/
TestPhysics.py
420 lines (329 loc) · 22 KB
/
TestPhysics.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
from PrimitivesPhysics import PrimitivesPhysics
from CapsuleNetwork import CapsuleNetwork
from Capsule import Capsule
from Observation import Observation
from RelationTriplet import RelationTriplet
from HyperParameters import HyperParameters
from AttributePool import AttributePool
import numpy as np
import math
import random
import scipy.misc
class TestPhysics(PrimitivesPhysics):
def init(self):
self._xPosOffset = self._attributePool.getAttributeOrderByName("Position-X")
self._yPosOffset = self._attributePool.getAttributeOrderByName("Position-Y")
self._sizeOffset = self._attributePool.getAttributeOrderByName("Size")
self._rotOffset = self._attributePool.getAttributeOrderByName("Rotation")
self._arOffset = self._attributePool.getAttributeOrderByName("Aspect-Ratio")
self._intOffset = self._attributePool.getAttributeOrderByName("Intensity")
self._strOffset = self._attributePool.getAttributeOrderByName("Strength")
def generateInteractionSequence(self, capsNet : CapsuleNetwork, width : int, height : int, folder : str, idname : str):
# Generate Images in the folder with name id + "." + sequence_index + file_format
# 0 = No Interaction
# 1 = Newtonian Collision
interType = random.randint(0, 1)
# 0 = Image Before
# 1 = Image at Interaction
# 2 = Image After
positionA = [None, None, None]
positionB = [None, None, None]
positionA[1] = np.array([0.5, 0.5]) #np.array([random.random(), random.random()])
# TODO: Assuming Circles for now
massA = random.random() * 0.2 + 0.1
massB = random.random() * 0.2 + 0.1
intA = random.random()
intB = random.random()
strA = min(random.random(), (0.333333 - massA) * 10.0 )
strB = min(random.random(), (0.333333 - massB) * 10.0 )
rotA = random.random()
rotB = random.random()
awayDir = np.array([random.random() - 0.5, random.random() - 0.5])
awayDir = awayDir / np.linalg.norm(awayDir)
velMod = 0.5
if 1 == 1: #interType == 10:
# No Interaction
maxDist = random.random()
awayVec = awayDir * ((massA + massB + (strA + strB) * 0.1 ) * 0.5 + maxDist + 0.02)
positionB[1] = positionA[1] + awayVec
# Velocities
velA = np.array([random.random() - 0.5, random.random() - 0.5]) * min(maxDist, velMod * random.random())
velB = np.array([random.random() - 0.5, random.random() - 0.5]) * min(maxDist, velMod * random.random())
positionA[0] = positionA[1] - velA
positionA[2] = positionA[1] + velA
positionB[0] = positionB[1] - velB
positionB[2] = positionB[1] + velB
elif interType == 1:
# Interaction
awayVec = awayDir * ((massA + massB + (strA + strB) * 0.1 ) * 0.5)
positionB[1] = positionA[1] + awayVec
# Velocities
velA = np.array([random.random() - 0.5, random.random() - 0.5]) * velMod * random.random()
velB = np.array([random.random() - 0.5, random.random() - 0.5]) * velMod * random.random()
if np.dot(velA, velB) < 0:
# Flying away from each other -> Reverse one Velocity
velA = -velA
if (np.dot(velA, awayDir) < 0 and np.dot(velB, awayDir) < 0 and np.linalg.norm(velB) < np.linalg.norm(velA)) or \
(np.dot(velA, awayDir) > 0 and np.dot(velB, awayDir) > 0 and np.linalg.norm(velA) < np.linalg.norm(velB)):
# A Flying away from B and B flying towards A (or B Flying away from A and A flying towards B)
# Only collide if B (A) is faster than A (B), thus we switch velocities
velTemp = velA
velA = velB
velB = velTemp
positionA[0] = positionA[1] - velA
positionB[0] = positionB[1] - velB
tempB = np.dot((velB - velA), awayVec) / (math.pow(np.linalg.norm(awayVec), 2.0))
resultVelB = velB - (2 * massA / (massA + massB)) * tempB * awayVec
tempA = np.dot((velA - velB), -awayVec) / (math.pow(np.linalg.norm(awayVec), 2.0))
resultVelA = velA - (2 * massB / (massA + massB)) * tempA * (-awayVec)
positionA[2] = positionA[1] + resultVelA
positionB[2] = positionB[1] + resultVelB
attributesA = [None, None, None]
attributesB = [None, None, None]
for i in range(3):
attributesA[i] = np.zeros(HyperParameters.MaximumAttributeCount)
