sess.run(init)
tf.global_variables_initializer().run()
if load_model == True:
print('Loading Model...')
ckpt = tf.train.get_checkpoint_state(path)
saver.restore(sess, ckpt.model_checkpoint_path)
updateTarget(targetOps,
sess) #Set the target network to be equal to the primary network.
tf.summary.FileWriter("logs/", self.sess.graph)
#所有episode的运行过程
for i in range(1, num_episodes):
episodeBuffer0 = priorized_experience_buffer() #首先明确好经验池的大小问题
environment = sumo.reset() #重置环境
current_state = sumo.getState() #拿到环境的状态信息
current_phases = list(init_phases) #将初始相位设置成list
wait_time_map = {}
wait_time = 0 #我添加的，因为wait_time没有定义
done = False
rAll = 0 #奖励值进行不断的累加
j = 0 #一个episode下步数累加器
flag = False
r_plot_i = []
q_plot_i = []
#这三个参数可以根据实际的来调整,先训练一会,然后累计前相位的相关参数
pre_queue = 12
pre_delay = 1000
pre_throughput = 20
print('step:' + str(i) + ',precent:%.3f' % (i / num_episodes * 100) + '%')
location=[0, 50])
car2 = car_env.car(id=1,
role='follower',
ingaged_in_platoon=False,
tar_interDis=car_env.DES_PLATOON_INTER_DISTANCE,
tar_speed=60.0 / 3.6,
location=[0, 25])
# 将新车加入车队
if len(Carlist) == 0:
Carlist.append(car1)
Carlist.append(car2)
# 设置参与车队的车辆，根据build_platoon，更新是否加入platoon的标志位
car_env.CarList_update_platoon_info(Carlist,
des_platoon_size=2,
build_platoon=True)
observation = car_env.reset(Carlist)
ep_r = 0
# 开始每个episode的运算
while True:
# 时间戳更新
time_tag += car_env.AI_DT
action = RL.choose_action(observation)
# 更新运动学参数。由于c++程序的3D和CarAI的时钟不同步，需要模仿那个程序进行多轮次更新
# 实际上多轮次更新得到的数据更贴近连续值
observation_, done, info = car_env.step_next(
Carlist, time_tag, action)
# 计算单步奖励