Python DoomGame.is_episode_finished方法代码示例

# 需要导入模块: from doom_py import DoomGame [as 别名]
# 或者: from doom_py.DoomGame import is_episode_finished [as 别名]
class DoomEnv(gym.Env, utils.EzPickle):
    metadata = {'render.modes': ['human', 'rgb_array'], 'video.frames_per_second': 35}

    def __init__(self, level):
        utils.EzPickle.__init__(self)
        self.previous_level = -1
        self.level = level
        self.game = DoomGame()
        self.loader = Loader()
        self.doom_dir = os.path.dirname(os.path.abspath(__file__))
        self.mode = 'fast'                          # 'human', 'fast' or 'normal'
        self.no_render = False                      # To disable double rendering in human mode
        self.viewer = None
        self.is_initialized = False                 # Indicates that reset() has been called
        self.curr_seed = 0
        self.action_space = spaces.HighLow(
            np.matrix([[0, 1, 0]] * 38 + [[-10, 10, 0]] * 2 + [[-100, 100, 0]] * 3, dtype=np.int8))
        self.allowed_actions = list(range(NUM_ACTIONS))
        self._seed()
        self._configure()

    def _configure(self, screen_resolution=ScreenResolution.RES_640X480):
        # Often agents end up downsampling the observations. Configuring Doom to
        # return a smaller image yields significant (~10x) speedups
        if screen_resolution == ScreenResolution.RES_640X480:
            self.screen_height = 480
            self.screen_width = 640
            self.screen_resolution = ScreenResolution.RES_640X480
        elif screen_resolution == ScreenResolution.RES_160X120:
            self.screen_height = 120
            self.screen_width = 160
            self.screen_resolution = ScreenResolution.RES_160X120

        self.observation_space = spaces.Box(low=0, high=255, shape=(self.screen_height, self.screen_width, 3))

    def _load_level(self):
        # Closing if is_initialized
        if self.is_initialized:
            self.is_initialized = False
            self.game.close()
            self.game = DoomGame()

        # Loading Paths
        if not self.is_initialized:
            self.game.set_vizdoom_path(self.loader.get_vizdoom_path())
            self.game.set_doom_game_path(self.loader.get_freedoom_path())

        # Common settings
        self._closed = False
        self.game.load_config(os.path.join(self.doom_dir, 'assets/%s' % DOOM_SETTINGS[self.level][CONFIG]))
        self.game.set_doom_scenario_path(self.loader.get_scenario_path(DOOM_SETTINGS[self.level][SCENARIO]))
        if DOOM_SETTINGS[self.level][MAP] != '':
            self.game.set_doom_map(DOOM_SETTINGS[self.level][MAP])
        self.game.set_doom_skill(DOOM_SETTINGS[self.level][DIFFICULTY])
        self.previous_level = self.level
        self.allowed_actions = DOOM_SETTINGS[self.level][ACTIONS]
        self.game.set_screen_resolution(self.screen_resolution)

        # Algo mode
        if 'human' != self.mode:
            self.game.set_window_visible(False)
            self.game.set_mode(Mode.PLAYER)
            self.no_render = False
            self.game.init()
            self._start_episode()
            self.is_initialized = True
            return self.game.get_state().image_buffer.copy()

        # Human mode
        else:
            self.game.add_game_args('+freelook 1')
            self.game.set_window_visible(True)
            self.game.set_mode(Mode.SPECTATOR)
            self.no_render = True
            self.game.init()
            self._start_episode()
            self.is_initialized = True
            self._play_human_mode()
            return np.zeros(shape=self.observation_space.shape, dtype=np.uint8)

    def _start_episode(self):
        if self.curr_seed > 0:
            self.game.set_seed(self.curr_seed)
            self.curr_seed = 0
        self.game.new_episode()
        return

    def _play_human_mode(self):
        while not self.game.is_episode_finished():
            self.game.advance_action()
            state = self.game.get_state()
            total_reward = self.game.get_total_reward()
            info = self._get_game_variables(state.game_variables)
            info["TOTAL_REWARD"] = round(total_reward, 4)
            print('===============================')
            print('State: #' + str(state.number))
            print('Action: \t' + str(self.game.get_last_action()) + '\t (=> only allowed actions)')
            print('Reward: \t' + str(self.game.get_last_reward()))
            print('Total Reward: \t' + str(total_reward))
            print('Variables: \n' + str(info))
#.........这里部分代码省略.........

# 需要导入模块: from doom_py import DoomGame [as 别名]
# 或者: from doom_py.DoomGame import is_episode_finished [as 别名]

#.........这里部分代码省略.........
                    self.game.init()
            except (ViZDoomUnexpectedExitException, ViZDoomErrorException):
                raise error.Error(
                    'VizDoom exited unexpectedly. This is likely caused by a missing multiprocessing lock. ' +
                    'To run VizDoom across multiple processes, you need to pass a lock when you configure the env ' +
                    '[e.g. env.configure(lock=my_multiprocessing_lock)], or create and close an env ' +
                    'before starting your processes [e.g. env = gym.make("DoomBasic-v0"); env.close()] to cache a ' +
                    'singleton lock in memory.')
            self._start_episode()
            self.is_initialized = True
            return self.game.get_state().image_buffer.copy()

        # Human mode
        else:
            self.game.add_game_args('+freelook 1')
            self.game.set_window_visible(True)
            self.game.set_mode(Mode.SPECTATOR)
            self.no_render = True
            with self.lock:
                self.game.init()
            self._start_episode()
            self.is_initialized = True
            self._play_human_mode()
            return np.zeros(shape=self.observation_space.shape, dtype=np.uint8)

    def _start_episode(self):
        if self.curr_seed > 0:
            self.game.set_seed(self.curr_seed)
            self.curr_seed = 0
        self.game.new_episode()
        return

    def _play_human_mode(self):
        while not self.game.is_episode_finished():
            self.game.advance_action()
            state = self.game.get_state()
            total_reward = self.game.get_total_reward()
            info = self._get_game_variables(state.game_variables)
            info["TOTAL_REWARD"] = round(total_reward, 4)
            print('===============================')
            print('State: #' + str(state.number))
            print('Action: \t' + str(self.game.get_last_action()) + '\t (=> only allowed actions)')
            print('Reward: \t' + str(self.game.get_last_reward()))
            print('Total Reward: \t' + str(total_reward))
            print('Variables: \n' + str(info))
            sleep(0.02857)  # 35 fps = 0.02857 sleep between frames
        print('===============================')
        print('Done')
        return

    def _step(self, action):
        if NUM_ACTIONS != len(action):
            logger.warn('Doom action list must contain %d items. Padding missing items with 0' % NUM_ACTIONS)
            old_action = action
            action = [0] * NUM_ACTIONS
            for i in range(len(old_action)):
                action[i] = old_action[i]
        # action is a list of numbers but DoomGame.make_action expects a list of ints
        if len(self.allowed_actions) > 0:
            list_action = [int(action[action_idx]) for action_idx in self.allowed_actions]
        else:
            list_action = [int(x) for x in action]
        try:
            reward = self.game.make_action(list_action)
            state = self.game.get_state()
            info = self._get_game_variables(state.game_variables)