def run(self, img_arr):
# Prepare image array.
img_arr = img_arr.reshape((1, ) + img_arr.shape)
# NN angle prediction.
angle_binned, _ = self.model.predict(img_arr)
angle_unbinned = dkutil.linear_unbin(angle_binned[0])
# TODO: Compute average (SMA) confidence.
angle_confidence = max(angle_binned[0])
# Fallback mode. If enabled, do obstacle detection.
if self.obstacle:
angle_obstacle = self.compute_angle_obstacle(img_arr)
# In case we detected obstacle use fallback angle.
if angle_obstacle:
angle_unbinned = angle_obstacle
print('obstacle detected, using fallback mode.')
# Angle postprocessing.
angle = self.compute_angle(angle_unbinned)
# By default we always run at failsafe speed.
throttle = self.config['throttle_failsafe']
# Speed-up if NN is confident in what it saw.
if float(angle_confidence) > float(
self.config['acceleration_confidence']):
throttle = self.compute_throttle(angle)
# Workaround to make car slower.
if self.skip_throttle():
throttle = 0
print('throttle:', throttle, 'angle:', angle)
return angle, throttle
def test_illegal_list(self):
with pytest.raises(ValueError):
linear_unbin([0] * 10)
def test_empty_list(self):
res = linear_unbin([0] * 15)
assert res == -1.0
def test_negative(self):
l = create_lbin(0)
res = linear_unbin(l)
assert res == -1.0
def test_positive(self):
l = create_lbin(14)
res = linear_unbin(l)
assert res == 1.0
def test_zero(self):
l = create_lbin(7)
res = linear_unbin(l)
assert res == 0.0
