action variable in a list [ ] to fix this issue." )
         self.step_async(actions)
         return self.step_wait()
 

tests/test_envs.py

@@ -127,6 +127,22 @@ def wrong_step(_action):
         check_env(env)
 
 
+def test_action_format ( env, action ):
+    """
+    Test if action format check works
+    :param env: (gym.Env)
+    :param action: (list, array)
+    """
+
+    env = gym.make('CartPole-v0')
+
+    with pytest.raises(TypeError):
+        env.step( action )
+
+    with pytest.raises(not TypeError):
+        env.step( [ action ] )
+
+
 def test_common_failures_step():
     """
     Test that common failure cases of the `step` method are caught
@@ -147,3 +163,6 @@ def test_common_failures_step():
     # Done is not a boolean
     check_step_assert_error(env, (env.observation_space.sample(), 0.0, 3.0, {}))
     check_step_assert_error(env, (env.observation_space.sample(), 0.0, 1, {}))
+
+    # action format must be [] or np.ndarray
+    test_action_format ( env, 0 )