-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdqncarla.py
1323 lines (1107 loc) · 53.9 KB
/
dqncarla.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#Code is modified from Felippe Roza's implementation which uses a
#Double DQN with Priority Experience Replay, credits: https://github.com/FelippeRoza/carla-rl
#Additionally referenced Simonini Thomas's Deep learning course
#Credits: https://github.com/simoninithomas/Deep_reinforcement_learning_Course
import glob
import os
import sys
try:
sys.path.append(glob.glob('../carla/dist/carla-*%d.%d-%s.egg' % (
sys.version_info.major,
sys.version_info.minor,
'win-amd64' if os.name == 'nt' else 'linux-x86_64'))[0])
except IndexError:
pass
import carla
from carla import ColorConverter as cc
import time
import random
import numpy as np
import argparse
import logging
import pygame
import tensorflow as tf
import queue
import datetime
from PIL import Image
from multiprocessing import Process
from collections import deque
import re
import weakref
import math
import collections
import pickle
import matplotlib.pyplot as plt
test_flag = 0
model_name = "DQNetwork"
state_size = [84, 84, 1]
# discrete action-space described as (throttle, steer, brake)
action_space = np.array([(0.0, 0.0, 1.0), (0.5, 0.0, 0.0), (1.0, 0.0, 0.0),
(0.5, 0.25, 0.0), (0.5, -0.25, 0.0), (0.5, 0.5, 0.0), (0.5, -0.5, 0.0)])
action_size = len(action_space)
learning_rate= 0.00025
# Training parameters
total_episodes = 50 # INTIALLY 5000
max_steps = 200
batch_size = 64
# Fixed Q target hyper parameters
max_tau = 5000 # tau is the C step where we update out target network -- INTIALLY 10000
# exploration hyperparamters for ep. greedy. startegy
explore_start = 1.0 # exploration probability at start
explore_stop = 0.01 # minimum exploration probability
decay_rate = 0.00005 # exponential decay rate for exploration prob
# Q LEARNING hyperparameters
gamma = 0.95 # Discounting rate
pretrain_length = 100 ## Number of experiences stored in the Memory when initialized for the first time --INTIALLY 100k
memory_size = 10000 # Number of experiences the Memory can keep --INTIALLY 100k
memory_save_path = "memory.pkl"
# ==============================================================================
# -- FadingText ----------------------------------------------------------------
# ==============================================================================
class FadingText(object):
def __init__(self, font, dim, pos):
self.font = font
self.dim = dim
self.pos = pos
self.seconds_left = 0
self.surface = pygame.Surface(self.dim)
def set_text(self, text, color=(255, 255, 255), seconds=2.0):
text_texture = self.font.render(text, True, color)
self.surface = pygame.Surface(self.dim)
self.seconds_left = seconds
self.surface.fill((0, 0, 0, 0))
self.surface.blit(text_texture, (10, 11))
def tick(self, _, clock):
delta_seconds = 1e-3 * clock.get_time()
self.seconds_left = max(0.0, self.seconds_left - delta_seconds)
self.surface.set_alpha(500.0 * self.seconds_left)
def render(self, display):
display.blit(self.surface, self.pos)
# ==============================================================================
# -- HelpText ------------------------------------------------------------------
# ==============================================================================
class HelpText(object):
"""Helper class to handle text output using pygame"""
def __init__(self, font, width, height):
lines = __doc__.split('\n')
self.font = font
self.line_space = 18
self.dim = (780, len(lines) * self.line_space + 12)
self.pos = (0.5 * width - 0.5 * self.dim[0], 0.5 * height - 0.5 * self.dim[1])
self.seconds_left = 0
self.surface = pygame.Surface(self.dim)
self.surface.fill((0, 0, 0, 0))
for n, line in enumerate(lines):
text_texture = self.font.render(line, True, (255, 255, 255))
self.surface.blit(text_texture, (22, n * self.line_space))
self._render = False
self.surface.set_alpha(220)
def toggle(self):
self._render = not self._render
def render(self, display):
if self._render:
display.blit(self.surface, self.pos)
def get_actor_display_name(actor, truncate=250):
name = ' '.join(actor.type_id.replace('_', '.').title().split('.')[1:])
return (name[:truncate - 1] + u'\u2026') if len(name) > truncate else name
def find_weather_presets():
rgx = re.compile('.+?(?:(?<=[a-z])(?=[A-Z])|(?<=[A-Z])(?=[A-Z][a-z])|$)')
name = lambda x: ' '.join(m.group(0) for m in rgx.finditer(x))
presets = [x for x in dir(carla.WeatherParameters) if re.match('[A-Z].+', x)]
return [(getattr(carla.WeatherParameters, x), name(x)) for x in presets]
class CollisionSensor(object):
def __init__(self, parent_actor, hud):
self.sensor = None
self.history = []
self._parent = parent_actor
self.hud = hud
world = self._parent.get_world()
bp = world.get_blueprint_library().find('sensor.other.collision')
self.sensor = world.spawn_actor(bp, carla.Transform(), attach_to=self._parent)
