-
Notifications
You must be signed in to change notification settings - Fork 120
/
estimator.cpp
1481 lines (1253 loc) · 44 KB
/
estimator.cpp
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
#include <algorithm>
#include <iomanip>
#include <iostream>
#include <fstream>
#include <tuple>
#include "Eigen/QR"
#include "glog/logging.h"
#include "estimator.h"
#include "feature.h"
#include "group.h"
#include "jac.h"
#include "mm.h"
#include "param.h"
#include "tracker.h"
#include "helpers.h"
#include "mapper.h"
#ifdef USE_G2O
#include "optimizer.h"
#endif
namespace xivo {
std::unique_ptr<Estimator> Estimator::instance_{nullptr};
EstimatorPtr Estimator::Create(const Json::Value &cfg) {
if (instance_) {
LOG(WARNING) << "Estimator already exists!";
} else {
instance_ = std::unique_ptr<Estimator>(new Estimator{cfg});
}
return instance_.get();
}
EstimatorPtr Estimator::instance() {
if (!instance_) {
LOG(FATAL) << "Estimator NOT created yet!";
}
return instance_.get();
}
static const Mat3 I3{Mat3::Identity()};
static const Mat3 nI3{-I3};
static const Mat2 I2{Mat2::Identity()};
static const Mat2 nI2{-I2};
static bool cmp(const std::unique_ptr<internal::Message> &m1,
const std::unique_ptr<internal::Message> &m2) {
return m1->ts() > m2->ts();
}
namespace internal {
void Inertial::Execute(Estimator *est) {
est->InertialMeasInternal(ts_, gyro_, accel_);
}
void Visual::Execute(Estimator *est) { est->VisualMeasInternal(ts_, img_); }
void VisualTrackerOnly::Execute(Estimator *est) { est->VisualMeasInternalTrackerOnly(ts_, img_); }
void VisualPointCloud::Execute(Estimator *est) {
est->VisualMeasPointCloudInternal(ts_, feature_ids_, xp_and_depths_);
}
void VisualPointCloudTrackerOnly::Execute(Estimator *est) {
est->VisualMeasPointCloudInternalTrackerOnly(ts_, feature_ids_, xp_and_depths_);
}
} // namespace internal
// destructor
Estimator::~Estimator() {
if (cfg_.get("print_calibration", false).asBool()) {
std::cout << "===== Auto-Calibration =====\n";
std::cout << "Rbc=\n" << X_.Rbc.matrix() << std::endl;
auto Wbc = X_.Rbc.log();
std::cout << "Wbc=" << Wbc.transpose() << std::endl;
std::cout << "Tbc=" << X_.Tbc.transpose() << std::endl;
std::cout << "td=" << X_.td << std::endl;
std::cout << "gyro.bias=" << X_.bg.transpose() << std::endl;
std::cout << "accel.bias=" << X_.ba.transpose() << std::endl;
std::cout << "Rsg=" << X_.Rsg.matrix() << std::endl;
auto Wsg = X_.Rsg.log();
std::cout << "Wsg=" << Wsg.transpose() << std::endl;
std::cout << "===== IMU intrinsics =====\n";
std::cout << "Ca=\n" << imu_.Ca() << std::endl;
std::cout << "Cg=\n" << imu_.Cg() << std::endl;
std::cout << "===== Camera intrinsics =====\n";
CameraManager::instance()->Print(std::cout);
}
if (worker_) {
worker_->join();
delete worker_;
}
}
Estimator::Estimator(const Json::Value &cfg)
: cfg_{cfg}, gauge_group_{-1}, worker_{nullptr}, timer_{"estimator"},
gauge_group_ptr_{nullptr} {
// /////////////////////////////
// Component flags
// /////////////////////////////
simulation_ = cfg_.get("simulation", false).asBool();
use_canvas_ = cfg_.get("use_canvas", true).asBool();
print_timing_ = cfg_.get("print_timing", false).asBool();
integration_method_ =
cfg_.get("integration_method", "unspecified").asString();
// OOS update options
use_OOS_ = cfg_.get("use_OOS", false).asBool();
use_compression_ = cfg_.get("use_compression", false).asBool();
compression_trigger_ratio_ =
cfg_.get("compression_trigger_ratio", 1.5).asDouble();
OOS_update_min_observations_ =
cfg_.get("OOS_update_min_observations", 5).asInt();
if (use_OOS_) {
LOG(FATAL) << "MSCKF not implemented";
}
// IMU clamping
Vec3 _vec_;
clamp_signals_ = cfg_.get("clamp_signals", false).asBool();
max_accel_ = GetVectorFromJson<number_t, 3>(cfg_, "max_accel");
max_gyro_ = GetVectorFromJson<number_t, 3>(cfg_, "max_gyro");
// one point ransac parameters
use_1pt_RANSAC_ = cfg_.get("use_1pt_RANSAC", false).asBool();
ransac_thresh_ = cfg_.get("1pt_RANSAC_thresh", 5).asDouble();
