import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import sys
sys.path.insert(1, 'env/')
from env import envs
from maddpg import MaDDPG
state_dim = 5
action_dim = 1
max_edge = 1
num_agents = 3
maddpg = MaDDPG(num_agents, state_dim, action_dim)
Env = envs.Environ(num_agents, max_edge)
obs = Env.reset()
current_state = obs
max_episode = 1000000
done_epoch = 0
#print(current_state)
max_epoch = 1000
catch_time = []
for episode in range(max_episode):
print('episode', episode)
#while (True):
#Env.re_create_env(num_agents)
import numpy as np
from ma_policy import MaPolicy
from maddpg import MaDDPG
from torch import round
DEFAULT_LEARN_STEPS = 10000
DEFAULT_TEST_STEPS = 10000
ENV_NAMES = "PongDuel-v0"
OUTPUT_FRAMES_DIR = 'output/video.npz'
if __name__ == '__main__':
model = MaDDPG(
policy=MaPolicy,
env=ENV_NAMES,
mapper=lambda actions: round(actions * 1.5 + 1).flatten().tolist(),
verbose=True)
model.learn(DEFAULT_LEARN_STEPS)
results = []
for interact in model.execute(DEFAULT_TEST_STEPS):
if isinstance(interact, np.ndarray):
results.append(interact)
if len(results) != 0:
np.savez(OUTPUT_FRAMES_DIR, frames=np.array(results))
def system_destroyed():
disfunction_height = 1.0
if (fabs(payload_euler_angle[0, 0]) > 1.0) or (fabs(
payload_euler_angle[1, 0]) > 1.0) or (payload_height <
disfunction_height):
return True
else:
return False
if __name__ == '__main__':
state_dim = 20
action_dim = 3
num_agents = 4
maddpg = MaDDPG(num_agents, state_dim, action_dim)
rospy.init_node('multi_UAV_gym', anonymous=False, log_level=rospy.INFO)
env = gym.make('multi_UAV-v0')
topic_name_odom = '/payload/ground_truth/odometry'
sub_payload_odometry = rospy.Subscriber(topic_name_odom, Odometry,
callback_payload_odometry)
#obs_shape_n = [env.observation_space[i].shape for i in range(4)]
maddpg.load_network()
#trainers = get_trainers(env, obs_shape_n, params)
f_reward = open(reward_file, 'r+')
episode = 0
for line in f_reward.readlines():
episode = episode + 1
f_reward.close()
from maddpg import MaDDPG
import numpy as np
from maddpg import MaDDPG
import numpy as np
state_dim = 3
action_dim =2
num_agents = 4
maddpg = MaDDPG(num_agents,state_dim,action_dim)
for ii in range(2000):
print('time step {}'.format(ii))
state =np.random.rand(num_agents,state_dim)
next_state =np.random.rand(num_agents,state_dim)
action = np.random.rand(num_agents, action_dim)
reward = np.random.rand(num_agents)
done = 0
takeaction = maddpg.noise_action(state)
maddpg.perceive(state,action,reward, next_state,done)
maddpg.close_session()
from maddpg import MaDDPG import numpy as np state_dim = 3 action_dim = 2 num_agents = 2 states = np.ones((4, state_dim)) states_batch = np.ones((2, 4, state_dim)) maddpg = MaDDPG(num_agents, state_dim, action_dim) maddpg.add_agents(2) print(maddpg.action(states)) actions = maddpg.target_actions(states_batch) noise_action = maddpg.noise_action(states) print(noise_action) maddpg.close_session() #print(maddpg.num_agents)
import numpy as np
import time
import sys
sys.path.insert(1, 'env/')
from env import envs
from maddpg import MaDDPG
# load the pre-trained network and plays the video slowly
state_dim = 5
action_dim = 1
max_edge = 1
num_agents = 3
maddpg = MaDDPG(num_agents, state_dim, action_dim)
# load saved network
maddpg.load_network()
Env = envs.Environ(num_agents, max_edge)
obs = Env.reset()
current_state = obs
max_time = 10000
#print(current_state)
for epoch in range(max_time):
print('epoch', epoch)
action = maddpg.action(current_state)
#print(action)
next_state, reward, done = Env.step(action)
#print(reward)
#maddpg.perceive(current_state,action,reward,next_state,done)
current_state = next_state