break
env.render()
cnt = cnt + 1
action = RL.choose_action(observation) # 选行为
action_onehot = np.zeros(env.action_space[0].n)
action_onehot[action] += 1.0
observation_, reward, done, info = env.step([action_onehot
]) # 获取下一个 state
observation_ = observation_[0]
reward = reward[0]
done = done[0]
# 保存这一组记忆
RL.store_transition(observation, action, reward, observation_)
# 学习
RL.learn()
observation = observation_
# display rewards
for agent in env.world.agents:
if not agent.name in reward_dic:
reward_dic[agent.name] = []
reward_dic[agent.name].append(env._get_reward(agent))
print(agent.name + " reward: %0.3f" % env._get_reward(agent))
for agent in env.world.agents:
y = reward_dic[agent.name]
x = np.linspace(1, len(y), len(y))
plt.plot(x, y)
plt.show()
# create interactive policies for each agent
policies = [InteractivePolicy(env, i) for i in range(env.n)]
# execution loop
obs_n = env.reset()
while True:
# query for action from each agent's policy
act_n = []
observation = obs_n[0]
length = observation.shape[1]
for i, policy in enumerate(policies):
if i < length - 1:
obs = observation[i, :]
act_n.append(policy.action(obs))
if i == length - 1:
act_n.append(policy.action(observation))
# step environment
obs_n, reward_n, done_n, _ = env.step(act_n)
# render all agent views
env.render()
#print action
print(act_n)
#get observation/state
env_obs = obs_n[0]
print(env_obs)
#get reward
env_reward = []
for agent in env.world.agents:
env_reward.append(env._get_reward(agent))
print(env_reward)
# load scenario from script
scenario = scenarios.load(args.scenario).Scenario()
# create world
world = scenario.make_world() # world is a class of world
# create multiagent environment
env = MultiAgentEnv(world,
scenario.reset_world,
scenario.reward,
scenario.observation,
info_callback=None,
shared_viewer=False)
# render call to create viewer window (necessary only for interactive policies)
env.render()
# create interactive policies for each agent
policies = [InteractivePolicy(env, i) for i in range(env.n)
] # create for each agent in the simulation world
# execution loop
obs_n = env.reset()
while True:
# query for action from each agent's policy
act_n = []
for i, policy in enumerate(policies):
act_n.append(policy.action(obs_n[i]))
# step environment
obs_n, reward_n, done_n, _ = env.step(act_n)
# render all agent views
env.render()
# display rewards
for agent in env.world.agents:
print(agent.name + " reward: %0.3f" % env._get_reward(agent))
env.render()
# create interactive policies for each agent
policies = [InteractivePolicy(env,i) for i in range(env.n)]
# execution loop
obs_n = env.reset()
rewards = np.zeros(len(env.world.agents))
print ('env.discrete_action_space:', env.discrete_action_space)
while True:
# query for action from each agent's policy
act_n = []
for i, policy in enumerate(policies):
act_n.append(policy.action(obs_n[i]))
# step environment
obs_n, reward_n, done_n, _ = env.step(act_n)
# render all agent views
env.render()
# display rewards
new_rewards = np.zeros(len(env.world.agents))
for i, agent in enumerate(env.world.agents):
new_rewards[i] = env._get_reward(agent)
if (np.abs(rewards - new_rewards) > .001).any():
print(rewards - new_rewards)
rewards = new_rewards
for i, r in enumerate(rewards):
print('agent {} reward: {:.3f}'.format(i, r))