ransac_prob_ = cfg_.get("1pt_RANSAC_prob", 0.95).asDouble();
ransac_Chi2_ = cfg_.get("1pt_RANSAC_Chi2", 5.89).asDouble();
// depth-initialization subfilter options
number_t tri_std = cfg_["subfilter"].get("visual_meas_std", 3.5).asDouble();
subfilter_options_.Rtri = tri_std * tri_std;
subfilter_options_.MH_thresh =
cfg_["subfilter"].get("MH_thresh", 5.991).asDouble();
subfilter_options_.ready_steps =
cfg_["subfilter"].get("ready_steps", 5).asInt();
// depth optimization options
use_depth_opt_ = cfg_.get("use_depth_opt", false).asBool();
refinement_options_.two_view =
cfg_["depth_opt"].get("two_view", false).asBool();
refinement_options_.use_hessian = cfg_["depth_opt"].get("use_hessian", false).asBool();
refinement_options_.max_iters = cfg_["depth_opt"].get("max_iters", 5).asInt();
refinement_options_.eps = cfg_["depth_opt"].get("eps", 1e-4).asDouble();
refinement_options_.damping =
cfg_["depth_opt"].get("damping", 1e-3).asDouble();
refinement_options_.max_res_norm =
cfg_["depth_opt"].get("max_res_norm", 2.0).asDouble();
refinement_options_.Rtri = subfilter_options_.Rtri;
triangulate_pre_subfilter_ =
cfg_.get("triangulate_pre_subfilter", false).asBool();
triangulate_options_.method = cfg_["triangulation"].get("method", "l1_angular").asString();
triangulate_options_.zmin =
cfg_["triangulation"].get("zmin", 0.05).asDouble();
triangulate_options_.zmax = cfg_["triangulation"].get("zmax", 5.0).asDouble();
triangulate_options_.max_theta_thresh = cfg_["triangulation"].get("max_theta_thresh", 0.1).asDouble() * M_PI / 180;
triangulate_options_.beta_thesh = cfg_["triangulation"].get("beta_thesh", 0.25).asDouble() * M_PI / 180;
adaptive_initial_depth_options_.median_weight =
cfg_["adaptive_initial_depth"].get("median_weight", 0.99).asDouble();
adaptive_initial_depth_options_.min_feature_lifetime =
cfg_["adaptive_initial_depth"].get("minimum_feature_lifetime", 5).asInt();
remove_outlier_counter_ = cfg_.get("remove_outlier_counter", 10).asInt();
group_degrees_fixed_ = cfg_.get("group_degrees_fixed", 4).asInt();
if ((group_degrees_fixed_ != 4) && (group_degrees_fixed_ != 6)) {
LOG(FATAL) << "group_degrees_fixed must be 4 or 6";
}
// load imu calibration
auto imu_calib = cfg_["imu_calib"];
// load accel axis misalignment first as a 3x3 matrix
Mat3 Ta =
GetMatrixFromJson<number_t, 3, 3>(imu_calib, "Car", JsonMatLayout::RowMajor);
Mat3 Ka; // accel scaling
Ka.diagonal() = GetVectorFromJson<number_t, 3>(imu_calib, "Cas");
Mat3 Ca{Ta * Ka};
// load gyro axis misalignment first as 3x3 matrix
Mat3 Tg =
GetMatrixFromJson<number_t, 3, 3>(imu_calib, "Cgr", JsonMatLayout::RowMajor);
Mat3 Kg; // gyro scaling
Kg.diagonal() = GetVectorFromJson<number_t, 3>(imu_calib, "Cgs");
Mat3 Cg{Tg * Kg};
// now update the IMU component
imu_ = IMU{Ca, Cg};
LOG(INFO) << "Imu calibration loaded";
g_ = GetMatrixFromJson<number_t, 3, 1>(cfg_, "gravity");
LOG(INFO) << "gravity loaded:" << g_.transpose();
// /////////////////////////////
// Initialize motion state
// /////////////////////////////
auto X = cfg_["X"];
try {
X_.Rsb = SO3::exp(GetVectorFromJson<number_t, 3>(X, "Wsb"));
} catch (const Json::LogicError &e) {
Mat3 Rsb_tmp = GetMatrixFromJson<number_t, 3, 3>(X, "Wsb", JsonMatLayout::RowMajor);
if (Sophus::isOrthogonal(Rsb_tmp) && (Rsb_tmp.determinant() > 0.0)) {
X_.Rsb = SO3(Rsb_tmp);
} else {
LOG(WARNING) << "Input value of Rsb is not orthogonal or its determinant is negative. Projecting to SO(3) group";
X_.Rsb = SO3::fitToSO3(Rsb_tmp);
}
}
X_.Tsb = GetVectorFromJson<number_t, 3>(X, "Tsb");
X_.Vsb = GetVectorFromJson<number_t, 3>(X, "Vsb");
X_.bg = GetVectorFromJson<number_t, 3>(X, "bg");
X_.ba = GetVectorFromJson<number_t, 3>(X, "ba");
if (cfg_.get("imu_tk_convention", false).asBool()) {
// For biases obtained by IMU-TK library,
// the calibrated meaurement is a_calib = K(a_raw + a_bias)
// whereas in our model a_calib=K * a_raw - a_bias
// thus we need convert that.
