def enable_display(self):
if not self.display:
self.display = fs.Display(self.map, self.robot)
self.display.update()
def __init__(self,
xml_env,
reward_func="distance_multigoal",
display=False,
light_sensor_range=200.,
light_sensor_mode="realistic"):
# Fastsim setup
# XML files typically contain relative names (for map) wrt their own path. Make that work
xml_dir, xml_file = os.path.split(xml_env)
if (xml_dir == ""):
xml_dir = "./"
oldcwd = os.getcwd()
os.chdir(xml_dir)
settings = fs.Settings(xml_file)
os.chdir(oldcwd)
self.map = settings.map()
self.background = np.expand_dims(netpbmfile.imread(
os.path.join(xml_dir, 'maze_hard.pbm')),
axis=-1)
self.background_size = [200, 200]
self.map_size = [600, 600]
self.robot = settings.robot()
if (display):
self.display = fs.Display(self.map, self.robot)
else:
self.display = None
self.maxVel = 4 # Same as in the C++ sferes2 experiment
# Lasers
lasers = self.robot.get_lasers()
n_lasers = len(lasers)
if (n_lasers > 0):
self.maxSensorRange = lasers[0].get_range(
) # Assume at least 1 laser ranger
else:
self.maxSensorRange = 0.
#Light sensors
self.ls_mode = light_sensor_mode
lightsensors = self.robot.get_light_sensors()
n_lightsensors = len(lightsensors)
if (n_lightsensors > 0):
self.maxLightSensorRange = light_sensor_range # Assume at least 1 laser ranger
else:
self.maxLightSensorRange = 0.
# State
self.initPos = self.get_robot_pos()
self.current_pos = self.get_robot_pos()
self.old_pos = self.get_robot_pos()
self.v1_motor_order = 0.
self.v2_motor_order = 0.
self.goal = self.map.get_goals()
self.goalPos = [[goal.get_x(), goal.get_y()] for goal in self.goal]
self.goalRadius = [goal.get_diam() / 2. for goal in self.goal]
self.observation_space = spaces.Box(
low=np.array([0.] * n_lasers + [0.] * 2 +
[0. if (self.ls_mode == "realistic") else -1] *
n_lightsensors),
high=np.array([self.maxSensorRange] * n_lasers + [
1. if
(self.ls_mode == "realistic") else self.maxLightSensorRange
] * 2 + [1.] * n_lightsensors,
dtype=np.float32))
self.action_space = spaces.Box(low=-self.maxVel,
high=self.maxVel,
shape=(2, ),
dtype=np.float32)
self.max_steps = 2000
self.step_count = 0
# Reward
self.reward_area = None
if (reward_func not in reward_functions):
raise RuntimeError("Unknown reward '%s'" % str(reward_func))
else:
self.reward_func = reward_functions[reward_func]