# We need to pass the lambda a weak reference to self to avoid circular
# reference.
weak_self = weakref.ref(self)
self.sensor.listen(lambda event: CollisionSensor._on_collision(weak_self, event))
def get_collision_history(self):
history = collections.defaultdict(int)
for frame, intensity in self.history:
history[frame] += intensity
return history
@staticmethod
def _on_collision(weak_self, event):
self = weak_self()
if not self:
return
actor_type = get_actor_display_name(event.other_actor)
self.hud.notification('Collision with %r' % actor_type)
impulse = event.normal_impulse
intensity = math.sqrt(impulse.x**2 + impulse.y**2 + impulse.z**2)
self.history.append((event.frame, intensity))
if len(self.history) > 4000:
self.history.pop(0)
# ==============================================================================
# -- LaneInvasionSensor --------------------------------------------------------
# ==============================================================================
class LaneInvasionSensor(object):
def __init__(self, parent_actor, hud):
self.sensor = None
self._parent = parent_actor
self.hud = hud
world = self._parent.get_world()
bp = world.get_blueprint_library().find('sensor.other.lane_invasion')
self.sensor = world.spawn_actor(bp, carla.Transform(), attach_to=self._parent)
# We need to pass the lambda a weak reference to self to avoid circular
# reference.
weak_self = weakref.ref(self)
self.sensor.listen(lambda event: LaneInvasionSensor._on_invasion(weak_self, event))
@staticmethod
def _on_invasion(weak_self, event):
self = weak_self()
if not self:
return
lane_types = set(x.type for x in event.crossed_lane_markings)
text = ['%r' % str(x).split()[-1] for x in lane_types]
self.hud.notification('Crossed line %s' % ' and '.join(text))
# ==============================================================================
# -- GnssSensor ----------------------------------------------------------------
# ==============================================================================
class GnssSensor(object):
def __init__(self, parent_actor):
self.sensor = None
self._parent = parent_actor
self.lat = 0.0
self.lon = 0.0
world = self._parent.get_world()
bp = world.get_blueprint_library().find('sensor.other.gnss')
self.sensor = world.spawn_actor(bp, carla.Transform(carla.Location(x=1.0, z=2.8)), attach_to=self._parent)
# We need to pass the lambda a weak reference to self to avoid circular
# reference.
weak_self = weakref.ref(self)
self.sensor.listen(lambda event: GnssSensor._on_gnss_event(weak_self, event))
@staticmethod
def _on_gnss_event(weak_self, event):
self = weak_self()
if not self:
return
self.lat = event.latitude
self.lon = event.longitude
# ==============================================================================
# -- IMUSensor -----------------------------------------------------------------
# ==============================================================================
class IMUSensor(object):
def __init__(self, parent_actor):
self.sensor = None
self._parent = parent_actor
self.accelerometer = (0.0, 0.0, 0.0)
self.gyroscope = (0.0, 0.0, 0.0)
self.compass = 0.0
world = self._parent.get_world()
bp = world.get_blueprint_library().find('sensor.other.imu')
self.sensor = world.spawn_actor(
bp, carla.Transform(), attach_to=self._parent)
# We need to pass the lambda a weak reference to self to avoid circular
# reference.
weak_self = weakref.ref(self)
self.sensor.listen(
lambda sensor_data: IMUSensor._IMU_callback(weak_self, sensor_data))
@staticmethod
def _IMU_callback(weak_self, sensor_data):
self = weak_self()
if not self:
return
limits = (-99.9, 99.9)
self.accelerometer = (
max(limits[0], min(limits[1], sensor_data.accelerometer.x)),
max(limits[0], min(limits[1], sensor_data.accelerometer.y)),
max(limits[0], min(limits[1], sensor_data.accelerometer.z)))
self.gyroscope = (
max(limits[0], min(limits[1], math.degrees(sensor_data.gyroscope.x))),
max(limits[0], min(limits[1], math.degrees(sensor_data.gyroscope.y))),
max(limits[0], min(limits[1], math.degrees(sensor_data.gyroscope.z))))
self.compass = math.degrees(sensor_data.compass)
# ==============================================================================
# -- RadarSensor ---------------------------------------------------------------
# ==============================================================================
class RadarSensor(object):
def __init__(self, parent_actor):
self.sensor = None
self._parent = parent_actor
self.velocity_range = 7.5 # m/s
world = self._parent.get_world()
self.debug = world.debug
bp = world.get_blueprint_library().find('sensor.other.radar')
bp.set_attribute('horizontal_fov', str(35))
bp.set_attribute('vertical_fov', str(20))
self.sensor = world.spawn_actor(
bp,
carla.Transform(
carla.Location(x=2.8, z=1.0),
carla.Rotation(pitch=5)),
attach_to=self._parent)
