def reset(self):
self.fitnesses = np.full((1, self.pop_size), -100.0) # init to -100
# array of population indexes
self.net_ids = np.arange(self.pop_size).reshape(1, self.pop_size)
# initialize robot angles (1, pop_size)
self.robot_angles = np.full(
(1, self.pop_size), utils.to_minus_pi_pi(self.conf.init_rotation))
# initialize sensor angles of robots
self.sensor_angles = np.zeros(
(len(self.robot.sensor_angles),
self.pop_size)) + utils.to_minus_pi_pi(self.robot.sensor_angles)
# initialize robot positions (1, pop_size, 2)
self.robot_positions = np.full((1, self.pop_size, 2),
self.map.start_point)
self.robot_bodies = np.add(self.robot.body,
np.rint(self.robot_positions)).astype(int)
self.sensor_lines = self.__update_sensor_lines()
# calculate angle errors and then normalize them
self.angle_errors = self.__update_angle_errors()
self.normalized_angle_errors = self.__normalize_angle_errors()
# calculate target distances and then normalize them
self.target_distances = self.__update_target_distances()
self.normalized_target_distances = self.__normalize_target_distances()
self.max_distance_from_start = np.zeros((1, self.pop_size))
def reset(self, pop_size=None):
if pop_size is not None:
self.pop_size = pop_size
self.fits = np.full((1, self.pop_size),
-100000.0) # init fitness to -100000
# initialize robot angles (1, pop_size)
self.robot_angles = np.full((1, self.pop_size),
utils.to_minus_pi_pi(self.init_rotation))
# initialize sensor angles of robots
self.sensor_angles = np.zeros(
(len(self.robot.sensor_angles),
self.pop_size)) + utils.to_minus_pi_pi(self.robot.sensor_angles)
# initialize robot positions (1, pop_size, 2)
self.robot_positions = np.full((1, self.pop_size, 2),
self.sim_map.start_point)
self.robot_bodies = np.add(self.robot.body,
np.rint(self.robot_positions)).astype(int)
self.sensor_lines = self.__update_sensor_lines()
# calculate angle errors and then normalize them
self.angle_errors = self.__update_angle_errors()
self.normalized_angle_errors = self.__normalize_angle_errors()
# calculate target distances and then normalize them
self.target_distances = self.__update_target_distances()
self.normalized_target_distances = self.__normalize_target_distances()
self.max_distance_from_start = np.zeros((1, self.pop_size))
self.collision_counters = np.zeros((1, self.pop_size))
self.cumulative_angle_errs = np.zeros((1, self.pop_size))
self.fig = None
self.img = None
def __rotate(self, delta_angles):
self.robot_angles = utils.to_minus_pi_pi(
np.add(self.robot_angles, delta_angles))
self.sensor_angles = utils.to_minus_pi_pi(
np.add(self.sensor_angles, delta_angles))
return self.robot_angles, self.sensor_angles
def __update_angle_errors(self):
self.angle_errors = utils.calc_angle_error(self.robot_positions,
self.map.end_point,
self.robot_angles)
self.angle_errors = utils.to_minus_pi_pi(self.angle_errors)
return self.angle_errors