agent1 = agent1_module.Agent()
agent2 = agent2_module.Agent()
agent1_obs_ind = agent1.get_obs_ind()
agent2_obs_ind = agent2.get_obs_ind()
# environment initiation
if args.log:
if args.log_path == 'default_log':
log_flag = args.agent1 + '_vs_' + args.agent2
else:
log_flag = args.log_path
else:
log_flag = False
env = Environment(map_path,
agent1_obs_ind,
agent2_obs_ind,
max_step=args.max_step,
render=True,
random_pos=args.random_pos,
log=log_flag)
# get map info
size_x, size_y = env.get_map_size()
side1_detector_num, side1_fighter_num, side2_detector_num, side2_fighter_num = env.get_unit_num(
)
agent1.set_map_info(size_x, size_y, side1_detector_num, side1_fighter_num)
agent2.set_map_info(size_x, size_y, side2_detector_num, side2_fighter_num)
# execution
step_cnt = 0
round_cnt = 0
agent1_crash_list = []
agent2_crash_list = []
def run(agent1_name,
agent2_name,
map_name,
round_num,
max_step,
random_pos=False):
"""
:param agent1_name: 红方名称
:param agent2_name: 蓝方名称
:param map_name: 地图名称
:param round_num: 对战局数
:param max_step: 单局最大step
:param random_pos: 随机起始位置
:return: agent1_win_times, agent2_win_times, draw_times, agent1_crash_times, agent2_crash_times, agent1_timeout_times, agent2_timeout_times, agent1_launch_failure_times, agent2_launch_failure_times
"""
side1_win_times = 0
side2_win_times = 0
draw_times = 0
log_flag = agent1_name + '_vs_' + agent2_name
agent1_launch_failed = False
agent2_launch_failed = False
round_cnt = 0
agent1_crash_list = []
agent2_crash_list = []
agent1_timeout_list = []
agent2_timeout_list = []
# file path constructing
map_path = 'maps/' + map_name + '.map'
agent1_path = 'agent/' + agent1_name + '/agent.py'
agent2_path = 'agent/' + agent2_name + '/agent.py'
if not os.path.exists(map_path):
print('Error: map file not exist!')
exit(-1)
if not os.path.exists(agent1_path):
print('Error: agent1 file not exist!')
exit(-1)
if not os.path.exists(agent2_path):
print('Error: agent2 file not exist!')
exit(-1)
# make env
env = Environment(map_path,
'raw',
'raw',
max_step=max_step,
render=True,
random_pos=random_pos,
log=log_flag)
# get map info
size_x, size_y = env.get_map_size()
side1_detector_num, side1_fighter_num, side2_detector_num, side2_fighter_num = env.get_unit_num(
)
# create agent
agent1 = AgentCtrl(agent1_name, size_x, size_y, side1_detector_num,
side1_fighter_num)
agent2 = AgentCtrl(agent2_name, size_x, size_y, side2_detector_num,
side2_fighter_num)
if not agent1.agent_init():
print('ERROR: Agent1 ' + agent1_name + ' init failed!')
agent1.terminate()
agent2.terminate()
agent1_launch_failed = True
if not agent2.agent_init():
print('ERROR: Agent2 ' + agent2_name + ' init failed!')
agent1.terminate()
agent2.terminate()
agent2_launch_failed = True
# 若此处一方启动失败，则认为该方全败，启动失败计round_num次，若双方启动失败，则认为双方平局round_num次，其他与前述相同。
if agent1_launch_failed and agent2_launch_failed:
return 0, 0, round_num, 0, 0, 0, 0, round_num, round_num
elif agent1_launch_failed:
return 0, round_num, 0, 0, 0, 0, 0, round_num, 0
elif agent2_launch_failed:
