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
#include "MultiKalmanBallLocator.h"

#include <Eigen/Core>
#include "Tools/Association.h"

MultiKalmanBallLocator::MultiKalmanBallLocator()
{
    // Modify number of models
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:remove_all_models",     "remove all models",                                                             false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:allow_just_one_model",  "allows only one model to be generated (all updates are applied to that model)", false);

    // Debug Drawings
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_real_ball_percept",          "draw the real incomming ball percept",                               false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_ball_on_field_before",       "draw the modelled ball on the field before prediction and update",   false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_ball_on_field",              "draw the modelled ball on the field before update",                  false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_ball_on_field_after",        "draw the modelled ball on the field after prediction and update",    false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_assignment",                 "draws the assignment of the ball percept to the filter",             false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_final_ball",                 "draws the final i.e. best model",                                    false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_final_ball_postion_at_rest", "draws the final i.e. best model's rest position",                    false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_covariance_ellipse",         "draws the ellipses representing the covariances",                    false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_last_known_ball",            "draws the last known ball", 	                                   false);

    // Plotting Related Debug Requests
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:plot_prediction_error",     "plots the prediction errors in x (horizontal angle) and y (vertical angle)", false);

    // Update Association Function Debug Requests
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:UpdateAssociationFunction:useEuclid",            "minimize Euclidian distance in measurement space",                                false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:UpdateAssociationFunction:useMahalanobis",       "minimize Mahalanobis distance in measurement space (no common covarince matrix)", false);
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:UpdateAssociationFunction:useMaximumLikelihood", "maximize likelihood of measurement in measurement space ",                        true );

    // Parameter Related Debug Requests
    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:reloadParameters",          "reloads the kalman filter parameters from the kfParameter object", false);

    DEBUG_REQUEST_REGISTER("MultiKalmanBallLocator:draw_trust_the_ball", "..", false);

    h.ballRadius = getFieldInfo().ballRadius;

    updateAssociationFunction = &likelihood;

    getDebugParameterList().add(&params);

    filter.reserve(10);  // set capacity for improved performance

    reloadParameters();
}

MultiKalmanBallLocator::~MultiKalmanBallLocator()
{
    getDebugParameterList().remove(&params);
}

void MultiKalmanBallLocator::execute()
{
    // allways reset the model first
    getBallModel().reset();

    // HACK: no updates in ready or when lifted
    if(getPlayerInfo().robotState == PlayerInfo::ready || getBodyState().isLiftedUp) {
      filter.clear();
      return;
    }

    DEBUG_REQUEST("MultiKalmanBallLocator:remove_all_models",
        filter.clear();
    );

    // delete some filter if they are too bad
    // NOTE: make sure at least one filter remains, so filter is not empty
    Filters::iterator iter = filter.begin();
    while(iter != filter.end() && filter.size() > 1) {
      // TODO: here we make a linear approximation depending n the distance - does it make problems? Can we have a better solution?
      double distance = Vector2d(iter->getState()(0), iter->getState()(2)).abs();
      double threshold_radius = params.area95Threshold_radius.factor * distance + params.area95Threshold_radius.offset;
      // HACK: we need ballSeenFilter here because the other condition didn't seem to work as expected
      // compare area of the position uncertainty ellipse with a circle representing the maximum uncertainty a ball can have
      // Note: pi is avoided on both sides of the inequality
      if(!iter->ballSeenFilter.value() && (*iter).getEllipseLocation().major * (*iter).getEllipseLocation().minor > threshold_radius*threshold_radius){
          iter = filter.erase(iter);
      } else {
          ++iter;
      }
    }

    // apply odometry on the filter state, to keep it in the robot's local coordinate system
    for(Filters::iterator iter = filter.begin(); iter != filter.end(); ++iter) {
        applyOdometryOnFilterState(*iter);
    }

    doDebugRequestBeforPredictionAndUpdate();

    // prediction
    double dt = getFrameInfo().getTimeInSeconds() - lastFrameInfo.getTimeInSeconds();
    ASSERT(dt > 0);
    for(Filters::iterator iter = filter.begin(); iter != filter.end(); ++iter) {
      predict(*iter, dt);
    }

    doDebugRequestBeforUpdate();

    // sensor update
    if(params.association.use_normal) {
        updateByPerceptsNormal();
    } else if(params.association.use_cool) {
        updateByPerceptsCool();
    } else if(params.association.use_greedy) {
        // need to handle bottom and top percepts independently because both cameras can observe the same ball
        updateByPerceptsGreedy(CameraInfo::Bottom);
        updateByPerceptsGreedy(CameraInfo::Top);
    }

