-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathenvironments.py
486 lines (383 loc) · 17.6 KB
/
environments.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
import cv2
import importlib
import os
import rospy
import signal
import sys
import numpy as np
from collections import deque
from gymnasium import Env
from gymnasium.spaces import Space, Box
from numpy.typing import NDArray
from . import PointOfViewTwins, DictToClass
from .utils import container_ip, airsim_launch, json_content
class Stack(Space):
def __init__(self, obs_type : str, stack : int, pre_aug : tuple):
self.__obs_type = obs_type
self.__shapes = {
"rgb" : (3*stack, *pre_aug),
"depth" : (3*stack, *pre_aug),
"segmentation" : (3*stack, *pre_aug),
"point_cloud" : (1024*stack, 3),
"tf" : (7*stack,)
}
self.__obs_space = {
"rgb": Box(low = 0, high = 255, shape= self.__shapes["rgb"], dtype=np.uint8),
"depth": Box(low = -2**63, high = 2**63, shape=self.__shapes["depth"], dtype=np.float32),
"segmentation": Box(low = 0, high = 255, shape=self.__shapes["segmentation"], dtype=np.uint8),
"point_cloud": Box(low = -2**63, high = 2**63, shape=self.__shapes["point_cloud"], dtype=np.float64),
"tf": Box(low = -2**63, high = 2**63, shape=self.__shapes["tf"], dtype=np.float64)
}
self.__stack = self._stack(stack)
# Space.__init__(self, shape=[*((img_shape, ) * (len(self.__obs_space.keys()) -1)),tf_shape])
# shapes = {key: shapes[key] for key in self.__obs_space.keys()}
# Space.__init__(self, shape=[*shapes.values()])
Space.__init__(self)
@property
def shape(self):
return {key: self.__shapes[key] for key in self.__obs_space.keys()}
@property
def stack(self):
stack = dict()
for k, obs in self.__stack.items():
# print(f"STACK_SHAPE : {len(obs)}")
stack[k] = np.concatenate(list(obs), axis=0)
# print(f"STACK_SHAPE : {len(stack[k])}")
# return stack.values()
return stack
@stack.setter
def stack(self, obs : dict):
for k, v in obs.items():
self.__stack[k].append(v)
if len(self.__stack[k]) == 1:
self.__stack[k] = self.__stack[k]*3
# rospy.logwarn(f"ENTREI NO SETTER --> stack {self.__stack}")
def _obs_space(self):
stereo = set({'segmentation', 'point_cloud'})
stereo_occupancy = set({'segmentation'})
panoptic = set({'point_cloud'})
leftover = None
if self.__obs_type.endswith('stereo'):
leftover = list(set(self.__obs_space) - stereo)
elif self.__obs_type.endswith('stereo_occupancy'):
leftover = list(set(self.__obs_space) - stereo_occupancy)
elif self.__obs_type.endswith('panoptic'):
leftover = list(set(self.__obs_space) - panoptic)
if not leftover:
obs = dict()
obs[self.__obs_type] = self.__obs_space[self.__obs_type]
obs['tf'] = self.__obs_space['tf']
self.__obs_space = obs
return self.__obs_space
obs = dict()
for k in leftover:
obs[k] = self.__obs_space[k]
self.__obs_space = obs
return self.__obs_space
def _stack(self, stack : int):
dstack = dict()
for k in self._obs_space():
dstack[k] = deque([], maxlen=stack)
return dstack
class Simulation:
action_range = DictToClass({
"x" : [-60, 60],
"y" : [-60, 60],
"z" : [-60, 60],
"yaw" : [-45, 45],
"gimbal_pitch" : [-45, 45]
})
vehicle = DictToClass({
"start_pose" : [0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
"base" : {
"name" : "boat_vehicle",
"altitude" : 3.0
}
})
twin = DictToClass({
"start_pose" : [0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
"base" : {
"name" : "boat_twin",
"altitude" : 3.0
}
})