X_.bg = -imu_.Cg() * X_.bg;
X_.ba = -imu_.Ca() * X_.ba;
}
try {
X_.Rbc = SO3::exp(GetVectorFromJson<number_t, 3>(X, "Wbc"));
} catch (const Json::LogicError &e) {
Mat3 Rbc_tmp = GetMatrixFromJson<number_t, 3, 3>(X, "Wbc", JsonMatLayout::RowMajor);
if (Sophus::isOrthogonal(Rbc_tmp) && (Rbc_tmp.determinant() > 0.0)) {
X_.Rbc = SO3(Rbc_tmp);
} else {
LOG(WARNING) << "Input value of Rbc is not orthogonal or its determinant is negative. Projecting to SO(3) group";
X_.Rbc = SO3::fitToSO3(Rbc_tmp);
}
}
X_.Tbc = GetVectorFromJson<number_t, 3>(X, "Tbc");
Vec3 Wsg;
Wsg.head<2>() = GetVectorFromJson<number_t, 2>(X, "Wsg");
X_.Rsg = SO3::exp(Wsg);
// temporal offset
#ifdef USE_ONLINE_TEMPORAL_CALIB
X_.td = X["td"].asDouble();
#endif
// initialize error state
err_.resize(kFullSize);
err_.setZero();
// make all group & feature slots available
std::fill(gsel_.begin(), gsel_.end(), false);
std::fill(fsel_.begin(), fsel_.end(), false);
LOG(INFO) << "Initial state loaded";
LOG(INFO) << X_;
auto P = cfg_["P"];
P_.setIdentity(kFullSize, kFullSize);
P_.block<3, 3>(Index::Wsb, Index::Wsb) *= P["Wsb"].asDouble();
P_.block<3, 3>(Index::Tsb, Index::Tsb) *= P["Tsb"].asDouble();
P_.block<3, 3>(Index::Vsb, Index::Vsb) *= P["Vsb"].asDouble();
P_.block<3, 3>(Index::bg, Index::bg) *= P["bg"].asDouble();
P_.block<3, 3>(Index::ba, Index::ba) *= P["ba"].asDouble();
P_.block<3, 3>(Index::Wbc, Index::Wbc) *= P["Wbc"].asDouble();
try {
P_.block<3, 3>(Index::Tbc, Index::Tbc) *= P["Tbc"].asDouble();
} catch (const std::exception&) {
auto Cov = GetVectorFromJson<number_t, 3>(P, "Tbc");
P_.block<3, 3>(Index::Tbc, Index::Tbc) *= Cov.asDiagonal();
}
P_.block<2, 2>(Index::Wsg, Index::Wsg) *= P["Wsg"].asDouble();
#ifdef USE_ONLINE_TEMPORAL_CALIB
P_(Index::td, Index::td) *= P["td"].asDouble();
#endif
#ifdef USE_ONLINE_IMU_CALIB
// online IMU calibration
P_.block<9, 9>(Index::Cg, Index::Cg) *= P["Cg"].asDouble();
P_.block<6, 6>(Index::Ca, Index::Ca) *= P["Ca"].asDouble();
#endif
// online camera intrinsics calibration
// initialize covariance for camera intrinsics
#ifdef USE_ONLINE_CAMERA_CALIB
int dim = Camera::instance()->dim();
try {
// homogeneous focal length and principal point error
P_.block(kCameraBegin, kCameraBegin, 4, 4) *= P["FC"].asDouble();
} catch (const std::exception &) {
// non-homogeneous focal length and principal point error
auto fc_var = GetVectorFromJson<number_t, 2>(P, "FC");
P_.block(kCameraBegin, kCameraBegin, 2, 2) *= fc_var[0];
P_.block(kCameraBegin+2, kCameraBegin+2, 2, 2) *= fc_var[1];
}
P_.block(kCameraBegin + 4, kCameraBegin + 4, dim - 4, dim - 4) *=
P["distortion"].asDouble();
P_.block(kCameraBegin + dim, kCameraBegin + dim, kMaxCameraIntrinsics - dim,
kMaxCameraIntrinsics - dim) *= 0;
#endif
// standard deviation -> covariance
// P_.block<kMotionSize, kMotionSize>(0, 0) *=
// P_.block<kMotionSize, kMotionSize>(0, 0);
P_ *= P_;
LOG(INFO) << "Initial covariance loaded";
// allocate spaces for Jacobians
F_.resize(kMotionSize, kMotionSize);
F_.setIdentity();
G_.resize(kMotionSize, 12);
G_.setZero();
auto Qmodel = cfg_["Qmodel"];
Qmodel_.setZero(kMotionSize, kMotionSize);
Qmodel_.block<3, 3>(Index::Wsb, Index::Wsb) = I3 * Qmodel["Wsb"].asDouble();
Qmodel_.block<3, 3>(Index::Wbc, Index::Wbc) = I3 * Qmodel["Wbc"].asDouble();
Qmodel_.block<2, 2>(Index::Wsg, Index::Wsg) = I2 * Qmodel["Wsg"].asDouble();
Qmodel_.block<kMotionSize, kMotionSize>(0, 0) *=
Qmodel_.block<kMotionSize, kMotionSize>(0, 0);
LOG(INFO) << "Covariance of process noises loaded";
// /////////////////////////////
// Initialize measurement noise
// /////////////////////////////
auto Qimu = cfg_["Qimu"];
Qimu_.setIdentity(12, 12);
Qimu_.block<3, 3>(0, 0) *= GetVectorFromJson<number_t, 3>(Qimu, "gyro").asDiagonal();
Qimu_.block<3, 3>(3, 3) *= GetVectorFromJson<number_t, 3>(Qimu, "accel").asDiagonal();
Qimu_.block<3, 3>(6, 6) *= GetVectorFromJson<number_t, 3>(Qimu, "gyro_bias").asDiagonal();
Qimu_.block<3, 3>(9, 9) *= GetVectorFromJson<number_t, 3>(Qimu, "accel_bias").asDiagonal();
Qimu_ *= Qimu_;