# We need a weak reference to self to avoid circular reference.
weak_self = weakref.ref(self)
self.sensor.listen(
lambda radar_data: RadarSensor._Radar_callback(weak_self, radar_data))
@staticmethod
def _Radar_callback(weak_self, radar_data):
self = weak_self()
if not self:
return
# To get a numpy [[vel, altitude, azimuth, depth],...[,,,]]:
# points = np.frombuffer(radar_data.raw_data, dtype=np.dtype('f4'))
# points = np.reshape(points, (len(radar_data), 4))
current_rot = radar_data.transform.rotation
for detect in radar_data:
azi = math.degrees(detect.azimuth)
alt = math.degrees(detect.altitude)
# The 0.25 adjusts a bit the distance so the dots can
# be properly seen
fw_vec = carla.Vector3D(x=detect.depth - 0.25)
carla.Transform(
carla.Location(),
carla.Rotation(
pitch=current_rot.pitch + alt,
yaw=current_rot.yaw + azi,
roll=current_rot.roll)).transform(fw_vec)
def clamp(min_v, max_v, value):
return max(min_v, min(value, max_v))
norm_velocity = detect.velocity / self.velocity_range # range [-1, 1]
r = int(clamp(0.0, 1.0, 1.0 - norm_velocity) * 255.0)
g = int(clamp(0.0, 1.0, 1.0 - abs(norm_velocity)) * 255.0)
b = int(abs(clamp(- 1.0, 0.0, - 1.0 - norm_velocity)) * 255.0)
self.debug.draw_point(
radar_data.transform.location + fw_vec,
size=0.075,
life_time=0.06,
persistent_lines=False,
color=carla.Color(r, g, b))
# ==============================================================================
# -- CameraManager -------------------------------------------------------------
# ==============================================================================
class CameraManager(object):
def __init__(self, parent_actor, hud, gamma_correction):
self.sensor = None
self.surface = None
self._parent = parent_actor
self.hud = hud
self.recording = False
bound_y = 0.5 + self._parent.bounding_box.extent.y
Attachment = carla.AttachmentType
self._camera_transforms = [
(carla.Transform(carla.Location(x=-5.5, z=2.5), carla.Rotation(pitch=8.0)), Attachment.SpringArm),
(carla.Transform(carla.Location(x=1.6, z=1.7)), Attachment.Rigid),
(carla.Transform(carla.Location(x=5.5, y=1.5, z=1.5)), Attachment.SpringArm),
(carla.Transform(carla.Location(x=-8.0, z=6.0), carla.Rotation(pitch=6.0)), Attachment.SpringArm),
(carla.Transform(carla.Location(x=-1, y=-bound_y, z=0.5)), Attachment.Rigid)]
self.transform_index = 1
self.sensors = [
['sensor.camera.rgb', cc.Raw, 'Camera RGB', {}],
['sensor.camera.depth', cc.Raw, 'Camera Depth (Raw)', {}],
['sensor.camera.depth', cc.Depth, 'Camera Depth (Gray Scale)', {}],
['sensor.camera.depth', cc.LogarithmicDepth, 'Camera Depth (Logarithmic Gray Scale)', {}],
['sensor.camera.semantic_segmentation', cc.Raw, 'Camera Semantic Segmentation (Raw)', {}],
['sensor.camera.semantic_segmentation', cc.CityScapesPalette,
'Camera Semantic Segmentation (CityScapes Palette)', {}],
['sensor.lidar.ray_cast', None, 'Lidar (Ray-Cast)', {'range': '50'}],
['sensor.camera.dvs', cc.Raw, 'Dynamic Vision Sensor', {}],
['sensor.camera.rgb', cc.Raw, 'Camera RGB Distorted',
{'lens_circle_multiplier': '3.0',
'lens_circle_falloff': '3.0',
'chromatic_aberration_intensity': '0.5',
'chromatic_aberration_offset': '0'}]]
world = self._parent.get_world()
bp_library = world.get_blueprint_library()
for item in self.sensors:
bp = bp_library.find(item[0])
if item[0].startswith('sensor.camera'):
bp.set_attribute('image_size_x', str(hud.dim[0]))
bp.set_attribute('image_size_y', str(hud.dim[1]))
if bp.has_attribute('gamma'):
bp.set_attribute('gamma', str(gamma_correction))
for attr_name, attr_value in item[3].items():
bp.set_attribute(attr_name, attr_value)
elif item[0].startswith('sensor.lidar'):
self.lidar_range = 50
for attr_name, attr_value in item[3].items():
bp.set_attribute(attr_name, attr_value)
if attr_name == 'range':
self.lidar_range = float(attr_value)
item.append(bp)
self.index = None
def toggle_camera(self):
self.transform_index = (self.transform_index + 1) % len(self._camera_transforms)
self.set_sensor(self.index, notify=False, force_respawn=True)
def set_sensor(self, index, notify=True, force_respawn=False):
index = index % len(self.sensors)
needs_respawn = True if self.index is None else \
(force_respawn or (self.sensors[index][2] != self.sensors[self.index][2]))
if needs_respawn:
if self.sensor is not None:
self.sensor.destroy()
self.surface = None
self.sensor = self._parent.get_world().spawn_actor(
self.sensors[index][-1],
self._camera_transforms[self.transform_index][0],
attach_to=self._parent,
attachment_type=self._camera_transforms[self.transform_index][1])