    // NOTE: (Heinrich) update the "ball seen" values
    for(Filters::iterator iter = filter.begin(); iter != filter.end(); ++iter) {
      bool updated = iter->getLastUpdateFrame().getFrameNumber() == getFrameInfo().getFrameNumber();
      (*iter).ballSeenFilter.setParameter(params.g0, params.g1);
      (*iter).ballSeenFilter.update(updated, 0.3, 0.7);
    }

    // estimate the best model
    if(params.use_covariance_based_selection) {
        bestModel = selectBestModelBasedOnCovariance();
    } else {
        bestModel = selectBestModel();
    }

    // fill the ball model representation
    if (bestModel != filter.end()) {
      provideBallModel(*bestModel);
    }

    doDebugRequest();

    lastFrameInfo     = getFrameInfo();
    lastRobotOdometry = getOdometryData();
}

void MultiKalmanBallLocator::updateByPerceptsNormal()
{
  // measurement
    for(MultiBallPercept::ConstABPIterator iter = getMultiBallPercept().begin(); iter != getMultiBallPercept().end(); iter++) {

        Eigen::Vector2d z;
        Vector2d p;

        // set correct camera matrix and info in functional
        if((*iter).cameraId == CameraInfo::Bottom)
        {
            //PLOT("MultiKalmanBallLocator:Measurement:Bottom:horizontal", z(0));
            //PLOT("MultiKalmanBallLocator:Measurement:Bottom:vertical",   z(1));

            h.camMat  = getCameraMatrix();
            h.camInfo = getCameraInfo();
        }
        else
        {
            //PLOT("MultiKalmanBallLocator:Measurement:Top:horizontal", z(0));
            //PLOT("MultiKalmanBallLocator:Measurement:Top:vertical",   z(1));

            h.camMat  = getCameraMatrixTop();
            h.camInfo = getCameraInfoTop();
        }

        // tansform measurement into angles
        Vector2d angles = CameraGeometry::pixelToAngles(h.camInfo,(*iter).centerInImage.x,(*iter).centerInImage.y);
        z << angles.x, angles.y;

        // needed if a new filter has to be created
        p = (*iter).positionOnField;

        // find best matching filter
        updateAssociationFunction->determineBestPredictor(filter, z, h);

        Filters::iterator bestPredictor = updateAssociationFunction->getBestPredictor();

        bool allowJustOneModel = false;
        DEBUG_REQUEST("MultiKalmanBallLocator:allow_just_one_model",
            allowJustOneModel = true;
        );

        // if no suitable filter found create a new one
        if((!allowJustOneModel && !updateAssociationFunction->inRange()) || bestPredictor == filter.end())
        {
            Eigen::Vector4d newState;
            newState << p.x, 0, p.y, 0;
            filter.push_back(BallHypothesis(getFrameInfo(), newState, processNoiseStdSingleDimension, measurementNoiseCovariances, initialStateStdSingleDimension));
        }
        else
        {
            DEBUG_REQUEST("MultiKalmanBallLocator:draw_assignment",
                FIELD_DRAWING_CONTEXT;
                PEN("FF0000", 10);
                const Eigen::Vector4d state = (*bestPredictor).getState();
                LINE(p.x,p.y,state(0),state(2));
                TEXT_DRAWING((p.x+state(0))/2,(p.y+state(2))/2,updateAssociationFunction->getScore());
            );

            // debug stuff -> should be in a DEBUG_REQUEST
            DEBUG_REQUEST("MultiKalmanBallLocator:plot_prediction_error",
                Eigen::Vector2d prediction_error;
                prediction_error = z - (*bestPredictor).getStateInMeasurementSpace(h);

                PLOT("MultiKalmanBallLocator:Innovation:x", prediction_error(0));
                PLOT("MultiKalmanBallLocator:Innovation:y", prediction_error(1));
            );

            (*bestPredictor).update(z,h,getFrameInfo());
        }
    }// end for
}

/*
 * Perform a greedy 1:1-matching of filters (hypothesis) and measurements (ball percept).
 * The association depends on the horizontal and vertical angles of the percept and
 * the hypothesis in the image. Each combination is assigned a score. The combination
 * with the best score is choosen and the filter updated with the measurement.
 * After that the second best and so on...
 */
void MultiKalmanBallLocator::updateByPerceptsGreedy(CameraInfo::CameraID camera)
{
    // set correct camera matrix and info in functional
    if(camera == CameraInfo::Bottom) {
        h.camMat  = getCameraMatrix();
        h.camInfo = getCameraInfo();
    } else {
        h.camMat  = getCameraMatrixTop();
        h.camInfo = getCameraInfoTop();
    }