attributesB[i] = np.zeros(HyperParameters.MaximumAttributeCount)
attributesA[i][self._xPosOffset] = positionA[i][0]
attributesA[i][self._yPosOffset] = positionA[i][1]
attributesA[i][self._sizeOffset] = massA
attributesA[i][self._intOffset] = intA
attributesA[i][self._strOffset] = strA
attributesA[i][self._rotOffset] = rotA
attributesA[i][self._arOffset] = 1.0
attributesB[i][self._xPosOffset] = positionB[i][0]
attributesB[i][self._yPosOffset] = positionB[i][1]
attributesB[i][self._sizeOffset] = massB
attributesB[i][self._intOffset] = intB
attributesB[i][self._strOffset] = strB
attributesB[i][self._rotOffset] = rotB
attributesB[i][self._arOffset] = 1.0
# Render Images and Save
circCaps = capsNet.getCapsuleByName("TestPrimitives.Circle")
for i in range(3):
attrDictA = {}
for j in range(len(attributesA[i])):
attrDictA[circCaps.getAttributeByName(self._attributePool.getAttributeNameByOrder(j))] = attributesA[i][j]
attrDictB = {}
for j in range(len(attributesB[i])):
attrDictB[circCaps.getAttributeByName(self._attributePool.getAttributeNameByOrder(j))] = attributesB[i][j]
observationA = Observation(circCaps, circCaps._routes[0], [], attrDictA, 1.0)
observationB = Observation(circCaps, circCaps._routes[0], [], attrDictB, 1.0)
obs = {circCaps : [observationA, observationB]}
imageReal, ignore1, ignore2 = capsNet.generateImage(width, height, obs)
pixels = [0.0] * (width * height * 3)
for yy in range(height):
for xx in range(width):
pixels[(yy * width + xx) * 3] = imageReal[(yy * width + xx) * 4]
pixels[(yy * width + xx) * 3 + 1] = imageReal[(yy * width + xx) * 4]
pixels[(yy * width + xx) * 3 + 2] = imageReal[(yy * width + xx) * 4]
scipy.misc.imsave(folder + idname + "." + str(i) + ".png", np.reshape(pixels, [height, width, 3]))
return
def generateRelation(self):
# Triplet Format:
# Sender -- Symbol | Attributes | Velocities | Static/Dynamic | Rigid/Elastic
# Receiver -- Symbol | Attributes | Velocities | Static/Dynamic | Rigid/Elastic
# Relation -- Distance | Degrees-Of-Freedom | Sender Normal | Receiver Normal
# Effect Format:
# Acceleration Vector | Angle Acceleration Vector
triplet = [0.0] * RelationTriplet.tripletLength()
effect = [0.0] * HyperParameters.DegreesOfFreedom
######### TRIPLET
totalObjectEntries = (HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2 * HyperParameters.DegreesOfFreedom)
if random.randint(0, 100) % 2 == 0:
triplet[2 * totalObjectEntries] = random.random() * HyperParameters.DistanceCutoff
hasCollision = True
else:
triplet[2 * totalObjectEntries] = random.random() + HyperParameters.DistanceCutoff
hasCollision = False
massSizeA = random.random()
massSizeB = random.random()
positionA = np.array([random.random(), random.random()])
differenceVector = np.array([random.random() - 0.5, random.random() - 0.5])
differenceVector = differenceVector / np.linalg.norm(differenceVector)
distanceVector = differenceVector * (triplet[2 * totalObjectEntries] + (massSizeA + massSizeB) * 0.5)
positionB = np.add(positionA, distanceVector)
# Slightly Off-Center Interactions
offRotA = random.random() * 2 * math.pi
offRotB = random.random() * 2 * math.pi
velocityA = random.random() * np.array([differenceVector[0] * math.cos(offRotA) - differenceVector[1] * math.sin(offRotA), differenceVector[0] * math.cos(offRotA) + differenceVector[1] * math.sin(offRotA)])
velocityB = random.random() * np.array([-differenceVector[0] * math.cos(offRotB) + differenceVector[1] * math.sin(offRotB), -differenceVector[0] * math.cos(offRotB) - differenceVector[1] * math.sin(offRotB)])
vMagA = random.random()
vMagB = random.random()
# Filling Sender Attributes
for i in range(HyperParameters.MaximumAttributeCount):
triplet[HyperParameters.MaximumSymbolCount + i] = random.random()
triplet[HyperParameters.MaximumSymbolCount + self._xPosOffset] = positionA[0]
triplet[HyperParameters.MaximumSymbolCount + self._yPosOffset] = positionA[1]
triplet[HyperParameters.MaximumSymbolCount + self._sizeOffset] = massSizeA
# Filling Receiver Attributes
for i in range(HyperParameters.MaximumAttributeCount):