# We need to pass the lambda a weak reference to self to avoid
# circular reference.
weak_self = weakref.ref(self)
self.sensor.listen(lambda image: CameraManager._parse_image(weak_self, image))
if notify:
self.hud.notification(self.sensors[index][2])
self.index = index
def next_sensor(self):
self.set_sensor(self.index + 1)
def toggle_recording(self):
self.recording = not self.recording
self.hud.notification('Recording %s' % ('On' if self.recording else 'Off'))
def render(self, display):
if self.surface is not None:
display.blit(self.surface, (0, 0))
@staticmethod
def _parse_image(weak_self, image):
self = weak_self()
if not self:
return
if self.sensors[self.index][0].startswith('sensor.lidar'):
points = np.frombuffer(image.raw_data, dtype=np.dtype('f4'))
points = np.reshape(points, (int(points.shape[0] / 3), 3))
lidar_data = np.array(points[:, :2])
lidar_data *= min(self.hud.dim) / (2.0 * self.lidar_range)
lidar_data += (0.5 * self.hud.dim[0], 0.5 * self.hud.dim[1])
lidar_data = np.fabs(lidar_data) # pylint: disable=E1111
lidar_data = lidar_data.astype(np.int32)
lidar_data = np.reshape(lidar_data, (-1, 2))
lidar_img_size = (self.hud.dim[0], self.hud.dim[1], 3)
lidar_img = np.zeros((lidar_img_size), dtype=np.uint8)
lidar_img[tuple(lidar_data.T)] = (255, 255, 255)
self.surface = pygame.surfarray.make_surface(lidar_img)
elif self.sensors[self.index][0].startswith('sensor.camera.dvs'):
# Example of converting the raw_data from a carla.DVSEventArray
# sensor into a NumPy array and using it as an image
dvs_events = np.frombuffer(image.raw_data, dtype=np.dtype([
('x', np.uint16), ('y', np.uint16), ('t', np.int64), ('pol', np.bool)]))
dvs_img = np.zeros((image.height, image.width, 3), dtype=np.uint8)
# Blue is positive, red is negative
dvs_img[dvs_events[:]['y'], dvs_events[:]['x'], dvs_events[:]['pol'] * 2] = 255
self.surface = pygame.surfarray.make_surface(dvs_img.swapaxes(0, 1))
else:
image.convert(self.sensors[self.index][1])
array = np.frombuffer(image.raw_data, dtype=np.dtype("uint8"))
array = np.reshape(array, (image.height, image.width, 4))
array = array[:, :, :3]
array = array[:, :, ::-1]
self.surface = pygame.surfarray.make_surface(array.swapaxes(0, 1))
if self.recording:
image.save_to_disk('_out/%08d' % image.frame)
class World(object):
def __init__(self, carla_world, hud, args):
self.world = carla_world
self.actor_role_name = args.rolename
try:
self.map = self.world.get_map()
except RuntimeError as error:
print('RuntimeError: {}'.format(error))
print(' The server could not send the OpenDRIVE (.xodr) file:')
print(' Make sure it exists, has the same name of your town, and is correct.')
sys.exit(1)
self.hud = hud
self.player = None
self.collision_sensor = None
self.lane_invasion_sensor = None
self.gnss_sensor = None
self.imu_sensor = None
self.radar_sensor = None
self.camera_manager = None
self._weather_presets = find_weather_presets()
self._weather_index = 0
self._actor_filter = args.filter
self._gamma = args.gamma
self.restart()
self.world.on_tick(hud.on_world_tick)
self.recording_enabled = False
self.recording_start = 0
def restart(self):
self.player_max_speed = 1.589
self.player_max_speed_fast = 3.713
# Keep same camera config if the camera manager exists.
cam_index = self.camera_manager.index if self.camera_manager is not None else 0
cam_pos_index = self.camera_manager.transform_index if self.camera_manager is not None else 0
