def _setup_spec(self):
action_spec = OrderedDict()
action_selection_spec = OrderedDict()
observation_vec_spec = OrderedDict()
action_spec['steer'] = Box(low=-1., high=1.)
action_spec['motor'] = Box(low=-1., high=1.)
action_space = Box(low=np.array([v.low[0] for k, v in action_spec.items()]),
high=np.array([v.high[0] for k, v in action_spec.items()]))
action_selection_spec['steer'] = Box(low=-1., high=1.)
action_selection_spec['motor'] = Box(low=0.3, high=1.) # min 0.3--> 1.4m/s == 4km/h, need 1km/h to detect collisions
action_selection_space = Box(low=np.array([v.low[0] for k, v in action_selection_spec.items()]),
high=np.array([v.high[0] for k, v in action_selection_spec.items()]))
assert (np.logical_and(action_selection_space.low >= action_space.low,
action_selection_space.high <= action_space.high).all())
num_channels = 0
for camera_name in self._params['cameras']:
camera_params = self._params[camera_name]
assert(camera_params['postprocessing'] is not None)
if camera_params['include_in_obs']:
if camera_params['postprocessing'] == 'SceneFinal':
num_channels += 1 if camera_params.get('grayscale', False) else 3
else:
num_channels += 1
observation_im_space = Box(low=0, high=255, shape=list(self._params['camera_size']) + [num_channels])
observation_vec_spec['coll'] = Discrete(1)
observation_vec_spec['coll_car'] = Discrete(1)
observation_vec_spec['coll_ped'] = Discrete(1)
observation_vec_spec['coll_oth'] = Discrete(1)
observation_vec_spec['heading'] = Box(low=-180., high=180.)
observation_vec_spec['speed'] = Box(low=-30., high=30.)
observation_vec_spec['accel_x'] = Box(low=-100., high=100.)
observation_vec_spec['accel_y'] = Box(low=-100., high=100.)
observation_vec_spec['accel_z'] = Box(low=-100., high=100.)
goal_spec = self._setup_goal_spec()
self.action_spec, self.action_selection_spec, self.observation_vec_spec, self.goal_spec, \
self.action_space, self.action_selection_space, self.observation_im_space = \
action_spec, action_selection_spec, observation_vec_spec, goal_spec, \
action_space, action_selection_space, observation_im_space
self.spec = EnvSpec(
observation_im_space=observation_im_space,
action_space=action_space,
action_selection_space=action_selection_space,
observation_vec_spec=observation_vec_spec,
action_spec=action_spec,
action_selection_spec=action_selection_spec,
goal_spec=goal_spec)
def __init__(self, params={}):
self._yaw_limits = params['yaw_limits']
self._obs_shape = params['obs_shape']
self._horizon = params['horizon']
self.action_spec = OrderedDict()
self.action_selection_spec = OrderedDict()
self.observation_vec_spec = OrderedDict()
self.goal_spec = OrderedDict()
self.action_spec['yaw'] = Box(low=-180, high=180)
self.action_space = Box(low=np.array([self.action_spec['yaw'].low[0]]),
high=np.array([self.action_spec['yaw'].high[0]]))
self.action_selection_spec['yaw'] = Box(low=self._yaw_limits[0], high=self._yaw_limits[1])
self.action_selection_space = Box(low = np.array([self.action_selection_spec['yaw'].low[0]]), high = np.array([self.action_selection_spec['yaw'].high[0]]))
self.observation_im_space = Box(low=0, high=255, shape=self._obs_shape)
self.observation_vec_spec['coll'] = Discrete(1)
self.spec = EnvSpec(
observation_im_space=self.observation_im_space,
action_space=self.action_space,
action_selection_space=self.action_selection_space,
observation_vec_spec=self.observation_vec_spec,
action_spec=self.action_spec,
action_selection_spec=self.action_selection_spec,
goal_spec=self.goal_spec
)
def _setup_spec(self):
self.action_spec = OrderedDict()
self.unnormalized_action_spec = OrderedDict()
self.observation_vec_spec = OrderedDict()
self.goal_spec = OrderedDict()
self.action_spec['steer'] = Box(low=-1., high=1.)
self.unnormalized_action_spec['steer'] = Box(
low=self._steer_limits[0], high=self._steer_limits[1])
if self._use_vel:
if self._fixed_speed:
self.action_spec['speed'] = Box(low=1., high=1.)
self.action_space = Box(low=np.array([-1., 1.]),
high=np.array([1., 1.]))
else:
self.action_spec['speed'] = Box(low=-1., high=1.)
self.action_space = Box(low=np.array([-1., -1.]),
high=np.array([1., 1.]))
self.unnormalized_action_spec['speed'] = Box(
low=self._speed_limits[0], high=self._speed_limits[1])
self.unnormalized_action_space = Box(
low=np.array([self._steer_limits[0], self._speed_limits[0]]),
high=np.array([self._steer_limits[1], self._speed_limits[1]]))
else:
self.action_spec['motor'] = Box(low=-1., high=1.)
self.action_space = Box(low=np.array([-1., -1.]),
high=np.array([1., 1.]))