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + i] = random.random()
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + self._xPosOffset] = positionB[0]
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + self._yPosOffset] = positionB[1]
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + self._sizeOffset] = massSizeB
# Filling Sender Velocity
for i in range(HyperParameters.MaximumAttributeCount):
triplet[HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount + i] = random.random()
triplet[HyperParameters.MaximumSymbolCount + HyperParameters.MaximumAttributeCount + self._xPosOffset] = (velocityA[0] + 1.0) / 2.0
triplet[HyperParameters.MaximumSymbolCount + HyperParameters.MaximumAttributeCount + self._yPosOffset] = (velocityA[1] + 1.0) / 2.0
triplet[HyperParameters.MaximumSymbolCount + HyperParameters.MaximumAttributeCount + self._rotOffset] = vMagA
# Filling Receiver Velocity
for i in range(HyperParameters.MaximumAttributeCount):
triplet[totalObjectEntries + HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount + i] = random.random()
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + HyperParameters.MaximumAttributeCount + self._xPosOffset] = (velocityB[0] + 1.0) / 2.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + HyperParameters.MaximumAttributeCount + self._yPosOffset] = (velocityB[1] + 1.0) / 2.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + HyperParameters.MaximumAttributeCount + self._rotOffset] = vMagB
# Train for all Symbols
senderSymbol = random.randint(0, HyperParameters.MaximumSymbolCount)
receiverSymbol = random.randint(0, HyperParameters.MaximumSymbolCount)
triplet[senderSymbol] = 1.0
triplet[totalObjectEntries + receiverSymbol] = 1.0
# Static / Dynamic
# For testing purposes, we also train the case, where the receiver can only rotate
windmill = random.randint(0, 100)
if windmill >= 75:
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount] = 1.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 1] = 1.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount] = 0.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 1] = 0.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2] = 1.0
triplet[totalObjectEntries + receiverSymbol] = 0.0
# Figure 8
triplet[totalObjectEntries + 3] = 1.0
elif windmill >= 50:
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount] = 0.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 1] = 0.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 1] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2] = 1.0
triplet[senderSymbol] = 0.0
# Figure 8
triplet[3] = 1.0
else:
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount] = 1.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 1] = 1.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 1] = 1.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 2] = 1.0
# Rigid / Elastic
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 3] = 0.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 4] = 0.0
triplet[HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 5] = 0.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 3] = 0.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 4] = 0.0
triplet[totalObjectEntries + HyperParameters.MaximumSymbolCount + 2 * HyperParameters.MaximumAttributeCount + 5] = 0.0
# Degrees-Of-Freedom
triplet[2 * totalObjectEntries + 1] = 1.0
triplet[2 * totalObjectEntries + 2] = 1.0
triplet[2 * totalObjectEntries + 3] = 1.0
# Sender Normal
triplet[2 * totalObjectEntries + 4] = (differenceVector[0] + 1.0) / 2.0
triplet[2 * totalObjectEntries + 5] = (differenceVector[1] + 1.0) / 2.0
# Receiver Normal
triplet[2 * totalObjectEntries + 6] = (-differenceVector[0] + 1.0) / 2.0
triplet[2 * totalObjectEntries + 7] = (-differenceVector[1] + 1.0) / 2.0
######### EFFECT
# We take the effects to be the sum of a Force F over time delta-t, i.e. Impuls I = F * delta-t
# Even though balls are close, they are going in different directions
if windmill >= 75:
