dqn_node.py 文件源码

def step(self, action, is_training=False):
    self.prevPoseX = self.poseX
    self.prevPoseY = self.poseY

    if action == -1:
      # Step with random action
      action = int(random.random()*(self.action_size))

    msg = Int8()
    msg.data = action
    self.pub_action_.publish( msg)

    if self.display:
      cv2.imshow("Screen", self.screen)
    #cv2.waitKey(9)

    dist = (self.poseX - self.goalX)**2 + (self.poseY - self.goalY)**2
    reward = (self.prevDist - dist)/10.0
    self.prevDist = dist

    if self.terminal == 1:
      reward -= 900
      #self.new_random_game()

    if dist < 0.9:
      reward += 300
      newStateMSG = EmptyMsg()
      self.pub_new_goal_.publish( newStateMSG)
      # cv2.waitKey(30)

    # Add whatever info you want
    info = ""

    #rospy.loginfo("Episede ended, reward: %g", reward)
    while(self.clock == self.lastClock):
      pass
    self.lastClock = self.clock

    if self.terminal == 2:
      self.sendTerminal = 1

    if self.terminal == 1:
#      rewd = Float32()
#      rewd.data = reward
#      self.pub_rew_.publish( rewd)
      self.terminal = 2

    return self.screen, reward, self.sendTerminal, info

    #observation, reward, terminal, info = self.env.step(action)
    #return self.preprocess(observation), reward, terminal, info