return round_num, 0, 0, 0, 0, 0, 0, 0, round_num
# execution
# input("Press the <ENTER> key to continue...")
for x in range(round_num):
if x != 0:
env.reset()
step_cnt = 0
round_cnt += 1
while True:
step_cnt += 1
# get obs
side1_obs_dict, side2_obs_dict = env.get_obs()
# get action
agent1_action, agent1_result = agent1.get_action(
side1_obs_dict, step_cnt)
if agent1_result == 0:
side1_detector_action = agent1_action['detector_action']
side1_fighter_action = agent1_action['fighter_action']
elif agent1_result == 1:
agent1_crash_list.append(round_cnt)
elif agent1_result == 2:
agent1_timeout_list.append(round_cnt)
agent2_action, agent2_result = agent2.get_action(
side2_obs_dict, step_cnt)
if agent2_result == 0:
side2_detector_action = agent2_action['detector_action']
side2_fighter_action = agent2_action['fighter_action']
elif agent2_result == 1:
agent2_crash_list.append(round_cnt)
elif agent2_result == 2:
agent2_timeout_list.append(round_cnt)
# execution
if agent1_result == 0 and agent2_result == 0:
env.step(side1_detector_action, side1_fighter_action,
side2_detector_action, side2_fighter_action)
elif agent1_result != 0 and agent2_result != 0:
env.set_surrender(2)
elif agent1_result != 0:
env.set_surrender(0)
else:
env.set_surrender(1)
# get done
if env.get_done():
# reward
o_detector_reward, o_fighter_reward, o_game_reward, e_detector_reward, e_fighter_reward, e_game_reward = env.get_reward(
)
if o_game_reward > e_game_reward:
side1_win_times += 1
elif o_game_reward < e_game_reward:
side2_win_times += 1
else:
draw_times += 1
break
agent1.terminate()
agent2.terminate()
return side1_win_times, side2_win_times, draw_times, len(
agent1_crash_list), len(agent2_crash_list), len(
agent1_timeout_list), len(agent2_timeout_list), 0, 0
ATTACK_IND_NUM = (
DETECTOR_NUM + FIGHTER_NUM
) * 2 + 1 # long missile attack + short missile attack + no attack
ACTION_NUM = COURSE_NUM * ATTACK_IND_NUM
LEARN_INTERVAL = 100
if __name__ == "__main__":
# create blue agent
blue_agent = Agent()
# get agent obs type
red_agent_obs_ind = 'feature'
blue_agent_obs_ind = blue_agent.get_obs_ind()
# make env
env = Environment(MAP_PATH,
red_agent_obs_ind,
blue_agent_obs_ind,
max_step=1000,
render=RENDER)
# get map info
size_x, size_y = env.get_map_size()
red_detector_num, red_fighter_num, blue_detector_num, blue_fighter_num = env.get_unit_num(
)
# set map info to blue agent
blue_agent.set_map_info(size_x, size_y, blue_detector_num,
blue_fighter_num)
red_detector_action = []
fighter_model = dqn.RLFighter(ACTION_NUM)
# execution
for x in range(MAX_EPOCH):
ATTACK_IND_NUM = (DETECTOR_NUM + FIGHTER_NUM) * 2 + 1 # long missile attack + short missile attack + no attack
ACTION_NUM = COURSE_NUM * ATTACK_IND_NUM
LEARN_INTERVAL = TARGET_REPLACE_ITER
BETAS = (0.9, 0.999)
EPS_clip = 0.2
K_epochs = 4
max_timesteps = 300
if __name__ == "__main__":
# create blue agent
blue_agent = Agent()
# get agent obs type
red_agent_obs_ind = 'ppo'
blue_agent_obs_ind = blue_agent.get_obs_ind()
# make env
env = Environment(MAP_PATH, red_agent_obs_ind, blue_agent_obs_ind, render=RENDER)
# get map info
size_x, size_y = env.get_map_size()
