def filter(self,
observation: ObservationType,
update_internal_state: bool = True) -> ObservationType:
# scipy works only with uint8
observation = observation.astype('uint8')
# rescale
if isinstance(self.output_observation_space, ImageObservationSpace):
observation = scipy.misc.imresize(
observation,
tuple(self.output_observation_space.shape),
interp=self.rescaling_interpolation_type.value)
else:
new_observation = []
for i in range(self.output_observation_space.shape[
self.output_observation_space.channels_axis]):
new_observation.append(
scipy.misc.imresize(
observation.take(
i, self.output_observation_space.channels_axis),
tuple(self.planar_map_output_shape),
interp=self.rescaling_interpolation_type.value))
new_observation = np.array(new_observation)
observation = new_observation.swapaxes(
0, self.output_observation_space.channels_axis)
return observation
def filter(self, observation: ObservationType, update_internal_state: bool=True) -> ObservationType:
observation = observation.astype('uint8')
# rescale
if isinstance(self.output_observation_space, ImageObservationSpace):
observation = resize(observation, tuple(self.output_observation_space.shape), anti_aliasing=False,
preserve_range=True).astype('uint8')
else:
new_observation = []
for i in range(self.output_observation_space.shape[self.output_observation_space.channels_axis]):
new_observation.append(resize(observation.take(i, self.output_observation_space.channels_axis),
tuple(self.planar_map_output_shape),
preserve_range=True).astype('uint8'))
new_observation = np.array(new_observation)
observation = new_observation.swapaxes(0, self.output_observation_space.channels_axis)
return observation
def filter(self, observation: ObservationType, update_internal_state: bool=True) -> ObservationType:
# scale to 0-1
observation = (observation - self.input_low) / (self.input_high - self.input_low)
# scale to 0-255
observation *= 255
observation = observation.astype('uint8')
return observation
def filter(self,
observation: ObservationType,
update_internal_state: bool = True) -> ObservationType:
# scipy works only with uint8
observation = observation.astype('uint8')
# rescale
observation = scipy.misc.imresize(
observation,
self.rescale_factor,
interp=self.rescaling_interpolation_type.value)
return observation
def filter(self,
observation: ObservationType,
update_internal_state: bool = True) -> ObservationType:
observation = observation.astype('uint8')
rescaled_output_size = tuple(
[int(self.rescale_factor * dim) for dim in observation.shape[:2]])
if len(observation.shape) == 3:
rescaled_output_size += (3, )
# rescale
observation = resize(observation,
rescaled_output_size,
anti_aliasing=False,
preserve_range=True).astype('uint8')
return observation
def filter(self,
observation: ObservationType,
update_internal_state: bool = True) -> ObservationType:
# Divide the raw lidar data into sectors and take the min from raw lidar data in each sector
observation = np.min(observation.reshape(
-1, int(observation.shape[0] / self._num_sectors)),
axis=1)
# Clip the observation data to clipping distance.
np.minimum(observation, self._clipping_dist, out=observation)
# Discretize the observation sector data
observation = np.floor(observation / self._discrete_range).astype(
np.int)
# One hot encode the discretized observation sector data
one_hot_encode = np.zeros(
(observation.size, self._num_values_per_sector + 1))
one_hot_encode[np.arange(observation.size), observation] = 1
# Get rid of last column as it means lidar didn't detect anything
one_hot_encode = np.delete(one_hot_encode, -1, axis=1)
result = one_hot_encode.reshape(-1).astype(float)
return result
def filter(self,
observation: ObservationType,
update_internal_state: bool = True) -> ObservationType:
observation = np.min(observation.reshape(-1, NUMBER_OF_LIDAR_SECTORS),
axis=1)
return (observation < SECTOR_LIDAR_CLIPPING_DIST).astype(float)
def filter(self,
observation: ObservationType,
update_internal_state: bool = True) -> ObservationType:
return observation.squeeze(axis=self.axis)
