def _update_observation(self):
return_obs = Observation()
return_obs.doubleArray = np.zeros(MDPState.nfeatures)
for effector, mapping in MDPState.description.iteritems():
pos = NaoWorldModel().get_effector_pos(effector)
for key, axis in mapping.iteritems():
return_obs.doubleArray[axis] = pos[MDPState.key_to_index(key)]
return return_obs
def start(self):
if self._random_start:
self._sensors = np.random.random(self._sensors.shape)
self._sensors *= (self._limits[:, 1] - self._limits[:, 0])
self._sensors += self._limits[:, 0]
else:
self._sensors = np.zeros(self._sensors.shape)
self._sensors[0] = -0.5
self._render(self._sensors[0])
return_obs = Observation()
return_obs.doubleArray = self._sensors.tolist()
return return_obs
def step(self, action):
return_ro = Reward_observation_terminal()
self._apply(action)
self._render(self._sensors[0])
return_ro.terminal = self._is_terminal()
return_ro.r = -1.
if return_ro.terminal:
return_ro.r = .0
if self._reward_noise > 0:
return_ro.r += np.random.normal(scale=self._reward_noise)
obs = Observation()
obs.doubleArray = self._sensors.tolist()
return_ro.o = obs
return return_ro