red_detector_num, red_fighter_num, blue_detector_num, blue_fighter_num = env.get_unit_num()
# set map info to blue agent
blue_agent.set_map_info(size_x, size_y, blue_detector_num, blue_fighter_num)
red_detector_action = []
fighter_model = ppo.PPOFighter(ACTION_NUM,LR,BETAS,GAMMA,EPS_clip,K_epochs,MAX_EPOCH,LEARN_INTERVAL,max_timesteps)
fighter_model.policy.load_state_dict(torch.load('model/ppo/model_000026500.pkl', map_location='cpu'))
fighter_model.policy_old.load_state_dict(torch.load('model/ppo/model_000026500.pkl', map_location='cpu'))
reward_sum = []
round_sum = []
# execution
for x in range(MAX_EPOCH):
else:
step_delay = 1 / args.fps
# environment initiation
if args.log:
if args.log_path == 'default_log':
log_flag = args.agent1 + '_vs_' + args.agent2
else:
log_flag = args.log_path
else:
log_flag = False
env = Environment(map_path,
'raw',
'raw',
max_step=args.max_step,
render=True,
random_pos=args.random_pos,
log=log_flag,
external_render=args.ext_render,
side1_name=args.agent1,
side2_name=args.agent2)
# get map info
size_x, size_y = env.get_map_size()
side1_detector_num, side1_fighter_num, side2_detector_num, side2_fighter_num = env.get_unit_num(
)
# create agent
agent1 = AgentCtrl(args.agent1, size_x, size_y, side1_detector_num,
side1_fighter_num, args.agent1_gpu)
agent2 = AgentCtrl(args.agent2, size_x, size_y, side2_detector_num,
side2_fighter_num, args.agent2_gpu)
if not agent1.agent_init():
dim_obs=OBS_NUM,
dim_act=ACTION_NUM,
batch_size=BATCH_SIZE,
capacity=CAPACITY,
replace_target_iter=TARGET_REPLACE_ITER,
episodes_before_train=EPISODES_BEFORE_TRAIN,
learning_rate=LR,
gamma=GAMMA,
scale_reward=SCALE_REWARD,
is_dispersed=IS_DISPERSED)
# gpu
FloatTensor = th.cuda.FloatTensor if maddpg.use_cuda else th.FloatTensor
# make env
env = Environment(MAP_PATH,
red_agent_obs_ind,
blue_agent_obs_ind,
render=RENDER,
max_step=MAX_STEP,
random_pos=True)
# get map info
size_x, size_y = env.get_map_size() # size_x == size_y == 1000
red_detector_num, red_fighter_num, blue_detector_num, blue_fighter_num = env.get_unit_num(
)
red_detector_action, blue_detector_action = [], [] # temp
for i_episode in range(MAX_EPOCH):
step_cnt = 0
env.reset()
total_reward = 0.0 # 每回合所有智能体的总体奖励
rr = np.zeros((FIGHTER_NUM, )) # 每回合每个智能体的奖励
# get obs
exit(-1)
# delay calc
if args.fps == 0:
step_delay = 0
else:
step_delay = 1 / args.fps
# environment initiation
if args.log:
if args.log_path == 'default_log':
log_flag = args.agent1 + '_vs_' + args.agent2
else:
log_flag = args.log_path
else:
log_flag = False
env = Environment(map_path, 'raw', 'raw', max_step=args.max_step, render=True,
random_pos=args.random_pos, log=log_flag, external_render=args.ext_render)
# get map info
size_x, size_y = env.get_map_size()
side1_detector_num, side1_fighter_num, side2_detector_num, side2_fighter_num = env.get_unit_num()
# create agent
agent1 = AgentCtrl(args.agent1, size_x, size_y, side1_detector_num, side1_fighter_num)
agent2 = AgentCtrl(args.agent2, size_x, size_y, side2_detector_num, side2_fighter_num)
if not agent1.agent_init():
print('ERROR: Agent1 init failed!')
agent1.terminate()
agent2.terminate()
exit(-1)
else:
print('Agent1 init success!')
if not agent2.agent_init():