# Get a random blueprint.
blueprint = random.choice(self.world.get_blueprint_library().filter(self._actor_filter))
blueprint.set_attribute('role_name', self.actor_role_name)
if blueprint.has_attribute('color'):
color = random.choice(blueprint.get_attribute('color').recommended_values)
blueprint.set_attribute('color', color)
if blueprint.has_attribute('driver_id'):
driver_id = random.choice(blueprint.get_attribute('driver_id').recommended_values)
blueprint.set_attribute('driver_id', driver_id)
if blueprint.has_attribute('is_invincible'):
blueprint.set_attribute('is_invincible', 'true')
# set the max speed
if blueprint.has_attribute('speed'):
self.player_max_speed = float(blueprint.get_attribute('speed').recommended_values[1])
self.player_max_speed_fast = float(blueprint.get_attribute('speed').recommended_values[2])
else:
print("No recommended values for 'speed' attribute")
# Spawn the player.
if self.player is not None:
spawn_point = self.player.get_transform()
spawn_point.location.z += 2.0
spawn_point.rotation.roll = 0.0
spawn_point.rotation.pitch = 0.0
self.destroy()
self.player = self.world.try_spawn_actor(blueprint, spawn_point)
while self.player is None:
if not self.map.get_spawn_points():
print('There are no spawn points available in your map/town.')
print('Please add some Vehicle Spawn Point to your UE4 scene.')
sys.exit(1)
spawn_points = self.map.get_spawn_points()
spawn_point = random.choice(spawn_points) if spawn_points else carla.Transform()
self.player = self.world.try_spawn_actor(blueprint, spawn_point)
# Set up the sensors.
self.collision_sensor = CollisionSensor(self.player, self.hud)
self.lane_invasion_sensor = LaneInvasionSensor(self.player, self.hud)
self.gnss_sensor = GnssSensor(self.player)
self.imu_sensor = IMUSensor(self.player)
self.camera_manager = CameraManager(self.player, self.hud, self._gamma)
self.camera_manager.transform_index = cam_pos_index
self.camera_manager.set_sensor(cam_index, notify=False)
actor_type = get_actor_display_name(self.player)
self.hud.notification(actor_type)
def next_weather(self, reverse=False):
self._weather_index += -1 if reverse else 1
self._weather_index %= len(self._weather_presets)
preset = self._weather_presets[self._weather_index]
self.hud.notification('Weather: %s' % preset[1])
self.player.get_world().set_weather(preset[0])
def toggle_radar(self):
if self.radar_sensor is None:
self.radar_sensor = RadarSensor(self.player)
elif self.radar_sensor.sensor is not None:
self.radar_sensor.sensor.destroy()
self.radar_sensor = None
def tick(self, clock):
self.hud.tick(self, clock)
def render(self, display):
self.camera_manager.render(display)
self.hud.render(display)
def destroy_sensors(self):
self.camera_manager.sensor.destroy()
self.camera_manager.sensor = None
self.camera_manager.index = None
def destroy(self):
if self.radar_sensor is not None:
self.toggle_radar()
actors = [
self.camera_manager.sensor,
self.collision_sensor.sensor,
self.lane_invasion_sensor.sensor,
self.gnss_sensor.sensor,
self.imu_sensor.sensor,
self.player]
for actor in actors:
if actor is not None:
actor.destroy()
class HUD(object):
def __init__(self, width, height):
self.dim = (width, height)
font = pygame.font.Font(pygame.font.get_default_font(), 20)
font_name = 'courier' if os.name == 'nt' else 'mono'
fonts = [x for x in pygame.font.get_fonts() if font_name in x]
default_font = 'ubuntumono'
mono = default_font if default_font in fonts else fonts[0]
mono = pygame.font.match_font(mono)
self._font_mono = pygame.font.Font(mono, 12 if os.name == 'nt' else 14)
self._notifications = FadingText(font, (width, 40), (0, height - 40))
self.help = HelpText(pygame.font.Font(mono, 16), width, height)
self.server_fps = 0
self.frame = 0
self.simulation_time = 0
self._show_info = True
self._info_text = []
self._server_clock = pygame.time.Clock()
def on_world_tick(self, timestamp):
self._server_clock.tick()