LOG(INFO) << "Covariance of IMU measurement noise loaded";
R_ = cfg_["visual_meas_std"].asDouble();
R_ *= R_;
Roos_ = cfg["oos_meas_std"].asDouble();
Roos_ *= Roos_;
Rlc_ = cfg["loop_closure_meas_std"].asDouble();
Rlc_ *= Rlc_;
LOG(INFO) << "R=" << R_ << "; Roos=" << Roos_ << "; Rlc=" << Rlc_;
// /////////////////////////////
// Load initial std on feature state
// /////////////////////////////
init_z_ = cfg_["initial_z"].asDouble();
init_std_x_ = cfg_["initial_std_x"].asDouble();
init_std_y_ = cfg_["initial_std_y"].asDouble();
init_std_x_ /= Camera::instance()->GetFocalLength();
init_std_y_ /= Camera::instance()->GetFocalLength();
init_std_z_ = cfg_["initial_std_z"].asDouble();
min_z_ = cfg_["min_depth"].asDouble();
max_z_ = cfg_["max_depth"].asDouble();
init_std_x_badtri_ = cfg_["initial_std_x_badtri"].asDouble();
init_std_y_badtri_ = cfg_["initial_std_y_badtri"].asDouble();
init_std_z_badtri_ = cfg_["initial_std_z_badtri"].asDouble();
LOG(INFO) << "Initial covariance for features loaded";
MeasurementUpdateInitialized_ = false;
// /////////////////////////////
// Outlier rejection options
// /////////////////////////////
use_MH_gating_ = cfg_.get("use_MH_gating", true).asBool();
min_required_inliers_ = cfg_.get("min_inliers", 5).asInt();
MH_thresh_ = cfg_.get("MH_thresh", 5.991).asDouble();
MH_thresh_multipler_ = cfg_.get("MH_adjust_factor", 1.1).asDouble();
// FIXME (xfei): used in HuberOnInnovation, but kinda overlaps with MH gating
outlier_thresh_ = cfg_.get("outlier_thresh", 1.1).asDouble();
feature_owner_change_cov_factor_ =
cfg_.get("filter_owner_change_cov_factor", 1.5).asDouble();
strict_criteria_timesteps_ = cfg_.get("strict_criteria_timesteps", 5).asInt();
// Feature Gauge Options
num_gauge_xy_features_ = cfg_.get("num_gauge_xy_features", 3).asInt();
number_t collinear_cross_prod_thresh =
cfg_.get("collinear_cross_prod_thresh", 1e-3).asDouble();
if ((num_gauge_xy_features_ < 0) || (num_gauge_xy_features_ > 3)) {
LOG(FATAL) << "Number of XY Gauge Features must be between 0 and 3";
}
// simulation options
sim_initialize_depths_ = false;
// reset initialization status
if (simulation_) {
gravity_init_counter_ = 0;
gravity_initialized_ = true;
} else {
gravity_init_counter_ = cfg_.get("gravity_init_counter", 20).asInt();
gravity_initialized_ = false;
}
vision_initialized_ = false;
// reset measurement counter
imu_counter_ = 0;
vision_counter_ = 0;
// reset various timestamps
last_imu_time_ = timestamp_t::zero();
curr_imu_time_ = timestamp_t::zero();
last_vision_time_ = timestamp_t::zero();
curr_vision_time_ = timestamp_t::zero();
last_time_ = timestamp_t::zero();
curr_time_ = timestamp_t::zero();
// random number generator
rng_ = std::unique_ptr<std::default_random_engine>(
new std::default_random_engine);
async_run_ = cfg_.get("async_run", false).asBool();
if (async_run_) {
Run();
}
}
void Estimator::Run() {
worker_ = new std::thread([this]() {
for (;;) {
std::unique_ptr<internal::Message> msg;
{
std::scoped_lock lck(buf_.mtx);
if (buf_.initialized && buf_.size() > InternalBuffer::MAX_SIZE) {
msg = std::move(buf_.front());
std::pop_heap(buf_.begin(), buf_.end(), cmp);
buf_.pop_back();
}
}
if (msg != nullptr) {
// std::cout << "executing\n";
msg->Execute(this);
}
}
});
}
bool Estimator::InitializeGravity() {
VLOG(0) << "attempt to initialize gravity";
if (!simulation_) {
if (gravity_init_buf_.size() < gravity_init_counter_) {
return false;
}
VLOG(0) << "initializing gravity";
// got enough stationary samples, estimate gravity
Vec3 mean_accel = std::accumulate(gravity_init_buf_.begin(),
gravity_init_buf_.end(), Vec3{0, 0, 0});
mean_accel /= gravity_init_buf_.size();
Vec3 accel_calib = imu_.Ca() * mean_accel - X_.ba;
// FromTwoVectors(a, b): returns R such that b=R*a
// we need R * accel + Rg * g_ == 0
// And R = Identity
// so accel = Rg * (-g_)
Eigen::AngleAxis<number_t> AAg(
Eigen::Quaternion<number_t>::FromTwoVectors(-g_, accel_calib));
Vec3 Wsg(AAg.axis() * AAg.angle());
Wsg(2) = 0;
X_.Rsg = SO3::exp(Wsg);
LOG(INFO) << "===== Wsg initialization =====";
