dqn_stage.py 文件源码

def __init__(self, max_random_start,
               observation_dims, data_format, display, use_cumulated_reward):

    self.max_random_start = max_random_start
    self.action_size = 28

    self.use_cumulated_reward = use_cumulated_reward
    self.display = display
    self.data_format = data_format
    self.observation_dims = observation_dims

    self.screen = np.zeros((self.observation_dims[0],self.observation_dims[1]), np.uint8)

    #self.dirToTarget = 0
    self.robotPose = Pose()
    self.goalPose = Pose()
    self.robotRot = 0.0
    self.prevDist = 0.0
    self.boom = False
    self.numWins = 0
    self.ep_reward = 0.0

    self.terminal = 0
    self.sendTerminal = 0

    self.readyForNewData = True

    rospy.wait_for_service('reset_positions')
    self.resetStage = rospy.ServiceProxy('reset_positions', EmptySrv)


    # Subscribers:
   # rospy.Subscriber('base_scan', LaserScan, self.scanCB,queue_size=1)
   # rospy.Subscriber('base_pose_ground_truth', Odometry, self.poseUpdateCB, queue_size=1)

    # trying time sync:
    sub_scan_ = message_filters.Subscriber('base_scan', LaserScan)
    sub_pose_ = message_filters.Subscriber('base_pose_ground_truth', Odometry)
    ts = message_filters.TimeSynchronizer( [sub_scan_, sub_pose_], 1)
    ts.registerCallback( self.syncedCB)

    # publishers:
    self.pub_vel_ = rospy.Publisher('cmd_vel', Twist, queue_size = 1)
    self.pub_rew_ = rospy.Publisher("lastreward",Float64,queue_size=10)
    self.pub_goal_ = rospy.Publisher("goal", Pose, queue_size = 15)