self.server_fps = self._server_clock.get_fps()
self.frame = timestamp.frame
self.simulation_time = timestamp.elapsed_seconds
def tick(self, world, clock):
self._notifications.tick(world, clock)
if not self._show_info:
return
t = world.player.get_transform()
v = world.player.get_velocity()
c = world.player.get_control()
compass = world.imu_sensor.compass
heading = 'N' if compass > 270.5 or compass < 89.5 else ''
heading += 'S' if 90.5 < compass < 269.5 else ''
heading += 'E' if 0.5 < compass < 179.5 else ''
heading += 'W' if 180.5 < compass < 359.5 else ''
colhist = world.collision_sensor.get_collision_history()
collision = [colhist[x + self.frame - 200] for x in range(0, 200)]
max_col = max(1.0, max(collision))
collision = [x / max_col for x in collision]
vehicles = world.world.get_actors().filter('vehicle.*')
self._info_text = [
'Server: % 16.0f FPS' % self.server_fps,
'Client: % 16.0f FPS' % clock.get_fps(),
'',
'Vehicle: % 20s' % get_actor_display_name(world.player, truncate=20),
'Map: % 20s' % world.map.name,
'Simulation time: % 12s' % datetime.timedelta(seconds=int(self.simulation_time)),
'',
'Speed: % 15.0f km/h' % (3.6 * math.sqrt(v.x**2 + v.y**2 + v.z**2)),
u'Compass:% 17.0f\N{DEGREE SIGN} % 2s' % (compass, heading),
'Accelero: (%5.1f,%5.1f,%5.1f)' % (world.imu_sensor.accelerometer),
'Gyroscop: (%5.1f,%5.1f,%5.1f)' % (world.imu_sensor.gyroscope),
'Location:% 20s' % ('(% 5.1f, % 5.1f)' % (t.location.x, t.location.y)),
'GNSS:% 24s' % ('(% 2.6f, % 3.6f)' % (world.gnss_sensor.lat, world.gnss_sensor.lon)),
'Height: % 18.0f m' % t.location.z,
'']
if isinstance(c, carla.VehicleControl):
self._info_text += [
('Throttle:', c.throttle, 0.0, 1.0),
('Steer:', c.steer, -1.0, 1.0),
('Brake:', c.brake, 0.0, 1.0),
('Reverse:', c.reverse),
('Hand brake:', c.hand_brake),
('Manual:', c.manual_gear_shift),
'Gear: %s' % {-1: 'R', 0: 'N'}.get(c.gear, c.gear)]
elif isinstance(c, carla.WalkerControl):
self._info_text += [
('Speed:', c.speed, 0.0, 5.556),
('Jump:', c.jump)]
self._info_text += [
'',
'Collision:',
collision,
'',
'Number of vehicles: % 8d' % len(vehicles)]
if len(vehicles) > 1:
self._info_text += ['Nearby vehicles:']
distance = lambda l: math.sqrt((l.x - t.location.x)**2 + (l.y - t.location.y)**2 + (l.z - t.location.z)**2)
vehicles = [(distance(x.get_location()), x) for x in vehicles if x.id != world.player.id]
for d, vehicle in sorted(vehicles):
if d > 200.0:
break
vehicle_type = get_actor_display_name(vehicle, truncate=22)
self._info_text.append('% 4dm %s' % (d, vehicle_type))
def toggle_info(self):
self._show_info = not self._show_info
def notification(self, text, seconds=2.0):
self._notifications.set_text(text, seconds=seconds)
def error(self, text):
self._notifications.set_text('Error: %s' % text, (255, 0, 0))
def render(self, display):
if self._show_info:
info_surface = pygame.Surface((220, self.dim[1]))
info_surface.set_alpha(100)
display.blit(info_surface, (0, 0))
v_offset = 4
bar_h_offset = 100
bar_width = 106
for item in self._info_text:
if v_offset + 18 > self.dim[1]:
break
if isinstance(item, list):
if len(item) > 1:
points = [(x + 8, v_offset + 8 + (1.0 - y) * 30) for x, y in enumerate(item)]
pygame.draw.lines(display, (255, 136, 0), False, points, 2)
item = None
v_offset += 18
elif isinstance(item, tuple):
if isinstance(item[1], bool):
rect = pygame.Rect((bar_h_offset, v_offset + 8), (6, 6))
pygame.draw.rect(display, (255, 255, 255), rect, 0 if item[1] else 1)
else:
rect_border = pygame.Rect((bar_h_offset, v_offset + 8), (bar_width, 6))
pygame.draw.rect(display, (255, 255, 255), rect_border, 1)
f = (item[1] - item[2]) / (item[3] - item[2])
if item[2] < 0.0:
rect = pygame.Rect((bar_h_offset + f * (bar_width - 6), v_offset + 8), (6, 6))
else:
rect = pygame.Rect((bar_h_offset, v_offset + 8), (f * bar_width, 6))
pygame.draw.rect(display, (255, 255, 255), rect)
item = item[0]
if item: # At this point has to be a str.