LOG(INFO) << "stationary accel samples=" << gravity_init_buf_.size();
LOG(INFO) << "accel " << accel_calib.transpose();
LOG(INFO) << "Wsg=" << Wsg.transpose();
LOG(INFO) << "g=" << g_.transpose();
LOG(INFO) << "The norm below should be small";
LOG(INFO) << "|Rsb*a+Rg*g|=" << (X_.Rsb * accel_calib + X_.Rsg * g_).norm();
}
return true;
}
void Estimator::InertialMeasInternal(const timestamp_t &ts, const Vec3 &gyro,
const Vec3 &accel) {
if (!GoodTimestamp(ts))
return;
++imu_counter_;
Vec3 gyro_new;
Vec3 accel_new;
Vec3 grav_s = X_.Rsg * g_;
Vec3 grav_b = X_.Rsb.inverse() * grav_s;
if (clamp_signals_) {
Vec3 accel_wout_grav = accel + grav_b;
for (int i=0; i < 3; i++) {
number_t sign_gyro = (gyro(i) > 0) ? 1.0 : -1.0;
number_t sign_accel = (accel_wout_grav(i) > 0) ? 1.0 : -1.0;
number_t gyro_mag = (abs(gyro(i)) > max_gyro_(i)) ? max_gyro_(i) : abs(gyro(i));
number_t accel_mag = (abs(accel_wout_grav(i)) > max_accel_(i)) ? max_accel_(i) : abs(accel_wout_grav(i));
gyro_new(i) = sign_gyro * gyro_mag;
accel_new(i) = sign_accel * accel_mag;
}
accel_new -= grav_b;
} else{
gyro_new = gyro;
accel_new = accel;
}
// initialize imu -- basically gravity
if (!gravity_initialized_) {
gravity_init_buf_.emplace_back(accel_new);
if (InitializeGravity()) {
curr_imu_time_ = last_time_ = ts;
curr_accel_ = last_accel_ = accel_new;
curr_gyro_ = last_gyro_ = gyro_new;
gravity_initialized_ = true;
gravity_init_buf_.clear();
LOG(INFO) << "IMU initialized";
}
} else {
// process inertials only after vision module initialized
if (vision_initialized_) {
last_time_ = curr_time_;
curr_time_ = ts;
curr_accel_ = accel_new;
curr_gyro_ = gyro_new;
last_imu_time_ = curr_imu_time_;
curr_imu_time_ = ts;
Propagate(false);
}
}
}
void Estimator::Propagate(bool visual_meas) {
#ifndef NDEBUG
CHECK(gravity_initialized_)
<< "state progagation with un-initialized imu module";
#endif
timer_.Tick("propagation");
number_t dt;
Vec3 accel0, gyro0; // initial condition for integration
dt = std::chrono::duration<number_t>(curr_time_ - last_time_).count();
if (dt == 0) {
if (!simulation_) {
LOG(WARNING) << "measurement timestamps coincide?";
}
return;
}
if (!visual_meas) {
// this is an imu meas
slope_accel_ = (curr_accel_ - last_accel_) / dt;
slope_gyro_ = (curr_gyro_ - last_gyro_) / dt;
accel0 = last_accel_;
gyro0 = last_gyro_;
last_accel_ = curr_accel_;
last_gyro_ = curr_gyro_;
} else {
// this is a visual meas
accel0 = last_accel_;
gyro0 = last_gyro_;
last_accel_ = accel0 + slope_accel_ * dt;
last_gyro_ = gyro0 + slope_gyro_ * dt;
}
if (dt > 0.030) {
LOG(WARNING) << "dt=" << dt << " > 30 ms";
}
if (integration_method_ == "PrinceDormand") {
PrinceDormand(gyro0, accel0, dt);
} else if (integration_method_ == "Fehlberg") {
Fehlberg(gyro0, accel0, dt);
} else if (integration_method_ == "RK4") {
RK4(gyro0, accel0, dt);
} else {
LOG(FATAL) << "Unknown integration method";
}
P_.block<kMotionSize, kMotionSize>(0, 0).noalias() += Qmodel_;
timer_.Tock("propagation");
}
void Estimator::Fehlberg(const Vec3 &gyro0, const Vec3 &accel0, number_t dt) {
throw NotImplemented();
}
void Estimator::ComposeMotion(State &X, const Vec3 &V,
const Eigen::Matrix<number_t, 6, 1> &gyro_accel,
number_t dt) {
Vec3 gyro = gyro_accel.head<3>();
Vec3 accel = gyro_accel.tail<3>();
Vec3 gyro_calib = imu_.Cg() * gyro - X.bg;
Vec3 accel_calib = imu_.Ca() * accel - X.ba;
// integrate the nominal state
X.Tsb += V * dt; //+ 0.5 * a * dt * dt;
X.Vsb += (X.Rsb * accel_calib + X.Rsg * g_) * dt;
X.Rsb *= SO3::exp(gyro_calib * dt);
X.Rsb.normalize();
}
void Estimator::ComputeMotionJacobianAt(
const State &X, const Eigen::Matrix<number_t, 6, 1> &gyro_accel) {
Vec3 gyro = gyro_accel.head<3>();
Vec3 accel = gyro_accel.tail<3>();
Vec3 gyro_calib = imu_.Cg() * gyro - X.bg; // \hat\omega in the doc
Vec3 accel_calib = imu_.Ca() * accel - X.ba; // \hat\alpha in the doc
// jacobian w.r.t. error state
Mat3 Rsb = X.Rsb.matrix();
Eigen::Matrix<number_t, 3, 9> dWsb_dCg;
for (int i = 0; i < 3; ++i) {
// NOTE: use the raw measurement (gyro) here. NOT the calibrated one
// (gyro_calib)!!!