centroide = "centroide"
def __init__(self, observation_type : str, ue4 : str, markers_name : DictToClass):
self.__ip = container_ip(ue4)
self.__airsim = None #airsim_launch(self.__ip)
self._parse_settings()
self.__twins = PointOfViewTwins(self.ip, self.vehicle, self.twin, observation_type)
self.__twins.set_detection(markers_name)
@property
def client(self):
return self.__twins
@property
def airsim(self):
return self.__airsim
@property
def ip(self):
return self.__ip
def _parse_settings(self):
settings_path = os.path.abspath(__file__).replace(__file__.split('/')[-1], 'settings/settings.json')
settings = dict(json_content(settings_path))
svehicle_names = list(settings['Vehicles'].keys())
svehicle = settings['Vehicles'][svehicle_names[0]]
svehicle_name = svehicle_names[0]
gx, gy, gz, groll, gpitch, gyaw = svehicle['X'], svehicle['Y'], svehicle['Z'], \
svehicle['Roll'], svehicle['Pitch'], svehicle['Yaw']
svehicle_camera_name = list(svehicle['Cameras'].keys())[0]
svehicle_camera_dim = svehicle['Cameras'][svehicle_camera_name]['CaptureSettings'][0]['Width'],\
svehicle['Cameras'][svehicle_camera_name]['CaptureSettings'][0]['Height']
svehicle_camera_fov = svehicle['Cameras'][svehicle_camera_name]['CaptureSettings'][0]["FOV_Degrees"]
stwin = settings['Vehicles'][svehicle_names[1]]
stwin_name = svehicle_names[1]
gsx, gsy, gsz, gsroll, gspitch, gsyaw = stwin['X'], stwin['Y'], stwin['Z'], \
stwin['Roll'], stwin['Pitch'], stwin['Yaw']
stwin_camera_name = list(stwin['Cameras'].keys())[0]
stwin_camera_dim = stwin['Cameras'][stwin_camera_name]['CaptureSettings'][0]['Width'],\
stwin['Cameras'][stwin_camera_name]['CaptureSettings'][0]['Height']
stwin_camera_fov = stwin['Cameras'][stwin_camera_name]['CaptureSettings'][0]["FOV_Degrees"]
v = {'name' : svehicle_name,
'global_pose' : [gx, gy, gz, groll, gpitch, gyaw],
'camera' : {
'name' : svehicle_camera_name,
'dim' : svehicle_camera_dim,
'fov' : svehicle_camera_fov
}
}
self.vehicle.update(v)
t = {'name' : stwin_name,
'global_pose' : [gsx, gsy, gsz, gsroll, gspitch, gsyaw],
'camera' : {
'name' : stwin_camera_name,
'dim' : stwin_camera_dim,
'fov' : stwin_camera_fov
}
}
# print(t)
self.twin.update(t)
import copy
class GymPointOfView(Simulation, Env):
task_name = "point-of-view"
max_episode_steps = 200
@staticmethod
def set_markers(markers_name, markers_num : int):
if isinstance(markers_name, str):
m = [ markers_name ]
m = m + [f"{markers_name}{i}" for i in range(1, markers_num+1)]
return set(m)
am = [ markers_name[0] ]
am = am + [f"{ markers_name[0]}{i}" for i in range(1, markers_num[0]+1)]
um = [ markers_name[1] ]
um = um + [f"{ markers_name[1]}{i}" for i in range(1, markers_num[1]+1)]
return set(am+um)
def __init__(self, observation_type : str,
observation_stack : int,
pre_aug : tuple,
domain : str,
ue4 : str,
markers : DictToClass,
target_range : DictToClass):
Simulation.__init__(self, observation_type, ue4, markers.name)
Env.__init__(self)
# rospy.init_node(f"gym-{self.task_name}")
self.observation_space = Stack(observation_type, observation_stack, pre_aug)
self.action_space = Box(low=-1, high=1, shape=(5,), dtype=np.float32)
self.current_step = 0
self.original_markers_len = markers.num
self.markers_len = markers.num
self.past_markers_len = markers.num
self.markers_backup = self.set_markers(markers.name, markers.num)
self.markers_need_to_visit = copy.deepcopy(self.markers_backup)