surface = self._font_mono.render(item, True, (255, 255, 255))
display.blit(surface, (8, v_offset))
v_offset += 18
self._notifications.render(display)
self.help.render(display)
#Define architecture for Deep Q-Network
class DQNetwork():
def __init__(self, state_size, action_size, learning_rate, name='DQNetwork'):
self.state_size = state_size
self.action_size = action_size
self.learning_rate = learning_rate
self.possible_actions = np.identity(self.action_size, dtype=int).tolist()
with tf.variable_scope(name):
#inputs define image fed into NN
self.inputs_ = tf.placeholder(tf.float32, [None, *state_size], name="inputs")
#actions define array containing tuple of actions taken by system
self.actions_ = tf.placeholder(tf.float32, [None, self.action_size], name="actions_")
self.target_Q = tf.placeholder(tf.float32, [None], name="target")
self.conv1 = tf.layers.conv2d(inputs=self.inputs_,
filters=32,
kernel_size=[8,8],
strides=[4,4],
padding="VALID",
kernel_initializer=tf.contrib.layers.xavier_initializer_conv2d(),
name="conv1")
self.conv1_out = tf.nn.elu(self.conv1, name="conv1_out")
self.conv2 = tf.layers.conv2d(inputs=self.conv1_out,
filters=64,
kernel_size=[4,4],
strides=[2,2],
padding="VALID",
kernel_initializer=tf.contrib.layers.xavier_initializer_conv2d(),
name="conv2")
self.conv2_out = tf.nn.elu(self.conv2, name="conv2_out")
self.conv3 = tf.layers.conv2d(inputs=self.conv2_out,
filters=64,
kernel_size=[3,3],
strides=[2,2],
padding="VALID",
kernel_initializer=tf.contrib.layers.xavier_initializer_conv2d(),
name="conv3")
#After multiple convolutions, use exponential linear unit
#Activation function since DQN predicts continuous set of q-vals
self.conv3_out = tf.nn.elu(self.conv3, name="conv3_out")
self.flatten = tf.contrib.layers.flatten(self.conv3_out)
self.fc = tf.layers.dense(inputs=self.flatten,
units=512,
activation=tf.nn.elu,
kernel_initializer=tf.contrib.layers.xavier_initializer(),
name="fc1")
self.output = tf.layers.dense(inputs=self.fc,
kernel_initializer=tf.contrib.layers.xavier_initializer(),
units=self.action_size,
activation=None)
self.Q = tf.reduce_sum(tf.multiply(self.output, self.actions_)) #predicted Q-value computed by DNN by associating output of DNN w/ action tuples
self.loss = tf.reduce_mean(tf.square(self.target_Q - self.Q)) #compute loss per each action-val
#return gradients for each weight of NN (change in weights after minimizing loss)
self.optimizer = tf.train.AdamOptimizer(self.learning_rate).minimize(self.loss)
def predict_action(self, sess, explore_start, explore_stop, decay_rate, decay_step, state):
# Epsilon greedy strategy: given state s, choose action a ep. greedy
exp_tradeoff = np.random.rand()
explore_probability = explore_stop + (explore_start - explore_stop) * np.exp(-decay_rate * decay_step)
if (explore_probability > exp_tradeoff):
action_int = np.random.choice(self.action_size)
action = self.possible_actions[action_int]
else:
# get action from Q-network: neural network estimates the Q values
Qs = sess.run(self.output, feed_dict={self.inputs_: state.reshape((1, *state.shape))})
# choose the best Q value from the discrete action space (argmax)
action_int = np.argmax(Qs)
action = self.possible_actions[int(action_int)]
return action_int, action, explore_probability
def map_from_control(control, action_space):
'''maps continuous control to discrete action values
used to convert control from autopilot to discrete values that the Q-network is using
It works by computing the discrete action with the smaller euclidian distance from the
continuous actions'''
control_vector = np.array([control.throttle, control.steer, control.brake])
distances = [] # euclidian distance list
for control in action_space:
distances.append(np.linalg.norm(control-control_vector)) # compute euclidian distance
return np.argmin(distances)
#Use experience replay to efficiently train agent on random samples, reduce correlation
class Memory():
def __init__(self, max_size, pretrain_length, action_space):
self.buffer = deque(maxlen = max_size)
self.pretrain_length = pretrain_length
self.action_space = action_space
self.action_size = len(action_space)
self.possible_actions = np.identity(self.action_size, dtype=int).tolist()
def add(self, experience):
self.buffer.append(experience)
def sample(self, batch_size):
buffer_size = len(self.buffer)
index = np.random.choice(np.arange(buffer_size),
size = batch_size,
replace = True)
return [self.buffer[i] for i in index]
def fill_memory(self, map, vehicle, camera_queue, sensors, autopilot = False):
print("Started to fill memory")
reset_environment(map, vehicle, sensors)
vehicle.set_autopilot()
for i in range(self.pretrain_length):
if i % 10 == 0:
print(i, "experiences stored")
state = process_image(camera_queue)
control = vehicle.get_control()
action_int = map_from_control(control, self.action_space)