dWsb_dCg.block<1, 3>(i, 3 * i) = gyro;
}
Eigen::Matrix<number_t, 3, 9> dV_dRCa = dAB_dA<3, 3>(accel);
Eigen::Matrix<number_t, 9, 9> dRCa_dCafm = dAB_dB<3, 3>(Rsb); // fm: full matrix
Eigen::Matrix<number_t, 9, 6> dCafm_dCa = dA_dAu<number_t, 3>(); // full matrix w.r.t. upper triangle
Eigen::Matrix<number_t, 3, 6> dV_dCa = dV_dRCa * dRCa_dCafm * dCafm_dCa;
Mat3 dWsb_dWsb = -SO3::hat(gyro_calib);
// static Mat3 dW_dbg = -I3;
// static Mat3 dT_dV = I3;
Mat3 dV_dWsb = -Rsb * SO3::hat(accel_calib);
Mat3 dV_dba = -Rsb;
Mat3 dV_dWsg = -Rsb * SO3::hat(g_); // effective dimension: 3x2, since Wg is 2-dim
// Mat2 dWg_dWg = Mat2::Identity();
F_.setZero(); // wipe out the delta added to F in the previous step
for (int j = 0; j < 3; ++j) {
F_.coeffRef(Index::Wsb + j, Index::bg + j) = -1; // dW_dbg
F_.coeffRef(Index::Tsb + j, Index::Vsb + j) = 1; // dT_dV
for (int i = 0; i < 3; ++i) {
// W
F_.coeffRef(Index::Wsb + i, Index::Wsb + j) = dWsb_dWsb(i, j);
// F_.coeffRef(Index::W + i, Index::bg + j) = dW_dbg(i, j);
// T
// F_.coeffRef(Index::T + i, Index::V + j) = dT_dV(i, j);
// V
F_.coeffRef(Index::Vsb + i, Index::Wsb + j) = dV_dWsb(i, j);
F_.coeffRef(Index::Vsb + i, Index::ba + j) = dV_dba(i, j);
if (j < 2) {
// NOTE: Wg is 2-dim, i.e., NO z-component
F_.coeffRef(Index::Vsb + i, Index::Wsg + j) = dV_dWsg(i, j);
}
}
}
#ifdef USE_ONLINE_IMU_CALIB
for (int j = 0; j < 9; ++j) {
for (int i = 0 ; i < 3; ++i) {
F_.coeffRef(Index::Wsb + i, Index::Cg + j) = dWsb_dCg(i, j);
}
}
for (int j = 0; j < 6; ++j) {
for (int i = 0; i < 3; ++i) {
F_.coeffRef(Index::Vsb + i, Index::Ca + j) = dV_dCa(i, j);
}
}
#endif
// Mat3 dW_dng = -I3;
// Mat3 dV_dna = -R;
// Mat3 dbg_dnbg = I3;
// Mat3 dba_dnba = I3;
// jacobian w.r.t. noise
G_.setZero();
for (int j = 0; j < 3; ++j) {
G_.coeffRef(Index::Wsb + j, j) = -1; // dWsb_dng
G_.coeffRef(Index::bg + j, 6 + j) = 1; // dbg_dnbg
G_.coeffRef(Index::ba + j, 9 + j) = 1; // dba_dnba
for (int i = 0; i < 3; ++i) {
G_.coeffRef(Index::Vsb + i, 3 + j) = -Rsb(i, j); // dV_dna
}
}
}
bool Estimator::GoodTimestamp(const timestamp_t &now) {
auto now_ms = std::chrono::duration_cast<std::chrono::milliseconds>(now);
auto curr_ms =
std::chrono::duration_cast<std::chrono::milliseconds>(curr_time_);
if (now_ms < curr_ms) {
LOG(WARNING) << StrFormat("now=%ld ms < curr=%ld ms", now_ms.count(),
curr_ms.count());
return false;
} else {
return true;
}
}
void Estimator::UpdateSystemClock(const timestamp_t &now) {
if (!vision_initialized_) {
if (gravity_initialized_) {
// only initialize vision module after gravity initialized
curr_time_ = now;
last_vision_time_ = curr_vision_time_;
curr_vision_time_ = now;
vision_initialized_ = true;
LOG(INFO) << "vision initialized";
}
} else {
last_time_ = curr_time_;
curr_time_ = now;
last_vision_time_ = curr_vision_time_;
curr_vision_time_ = now;
}
}
void Estimator::RemoveGroupFromState(GroupPtr g) {
#ifndef NDEBUG
CHECK(g->instate()) << "free a group not instate";
CHECK(g->sind() != -1) << "invalid state index";
CHECK(gsel_[g->sind()]) << "Group not in state?!";
#endif
VLOG(0) << "removing group #" << g->id();
// change the covariance and error state
int index = g->sind();
gsel_[index] = false;
g->SetSind(-1);
g->SetStatus(GroupStatus::FLOATING);
int offset = kGroupBegin + 6 * index;
int size = err_.rows();
err_.segment<6>(offset).setZero();
P_.block(offset, 0, 6, size).setZero();
