def agent():
env = RemoteEnv(IN_PORT=8080, OUT_PORT=7070, host="127.0.0.1")
env.open(0)
for i in range(10000000):
#state = env.step("restart")
state = env.step("restart")
prev_energy = state['energy']
done = state['done']
print("OK")
while not done:
action = int(input('action'))
#action = np.random.choice([0, 1])
reward_sum = 0
touched = np.zeros(8)
for i in range(8):
state = env.step("act", action)
energy = state['energy']
touched[i] = state['touched']
reward_sum += state['reward']
if state['done']:
break
done = state['done']
prev_energy = energy
frame = state['frame']
frame = image_decode(state['frame'], 20, 20)
print(frame)
print('Reward: ', reward_sum)
print('Touched: ', touched)
print('Signal: ', state['signal'])
print('Done: ', state['done'])
prev_energy = energy
if i >= 2:
break
env.close()
def get_frame_from_fields(fields, touchs): imgdata = image_decode(fields['frame'], 20, 20) proprioceptions = np.zeros(ARRAY_SIZE) + 0.0001 for i in range(TOUCH_SIZE): proprioceptions[i] = touchs[i] proprioception = np.array(proprioceptions, dtype=np.float32) imgdata = np.vstack([imgdata, proprioception]) return imgdata
def get_frame_from_fields(frame, touchs): imgdata = image_decode(frame, 20, 20, np.uint8) proprioceptions = np.zeros(ARRAY_SIZE) for i in range(TOUCH_SIZE): proprioceptions[i] = touchs[i] proprioception = np.array(proprioceptions, dtype=np.uint8) imgdata = np.vstack([imgdata, proprioception]) return imgdata
def reset(self, env):
self.h = deque(maxlen=H_SIZE)
env_info = env.remoteenv.step("restart")
time.sleep(1)
for i in range(np.random.choice(15)):
env_info = env.one_step(np.random.choice([0, 1, 2]))
f = image_decode(env_info['frame'], 20, 20, dtype=np.float32)
for _ in range(H_SIZE):
self.h.append(f)
return Agent.__make_state__(self.h, np.array(env_info['reward_hist']))
def reset(self, env):
env.remoteenv.step("SetNMoves", 1)
env_info = env.remoteenv.step("restart")
for i in range(np.random.choice(15)):
env_info = env.one_step(np.random.choice([0, 1, 2]))
img = image_decode(env_info['frame'], 20, 20)
self.seq.append(img)
self.seq.append(img)
self.seq.append(img)
self.seq.append(img)
return Agent.__make_state__(self.seq)
def act(self, env, action, info=None):
envinfo = env.one_step(action)
sum_reward = envinfo['reward']
for k in range(1):
if envinfo['done']:
break
else:
envinfo = env.one_step(3)
sum_reward += envinfo['reward']
f = image_decode(envinfo['frame'], 20, 20, dtype=np.float32)
self.h.append(f)
return Agent.__make_state__(self.h, np.array(
envinfo['reward_hist'])), sum_reward, envinfo['done'], envinfo
def get_frame_from_fields(fields):
imgdata = image_decode(fields['frame'], 20, 20)
return imgdata
def transform(frame):
return image_decode(frame, 20, 20)
def step(self, action):
obs, reward, done, info = self.env.step(action)
return image_decode(obs), reward, done, info
def reset(self):
state = self.env.reset()
self.state = image_decode(state)
def to_image(img):
imgdata = image_decode(img, 0, 0)
return imgdata
def act(self, env, action, info=None):
reward_sum, done, env_info = self.__step(env, action)
img = image_decode(env_info['frame'], 20, 20)
self.seq.append(img)
frameseq = Agent.__make_state__(self.seq)
return frameseq, reward_sum, done, env_info