action = self.possible_actions[action_int]
time.sleep(0.25)
reward = compute_reward(vehicle, sensors)
done = isDone(reward)
next_state = process_image(camera_queue)
experience = state, action, reward, next_state, done
self.add(experience)
if done:
reset_environment(map, vehicle, sensors)
else:
state = next_state
print('Finished filing memory. %s experiences stored.' % self.pretrain_length)
vehicle.set_autopilot(enabled = False)
def save_memory(self, filename, object):
handle = open(filename, "wb")
pickle.dump(object, handle)
def load_memory(self, filename):
with open(filename, 'rb') as f:
return pickle.load(f)
#carla sensors special actors to measure & stream data using listen() method
#can retrieve data upon timestep/action; must be attached to parent actor (vehicle)
#listen() method employs lambda func which recursively callback
class Sensors(object):
"""Class to keep track of all sensors added to the vehicle"""
def __init__(self, world, vehicle):
super(Sensors, self).__init__()
self.world = world
self.vehicle = vehicle
self.camera_queue = queue.Queue() # queue to store images from buffer
self.collision_flag = False # Flag for colision detection
self.lane_crossed = False # Flag for lane crossing detection
self.lane_crossed_type = '' # Which type of lane was crossed
self.camera_rgb = self.add_sensors(world, vehicle, 'sensor.camera.rgb')
self.collision = self.add_sensors(world, vehicle, 'sensor.other.collision')
self.lane_invasion = self.add_sensors(world, vehicle, 'sensor.other.lane_invasion', sensor_tick = '0.5')
self.sensor_list = [self.camera_rgb, self.collision, self.lane_invasion]
#sensor uses lambda func to constantly retrieve data and return it
#secondary func that uses lambda func will then manipulate data for use
#such as by setting collision flag on
self.collision.listen(lambda collisionEvent: self.track_collision(collisionEvent))
self.camera_rgb.listen(lambda image: self.camera_queue.put(image))
self.lane_invasion.listen(lambda event: self.on_invasion(event))
def add_sensors(self, world, vehicle, type, sensor_tick = '0.0'):
sensor_bp = self.world.get_blueprint_library().find(type)
try:
sensor_bp.set_attribute('sensor_tick', sensor_tick)
except:
pass
if type == 'sensor.camera.rgb':
sensor_bp.set_attribute('image_size_x', '100')
sensor_bp.set_attribute('image_size_y', '100')
sensor_transform = carla.Transform(carla.Location(x=1.5, z=2.4))
sensor = self.world.spawn_actor(sensor_bp, sensor_transform, attach_to=vehicle)
return sensor
def track_collision(self, collisionEvent):
'''Whenever a collision occurs, the flag is set to True'''
self.collision_flag = True
def reset_sensors(self):
'''Sets all sensor flags to False'''
self.collision_flag = False
self.lane_crossed = False
self.lane_crossed_type = ''
def on_invasion(self, event):
'''Whenever the car crosses the lane, the flag is set to True'''
lane_types = set(x.type for x in event.crossed_lane_markings)
text = ['%r' % str(x).split()[-1] for x in lane_types]
self.lane_crossed_type = text[0]
self.lane_crossed = True
def destroy_sensors(self):
'''Destroy all sensors (Carla actors)'''
for sensor in self.sensor_list:
sensor.destroy()
def reset_environment(map, vehicle, sensors):
vehicle.apply_control(carla.VehicleControl(throttle=0.0, brake=1.0))
time.sleep(1)
spawn_points = map.get_spawn_points()
spawn_point = random.choice(spawn_points) if spawn_points else carla.Transform()
vehicle.set_transform(spawn_point)
time.sleep(2)
sensors.reset_sensors()
def process_image(queue):
'''get the image from the buffer and process it. It's the state for vision-based systems'''
image = queue.get()
array = np.frombuffer(image.raw_data, dtype=np.dtype("uint8"))
array = np.reshape(array, (image.height, image.width, 4))
array = array[:, :, :3]
array = array[:, :, ::-1]
image = Image.fromarray(array).convert('L') # grayscale conversion
image = np.array(image.resize((84, 84))) # convert to numpy array
image = np.reshape(image, (84, 84, 1)) # reshape image
return image
def map_action(action, action_space):
""" maps discrete actions into actual values to control the car"""
control = carla.VehicleControl()
control_sequence = action_space[action]
control.throttle = control_sequence[0]
control.steer = control_sequence[1]
control.brake = control_sequence[2]
return control
def compute_reward(vehicle, sensors):
max_speed = 14
min_speed = 2
speed = vehicle.get_velocity()
vehicle_speed = np.linalg.norm([speed.x, speed.y, speed.z])
speed_reward = (abs(vehicle_speed) - min_speed) / (max_speed - min_speed)
lane_reward = 0
if (vehicle_speed > max_speed) or (vehicle_speed < min_speed):
speed_reward = -0.05
if sensors.lane_crossed:
if sensors.lane_crossed_type == "'Broken'" or sensors.lane_crossed_type == "'NONE'":
lane_reward = -0.5
sensors.lane_crossed = False
if sensors.collision_flag:
return -1
else:
return speed_reward + lane_reward