P_.block(0, offset, size, 6).setZero();
}
void Estimator::RemoveFeatureFromState(FeaturePtr f) {
#ifndef NDEBUG
CHECK((f->instate() && (f->track_status() == TrackStatus::DROPPED)) ||
(f->status() == FeatureStatus::REJECTED_BY_FILTER) ||
(std::count(affected_groups_.begin(), affected_groups_.end(), f->ref()) > 0));
CHECK(f->sind() != -1) << "invalid state index";
CHECK(fsel_[f->sind()]) << "Feature not in state?!";
#endif
VLOG(0) << "removing feature #" << f->id();
int index = f->sind();
fsel_[index] = false;
f->SetSind(-1);
int offset = kFeatureBegin + 3 * index;
int size = err_.rows();
err_.segment<3>(offset).setZero();
P_.block(offset, 0, 3, size).setZero();
P_.block(0, offset, size, 3).setZero();
}
void Estimator::AddGroupToState(GroupPtr g) {
#ifndef NDEBUG
CHECK(!g->instate()) << "group already in state";
CHECK(g->sind() == -1) << "group slot already allocated";
#endif
// change the covariance and error state
int index;
// find empty slot
for (index = 0; index < gsel_.size() && gsel_[index]; ++index)
;
if (index < gsel_.size()) {
gsel_[index] = true;
g->SetSind(index);
g->SetStatus(GroupStatus::INSTATE);
int offset = kGroupBegin + 6 * index;
// with gsb=(Rsb, Tsb) as the augmented state
// augmentation is much simpler
err_.segment<3>(offset) = err_.segment<3>(Index::Wsb);
err_.segment<3>(offset + 3) = err_.segment<3>(Index::Tsb);
P_.block(offset, 0, 3, err_.size()) =
P_.block(Index::Wsb, 0, 3, err_.size());
P_.block(0, offset, err_.size(), 3) =
P_.block(0, Index::Wsb, err_.size(), 3);
P_.block(offset + 3, 0, 3, err_.size()) =
P_.block(Index::Tsb, 0, 3, err_.size());
P_.block(0, offset + 3, err_.size(), 3) =
P_.block(0, Index::Tsb, err_.size(), 3);
VLOG(0) << StrFormat("group #%d inserted @ %d/%d", g->id(), index,
kMaxGroup);
} else {
throw std::runtime_error("Failed to find slot in state for group.");
}
}
void Estimator::AddFeatureToState(FeaturePtr f) {
#ifndef NDEBUG
CHECK(!f->instate()) << "feature already in state";
CHECK(f->sind() == -1) << "feature slot already allocated";
#endif
// change the covariance and error state
int index;
// find empty slot
for (index = 0; index < fsel_.size() && fsel_[index]; ++index)
;
if (index < fsel_.size()) {
fsel_[index] = true;
f->SetStatus(FeatureStatus::INSTATE);
f->SetSind(index);
f->FillCovarianceBlock(P_);
VLOG(0) << StrFormat("feature #%d inserted @ %d/%d", f->id(), index,
kMaxFeature);
} else {
throw std::runtime_error("Failed to find slot in state for feature.");
}
}
void Estimator::PrintErrorStateNorm() {
VLOG(0) << StrFormat(
"|Wsb|=%0.8f, |Tsb|=%0.8f, |Vsb|=%0.8f, "
"|bg|=%0.8f, |ba|=%0.8f, |Wbc|=%0.8f, |Tbc|=%0.8f, |Wsg|=%0.8f\n",
err_.segment<3>(Index::Wsb).norm(), err_.segment<3>(Index::Tsb).norm(),
err_.segment<3>(Index::Vsb).norm(), err_.segment<3>(Index::bg).norm(),
err_.segment<3>(Index::ba).norm(), err_.segment<3>(Index::Wbc).norm(),
err_.segment<3>(Index::Tbc).norm(), err_.segment<2>(Index::Wsg).norm());
for (auto g : instate_groups_) {
#ifndef NDEBUG
CHECK(gsel_[g->sind()]) << "instate group not actually instate";
#endif
VLOG(0) << StrFormat(
"g#%d |W|=%0.8f, |T|=%0.8f\n", g->id(),
err_.segment<3>(kGroupBegin + 6 * g->sind()).norm(),
err_.segment<3>(kGroupBegin + 6 * g->sind() + 3).norm());
}
for (auto f : instate_features_) {
#ifndef NDEBUG
CHECK(fsel_[f->sind()]) << "instate feature not yet instate";
#endif