# Ball - Windmill interaction.
# This is not real physics, just something that simulates plausible looking interactions. If we knew the real physics anyways,
# we could just implement that.
# Fake linear momentums from angular momentums.
vMagB = ((vMagB * 2.0) - 1.0) / (massSizeB * 0.5)
velocityB = vMagB * np.array([-differenceVector[1], differenceVector[0]])
massSizeA = massSizeA * massSizeB / 2.0
massSizeB = massSizeB * massSizeB / 6.0
elif windmill >= 50:
vMagA = ((vMagA * 2.0) - 1.0) / (massSizeA * 0.5)
velocityA = vMagA * np.array([-differenceVector[1], differenceVector[0]])
massSizeB = massSizeA * massSizeB / 2.0
massSizeA = massSizeA * massSizeA / 6.0
if not (np.dot(velocityA, differenceVector) < 0 and (np.dot(velocityB, differenceVector) > 0 or np.linalg.norm(velocityB) < np.linalg.norm(velocityA))) and \
not (np.dot(velocityB, -differenceVector) < 0 and (np.dot(velocityA, -differenceVector) > 0 or np.linalg.norm(velocityA) < np.linalg.norm(velocityB))) and \
hasCollision == True:
tempB = np.dot((velocityB - velocityA), distanceVector) / (math.pow(np.linalg.norm(distanceVector), 2.0))
resultVelocityB = velocityB - (2 * massSizeA / (massSizeA + massSizeB)) * tempB * distanceVector
resultAccelB = (resultVelocityB - velocityB) / HyperParameters.TimeStep
if windmill >= 75:
angDir = np.array([-differenceVector[1], differenceVector[0]])
effect[0] = 0.5
effect[1] = 0.5
effect[2] = ((np.linalg.norm(resultAccelB * angDir) / HyperParameters.AccelerationScale) + 1.0) / 2.0
else:
# Scaling the Force Vectors
effect[0] = ((resultAccelB[0] / HyperParameters.AccelerationScale) + 1.0) / 2.0
effect[1] = ((resultAccelB[1] / HyperParameters.AccelerationScale) + 1.0) / 2.0
effect[2] = 0.5
else:
effect[0] = 0.5
effect[1] = 0.5
effect[2] = 0.5
return triplet, effect
def generateInteraction(self):
# Aggregate Format:
# Receiver -- Attributes | Symbol | Velocities | Static/Dynamic | Rigid/Elastic
# Effects -- Summed Effect Acceleration Vector | Summed Effect Angle Acceleration Vector
# External -- External Acceleration Vector | External Angle Acceleration Vector
# Attributes Format:
# Receiver -- Attributes | Accelerations
aggregate = [0.0] * (HyperParameters.MaximumAttributeCount * 2 + HyperParameters.MaximumSymbolCount + HyperParameters.DegreesOfFreedom * 4)
attributes = [0.0] * HyperParameters.MaximumAttributeCount * 2
######### AGGREGATE
# Attributes
for i in range(HyperParameters.MaximumAttributeCount):
aggregate[i] = random.random()
# Symbol
aggregate[HyperParameters.MaximumAttributeCount + random.randint(0, HyperParameters.MaximumSymbolCount)] = 1.0
# Velocities
offset = HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount
for i in range(HyperParameters.MaximumAttributeCount):
aggregate[offset + i] = random.random()
# Static/Dynamic // Rigid/Elastic
offset = offset + HyperParameters.MaximumAttributeCount
