def _update_observation(self):
return_obs = Observation()
y_ankle = NaoWorldModel().get_joint_pos(self._ankle_roll)[1]
y_hip = NaoWorldModel().get_joint_pos(self._hip_roll)[1]
displacement = (y_ankle - y_hip) * 1000 # convert to millimeters
if self._feature_rep == 'rl':
image_x = NaoWorldModel().get_object("ball").image_center.x
return_obs.intArray = [image_x, displacement]
else:
s = [0] * MDPState.nfeatures
bin_num = int(math.floor((displacement - MDPState.min_features) / self._bin_width))
try:
s[bin_num] = 1
except IndexError, e:
exit(e)
return_obs.intArray = s
def start(self):
if self.fixed_start_state:
state_valid = self.set_agent_state(self.start_row, self.start_col)
if not state_valid:
print "The fixed start state was NOT valid: " + str(int(self.start_row)) + "," + str(int(self.start_row))
self.set_random_state()
else:
self.set_random_state()
return_obs = Observation()
return_obs.intArray = [self.calculate_flat_state()]
return return_obs
def step(self, action):
# Make sure the action is valid
assert len(action.intArray) == 1, "Expected 1 integer action."
assert action.intArray[0] >= 0, "Expected action to be in [0,3]"
assert action.intArray[0] < 4, "Expected action to be in [0,3]"
self.update_position(action.intArray[0])
obs = Observation()
obs.intArray = [self.calculate_flat_state()]
return_ro = Reward_observation_terminal()
return_ro.r = self.calculate_reward()
return_ro.o = obs
return_ro.terminal = self.check_current_terminal()
return return_ro
