ddqn_agent.py 文件源码

def agent_start(self, observation):

        # Preprocess
        tmp = np.bitwise_and(np.asarray(observation.intArray[128:]).reshape([210, 160]), 0b0001111)  # Get Intensity from the observation
        obs_array = (spm.imresize(tmp, (110, 84)))[110-84-8:110-8, :]  # Scaling

        # Initialize State
        self.state = np.zeros((4, 84, 84), dtype=np.uint8)
        self.state[0] = obs_array
        state_ = cuda.to_gpu(np.asanyarray(self.state.reshape(1, 4, 84, 84), dtype=np.float32))

        # Generate an Action e-greedy
        returnAction = Action()
        action, Q_now = self.DDQN.e_greedy(state_, self.epsilon)
        returnAction.intArray = [action]

        # Update for next step
        self.lastAction = copy.deepcopy(returnAction)
        self.last_state = self.state.copy()
        self.last_observation = obs_array

        return returnAction