self.range_to_get_target = markers.range_to_get
self.target_range = target_range
self.domain = domain
self.heliport = True if domain.endswith('air') else False
self.__pre_aug = pre_aug
def _normalize_action(self, action):
def _normalize_value(x, min_val, max_val, a, b):
return ((x - min_val) / (max_val - min_val)) * (b - a) + a
x, y, z, yaw_angle, gimbal_pitch_angle = action
xmin, xmax = self.action_range.x
ymin, ymax = self.action_range.y
zmin, zmax = self.action_range.z
yaw_min, yaw_max =self.action_range.yaw
gimbal_pitch_min, gimbal_pitch_max = self.action_range.gimbal_pitch
px = _normalize_value(x, -1, 1, xmin, xmax)
py = _normalize_value(y, -1, 1, ymin, ymax)
pz = _normalize_value(-1 * z, -1, 1, zmin, zmax)
yaw = _normalize_value(yaw_angle, -1, 1, yaw_min, yaw_max)
gimbal_pitch = _normalize_value(gimbal_pitch_angle, -1, 1, gimbal_pitch_min, gimbal_pitch_max)
return px, py, pz, yaw, gimbal_pitch
def _get_obs(self):
def _pre_aug_obs_shape(obs : NDArray, dim : tuple, type= 'image'):
if type.endswith('depth'):
img_ = obs.copy()
nan_location = np.isnan(img_)
img_[nan_location] = np.nanmax(img_)
norm_image = (img_)*255./5.
norm_image[0,0] = 255.
norm_image = norm_image.astype('uint8')
norm_image = cv2.cvtColor(norm_image, cv2.COLOR_GRAY2BGR)
return cv2.resize(norm_image.copy(), dim, interpolation = cv2.INTER_AREA).transpose(2, 0, 1)
if type.endswith('point_cloud'):
centroid = np.mean(obs, axis=0)
xy_projection = obs[:, :2]
center_index = np.argmin(np.linalg.norm(xy_projection - centroid[:2], axis=1))
half = 512
obs = obs[center_index - half : center_index + half, :]
print(obs.shape, center_index - half, center_index + half)
return obs
return cv2.resize(obs.copy(), dim, interpolation = cv2.INTER_AREA).transpose(2, 0, 1)
def _parse_obs(obs : dict):
_obs = dict()
for k, v in obs.items():
if k.endswith('tf'):
_obs[k] = v
elif k.endswith('point_cloud'):
# max_points = self.__pre_aug[0] * self.__pre_aug[1]
# _obs[k] = np.pad(v, ((0, max_points - v.shape[0]), (0, 0)), mode='constant', constant_values=0)
_obs[k] = _pre_aug_obs_shape(v, self.__pre_aug, 'point_cloud')
elif k.endswith('depth'):
_obs[k] = _pre_aug_obs_shape(v, self.__pre_aug, 'depth')
else:
_obs[k] = _pre_aug_obs_shape(v, self.__pre_aug)
return _obs
observation = _parse_obs(self.client.observation)
self.observation_space.stack = observation
return self.observation_space.stack
def _get_info(self):
info = f"Current markers lenght : {self.markers_len}"
return {'info' : info}
def _get_state(self):
viewd_markers, distances = self.client.detections()
# print(f'vm {viewd_markers}')
d = self.client.detection_distance()
done = False
reset_pose = False
distance = 0
self.markers_len = self.past_markers_len
if distances:
self.markers_need_to_visit -= set(viewd_markers)
self.markers_len = len(self.markers_need_to_visit)
dmin, dmax = self.range_to_get_target
if (abs(d - min(distances))) < dmin or abs(max(distances) - d) > dmax:
distance = 1
reset_pose = True
if not self.markers_len or (self.original_markers_len - self.markers_len) >= .97*self.original_markers_len :
done = True
else:
reset_pose = True
return self._get_obs(), self.markers_len, distance, reset_pose, done, self._get_info()
def _reward(self, markers_len, distance, reset_pose, done):
# print(f"original : {self.original_markers_len} {markers_len}")
# print(f"CURRENT_STEP : {self.current_step} --- MAX_STEPS : {self.max_episode_steps}")
alpha = 1e-2