aggregate[offset] = float(random.randint(0, 1))
aggregate[offset + 1] = float(random.randint(0, 1))
aggregate[offset + 2] = float(random.randint(0, 1))
aggregate[offset + 3] = 0.0
aggregate[offset + 4] = 0.0
aggregate[offset + 5] = 0.0
# Effects
offset = offset + 2 * HyperParameters.DegreesOfFreedom
aggregate[offset] = random.random()
aggregate[offset + 1] = random.random()
aggregate[offset + 2] = random.random()
# External
aggregate[offset + 3] = random.random()
aggregate[offset + 4] = random.random()
aggregate[offset + 5] = random.random()
totalAccel = (np.array([aggregate[offset], aggregate[offset + 1], aggregate[offset + 2]]) * 2.0 - 1.0) * HyperParameters.AccelerationScale
totalAccel = totalAccel + (np.array([aggregate[offset + 3], aggregate[offset + 4], aggregate[offset + 5]]) * 2.0 - 1.0) * HyperParameters.AccelerationScale
######### RECEIVER
offsetV = HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount
for i in range(HyperParameters.MaximumAttributeCount):
# Attributes:
attributes[i] = aggregate[i] + ((aggregate[offsetV + i] * 2.0) - 1.0) * HyperParameters.TimeStep
# For non-Positions
# Accelerations:
attributes[HyperParameters.MaximumAttributeCount + i] = 0.5
# Only apply force acceleration to position and rotation
attributes[self._xPosOffset] = attributes[self._xPosOffset] + 0.5 * totalAccel[0] * HyperParameters.TimeStep * HyperParameters.TimeStep
attributes[self._yPosOffset] = attributes[self._yPosOffset] + 0.5 * totalAccel[1] * HyperParameters.TimeStep * HyperParameters.TimeStep
attributes[self._rotOffset] = attributes[self._rotOffset] + 0.5 * totalAccel[2] * HyperParameters.TimeStep * HyperParameters.TimeStep
attributes[HyperParameters.MaximumAttributeCount + self._xPosOffset] = ((totalAccel[0] / HyperParameters.AccelerationScale) + 1.0) * 0.5
attributes[HyperParameters.MaximumAttributeCount + self._yPosOffset] = ((totalAccel[1] / HyperParameters.AccelerationScale) + 1.0) * 0.5
attributes[HyperParameters.MaximumAttributeCount + self._rotOffset] = ((totalAccel[2] / HyperParameters.AccelerationScale) + 1.0) * 0.5
# Static.. Undo changes
if aggregate[2 * HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount] < 0.1:
attributes[self._xPosOffset] = aggregate[self._xPosOffset]
attributes[HyperParameters.MaximumAttributeCount + self._xPosOffset] = 0.5
if aggregate[2 * HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount + 1] < 0.1:
attributes[self._yPosOffset] = aggregate[self._yPosOffset]
attributes[HyperParameters.MaximumAttributeCount + self._yPosOffset] = 0.5
if aggregate[2 * HyperParameters.MaximumAttributeCount + HyperParameters.MaximumSymbolCount + 2] < 0.1:
attributes[self._rotOffset] = aggregate[self._rotOffset]
attributes[HyperParameters.MaximumAttributeCount + self._rotOffset] = 0.5
return aggregate, attributes