VLOG(0) << StrFormat(
"f#%d |X|=%0.8f\n", f->id(),
err_.segment<3>(kFeatureBegin + 3 * f->sind()).norm());
}
}
void Estimator::AbsorbError(const VecX &err) {
// motion state
this->UpdateState(err.head<kMotionSize>());
#ifdef USE_ONLINE_IMU_CALIB
// update IMU state
Eigen::Matrix<number_t, 15, 1> dCaCg;
dCaCg << err.segment<6>(Index::Ca), err.segment<9>(Index::Cg);
imu_.UpdateState(dCaCg);
#endif
#ifdef USE_ONLINE_CAMERA_CALIB
// update camera instrinsics
Camera::instance()->UpdateState(
err.segment<kMaxCameraIntrinsics>(kCameraBegin));
#endif
// Camera::instance()->Print(std::cout);
// std::cout << "Ca=\n" << imu_.Ca() << std::endl;
// std::cout << "Cg=\n" << imu_.Cg() << std::endl;
// std::cout << "td=" << err(Index::td) << std::endl;
// augmented state
for (auto g : instate_groups_) {
#ifndef NDEBUG
CHECK(g->sind() != -1);
#endif
int offset = kGroupBegin + 6 * g->sind();
g->UpdateState(err.segment<6>(offset));
// if (g->id() == gauge_group_) {
// std::cout << "gauge group:" << err.segment<6>(offset).transpose() <<
// std::endl;
// }
}
for (auto f : in_current_ekf_update_) {
#ifndef NDEBUG
CHECK(f->sind() != -1);
#endif
int offset = kFeatureBegin + 3 * f->sind();
f->UpdateState(err.segment<3>(offset));
}
}
void Estimator::AbsorbError() {
AbsorbError(err_);
err_.setZero();
}
void Estimator::MaintainBuffer() {
if (!buf_.initialized) {
if (buf_.size() >= InternalBuffer::MAX_SIZE) {
std::make_heap(buf_.begin(), buf_.end(), cmp);
buf_.initialized = true;
}
} else {
std::push_heap(buf_.begin(), buf_.end(), cmp);
}
if (!async_run_) {
// execute here
if (buf_.initialized && buf_.size() > InternalBuffer::MAX_SIZE) {
buf_.front()->Execute(this);
std::pop_heap(buf_.begin(), buf_.end(), cmp);
buf_.pop_back();
}
}
}
void Estimator::VisualMeas(const timestamp_t &ts_raw, const cv::Mat &img) {
timestamp_t ts{ts_raw};
#ifdef USE_ONLINE_TEMPORAL_CALIB
if (X_.td >= 0) {
ts += timestamp_t(uint64_t(X_.td * 1e9)); // seconds -> nanoseconds
} else {
ts -= timestamp_t(uint64_t(-X_.td * 1e9)); // seconds -> nanoseconds
}
#endif
if (async_run_) {
std::scoped_lock lck(buf_.mtx);
buf_.push_back(std::make_unique<internal::Visual>(ts, img));
MaintainBuffer();
} else {
buf_.push_back(std::make_unique<internal::Visual>(ts, img));
MaintainBuffer();
}
}
void Estimator::VisualMeasTrackerOnly(const timestamp_t &ts_raw, const cv::Mat &img) {
timestamp_t ts{ts_raw};
#ifdef USE_ONLINE_TEMPORAL_CALIB
if (X_.td >= 0) {
ts += timestamp_t(uint64_t(X_.td * 1e9)); // seconds -> nanoseconds
} else {
ts -= timestamp_t(uint64_t(-X_.td * 1e9)); // seconds -> nanoseconds
}
#endif
if (async_run_) {
std::scoped_lock lck(buf_.mtx);
buf_.push_back(std::make_unique<internal::VisualTrackerOnly>(ts, img));
MaintainBuffer();
} else {
buf_.push_back(std::make_unique<internal::VisualTrackerOnly>(ts, img));
MaintainBuffer();
}
}
void Estimator::VisualMeasPointCloud(
const timestamp_t &ts_raw,
const VecXi &feature_ids,
const MatX3 &xp_and_depths)
{
timestamp_t ts{ts_raw};
#ifdef USE_ONLINE_TEMPORAL_CALIB
if (X_.td >= 0) {
ts += timestamp_t(uint64_t(X_.td * 1e9)); // seconds -> nanoseconds
} else {
ts -= timestamp_t(uint64_t(-X_.td * 1e9)); // seconds -> nanoseconds
}
#endif
if (async_run_) {
std::scoped_lock lck(buf_.mtx);
buf_.push_back(std::make_unique<internal::VisualPointCloud>(
ts, feature_ids, xp_and_depths));
MaintainBuffer();
} else {