beta = 1e-3 * self.original_markers_len
gamma = 2 * beta
delta = 1e-1
if done:
return delta * self.original_markers_len, done
if self.current_step == self.max_episode_steps -1 :
return 0, True
if reset_pose:
self.client.go_home()
if distance:
# print(f'medium : {-10 * alpha}')
return -alpha, done
# print(f'minimum : {-20 * alpha}')
return -beta, done
if markers_len == self.past_markers_len:
return 0, done
self.past_markers_len = markers_len
# print(f"bigger : {(self.original_markers_len - markers_len) * beta}")
return alpha * (self.original_markers_len - markers_len), done
def reset_random(self):
pose = self.client.random_pose(self.action_range.x, self.action_range.y,
self.target_range.x, self.target_range.y,
self.centroide, self.heliport)
self.client.home = pose
self.client.go_home()
def reset(self, seed= None, options= None):
seed = np.random.seed()
super().reset(seed=seed)
self.current_step = 0
self.client.go_home()
self.markers_need_to_visit = copy.deepcopy(self.markers_backup)
return self._get_obs(), self._get_info()
def step(self, action):
norm_action = self._normalize_action(action)
self.client.next_point_of_view(norm_action)
observation, len_markers, distance, reset_pose, done, info = self._get_state()
reward, done = self._reward(len_markers, distance, reset_pose, done)
self.current_step += 1
return observation, reward, done, info
def close(self):
os.killpg(self.airsim, signal.SIGINT)
sys.exit()
class AirPointOfView:
domain = 'air'
markers = DictToClass({'name' : 'Cube'})
altitude = -1.0
class UnderwaterPointOfView:
domain = 'underwater'
markers = DictToClass({'name' : 'Sphere'})
altitude = [-30, -6]
class HybridPointOfView:
domain = 'hybrid'
markers = DictToClass({'name' : ('Cube', 'Sphere')})
altitude = -30
class BasicAirPOV(AirPointOfView, GymPointOfView):
markers = AirPointOfView.markers.update({
"num" : 79,
"range_to_get" : [2, 80]
})
target_range = DictToClass({
"x" : [2, 80],
"y" : [2, 80]
})
def __init__(self, observation_type : str,
observation_stack : int,
pre_aug : tuple,
max_ep_steps : int,
ue4 : str) -> None:
AirPointOfView.__init__(self)
GymPointOfView.__init__(self, observation_type, observation_stack, pre_aug,
self.domain, ue4, self.markers, self.target_range)
self.max_episode_steps = max_ep_steps
class BasicUnderwaterPOV(UnderwaterPointOfView, GymPointOfView):
markers = UnderwaterPointOfView.markers.update({
"num" : 73,
"range_to_get" : [2, 160]
})
target_range = DictToClass({
"x" : [62, 100],
"y" : [62, 100]
})
def __init__(self, observation_type : str,
observation_stack : int,
pre_aug : tuple,
max_ep_steps : int,
ue4 : str) -> None:
UnderwaterPointOfView.__init__(self)
GymPointOfView.__init__(self, observation_type, observation_stack, pre_aug,
self.domain, ue4, self.markers, self.target_range)
self.max_episode_steps = max_ep_steps
class BasicHybridPOV(HybridPointOfView, GymPointOfView):
markers = HybridPointOfView.markers.update({
"num" : (79, 73),
"range_to_get" : [2, 160]
})
target_range = DictToClass({
"x" : [62, 100],
"y" : [62, 100]
})
def __init__(self, observation_type : str,
observation_stack : int,
pre_aug : tuple,
max_ep_steps : int,
ue4 : str) -> None:
HybridPointOfView.__init__(self)
GymPointOfView.__init__(self, observation_type, observation_stack, pre_aug,
self.domain, ue4, self.markers, self.target_range)
self.max_episode_steps = max_ep_steps