self.unnormalized_action_spec['motor'] = Box(low=-self._accelClamp,
high=self._accelClamp)
self.unnormalized_action_space = Box(
low=np.array([self._steer_limits[0], -self._accelClamp]),
high=np.array([self._steer_limits[1], self._accelClamp]))
self.observation_im_space = Box(low=0,
high=255,
shape=tuple(
self._get_observation()[0].shape))
self.observation_vec_spec['coll'] = Discrete(1)
self.observation_vec_spec['heading'] = Box(low=0, high=2 * 3.14)
self.observation_vec_spec['speed'] = Box(low=-4.0, high=4.0)
def _setup_spec(self):
self.action_spec = OrderedDict()
self.action_selection_spec = OrderedDict()
self.observation_vec_spec = OrderedDict()
self.goal_spec = OrderedDict()
self.action_spec['steer'] = Box(low=-1., high=1.)
self.action_spec['speed'] = Box(low=-0.3, high=0.3)
self.action_space = Box(low=np.array([
self.action_spec['steer'].low[0], self.action_spec['speed'].low[0]
]),
high=np.array([
self.action_spec['steer'].high[0],
self.action_spec['speed'].high[0]
]))
self.action_selection_spec['steer'] = Box(low=self._steer_limits[0],
high=self._steer_limits[1])
self.action_selection_spec['speed'] = Box(low=self._speed_limits[0],
high=self._speed_limits[1])
self.action_selection_space = Box(
low=np.array([
self.action_selection_spec['steer'].low[0],
self.action_selection_spec['speed'].low[0]
]),
high=np.array([
self.action_selection_spec['steer'].high[0],
self.action_selection_spec['speed'].high[0]
]))
assert (np.logical_and(
self.action_selection_space.low >= self.action_space.low,
self.action_selection_space.high <= self.action_space.high).all())
self.observation_im_space = Box(low=0, high=255, shape=self._obs_shape)
self.observation_vec_spec['coll'] = Discrete(1)
self.observation_vec_spec['heading'] = Box(low=0, high=2 * 3.14)
self.observation_vec_spec['speed'] = Box(low=-0.4, high=0.4)
self.observation_vec_spec['steer'] = Box(low=-1., high=1.)
self.observation_vec_spec['motor'] = Box(low=-1., high=1.)
def _setup_spec(self):
self.action_spec = OrderedDict()
self.action_selection_spec = OrderedDict()
self.observation_vec_spec = OrderedDict()
self.goal_spec = OrderedDict()
self.action_spec['yaw'] = Box(low=-180, high=180)
self.action_space = Box(low=np.array([self.action_spec['yaw'].low[0]]),
high=np.array(
[self.action_spec['yaw'].high[0]]))
self.action_selection_spec['yaw'] = Box(low=self._yaw_limits[0],
high=self._yaw_limits[1])
self.action_selection_space = Box(
low=np.array([self.action_selection_spec['yaw'].low[0]]),
high=np.array([self.action_selection_spec['yaw'].high[0]]))
assert (np.logical_and(
self.action_selection_space.low >= self.action_space.low,
self.action_selection_space.high <= self.action_space.high).all())
self.observation_im_space = Box(low=0, high=255, shape=self._obs_shape)
self.observation_vec_spec['coll'] = Discrete(1)
def _setup_spec(self):
self.action_spec = OrderedDict()
self.action_selection_spec = OrderedDict()
self.observation_vec_spec = OrderedDict()
self.goal_spec = OrderedDict()
self.action_spec['steer'] = Box(low=-45., high=45.)
self.action_selection_spec['steer'] = Box(low=self._steer_limits[0], high=self._steer_limits[1])
if self._use_vel:
self.action_spec['speed'] = Box(low=-4., high=4.)
self.action_space = Box(low=np.array([self.action_spec['steer'].low[0], self.action_spec['speed'].low[0]]),
high=np.array([self.action_spec['steer'].high[0], self.action_spec['speed'].high[0]]))
self.action_selection_spec['speed'] = Box(low=self._speed_limits[0], high=self._speed_limits[1])
self.action_selection_space = Box(low=np.array([self.action_selection_spec['steer'].low[0],
self.action_selection_spec['speed'].low[0]]),
high=np.array([self.action_selection_spec['steer'].high[0],
self.action_selection_spec['speed'].high[0]]))
else:
self.action_spec['motor'] = Box(low=-self._accelClamp, high=self._accelClamp)
self.action_space = Box(low=np.array([self.action_spec['steer'].low[0], self.action_spec['motor'].low[0]]),
high=np.array([self.action_spec['steer'].high[0], self.action_spec['motor'].high[0]]))
self.action_selection_spec['motor'] = Box(low=self._motor_limits[0], high=self._motor_limits[1])
self.action_selection_space = Box(low=np.array([self.action_selection_spec['steer'].low[0],
self.action_selection_spec['motor'].low[0]]),
high=np.array([self.action_selection_spec['steer'].high[0],
self.action_selection_spec['motor'].high[0]]))
assert (np.logical_and(self.action_selection_space.low >= self.action_space.low - 1e-4,
self.action_selection_space.high <= self.action_space.high + 1e-4).all())
self.observation_im_space = Box(low=0, high=255, shape=tuple(self._get_observation()[0].shape))
self.observation_vec_spec['coll'] = Discrete(1)
self.observation_vec_spec['heading'] = Box(low=0, high=2 * 3.14)
self.observation_vec_spec['speed'] = Box(low=-4.0, high=4.0)