    // phase 1: filter percepts
    std::vector<Eigen::Vector2d, Eigen::aligned_allocator<Eigen::Vector2d> > measurements;
    std::vector<Vector2d> positions;

    for(const MultiBallPercept::BallPercept& percept: getMultiBallPercept().getPercepts()){
        Eigen::Vector2d z;

        if(percept.cameraId == camera){
            // transform measurement into angles
            Vector2d angles = CameraGeometry::pixelToAngles(h.camInfo, percept.centerInImage.x, percept.centerInImage.y);
            z << angles.x, angles.y;
            measurements.push_back(z);

            // needed if a new filter has to be created
            positions.push_back(percept.positionOnField);
        }
    }

    // abort if there are no measurements for the current camera
    if(measurements.empty()) return;

    // if no filter exists, skip directly to phase 4
    if(!filter.empty())
    {
        // phase 2: calculate #measurement x #filter score-matrix
        // Note: it depends on the association function whether a low or a high score is better
        Eigen::MatrixXd scores = Eigen::MatrixXd::Zero(measurements.size(), filter.size());
        for(size_t i = 0; i < measurements.size(); ++i){
            for(size_t j = 0; j < filter.size(); ++j){
                scores(i,j) = updateAssociationFunction->associationScore(filter[j], measurements[i], h);
            }
        }

        // create index vector for filters
        std::vector<int> filterIndices;
        for(size_t i = 0; i < filter.size(); ++i){
            filterIndices.push_back((int)i);
        }

        // phase 3: greedyly update each filter by best matching percept
        while(!(filterIndices.empty() || measurements.empty())){
            // phase 3.1: find location of best entry
            // which is depending on the updateAssociationFunction a global maximum or global minimum
            Eigen::Index maxColIdx, minColIdx, maxRowIdx, minRowIdx, bestColIdx, bestRowIdx;
            double max = scores.maxCoeff(&maxRowIdx, &maxColIdx);
            double min = scores.minCoeff(&minRowIdx, &minColIdx);

            if(updateAssociationFunction->better(min, max)){
                bestColIdx = minColIdx;
                bestRowIdx = minRowIdx;
            } else {
                bestColIdx = maxColIdx;
                bestRowIdx = maxRowIdx;
            }

            // phase 3.2: update filter
            bool allowJustOneModel = false;
            DEBUG_REQUEST("MultiKalmanBallLocator:allow_just_one_model",
                allowJustOneModel = true;
            );

            // if the score is not in the valid range continue with phase 4
            // because no other possible matching could be better than the current one
            if(!allowJustOneModel && !updateAssociationFunction->inRange(scores(bestRowIdx,bestColIdx))) {
                break;
            } else {
                DEBUG_REQUEST("MultiKalmanBallLocator:draw_assignment",
                    FIELD_DRAWING_CONTEXT;
                    PEN("FF0000", 10);
                    const Eigen::Vector4d state = filter[filterIndices[bestColIdx]].getState();
                    LINE(positions[bestRowIdx].x, positions[bestRowIdx].y, state(0), state(2));
                    TEXT_DRAWING((positions[bestRowIdx].x + state(0)) / 2, (positions[bestRowIdx].y + state(2)) / 2, scores(bestRowIdx,bestColIdx));
                );

                DEBUG_REQUEST("MultiKalmanBallLocator:plot_prediction_error",
                    Eigen::Vector2d prediction_error;
                    prediction_error = measurements[bestRowIdx] - filter[filterIndices[bestColIdx]].getStateInMeasurementSpace(h);

                    PLOT("MultiKalmanBallLocator:Innovation:x", prediction_error(0));
                    PLOT("MultiKalmanBallLocator:Innovation:y", prediction_error(1));
                );

                filter[filterIndices[bestColIdx]].update(measurements[bestRowIdx], h, getFrameInfo());
            }

            // phase 3.3: remove matching pair of hypothesis and measurement
            // from the set of matchable options, that means:
            // - remove bestColIdx from filterIndices
            // - remove the measurement and its position from measurements and positions
            // - remove their corresponding row and column in the score matrix
            filterIndices.erase(filterIndices.begin() + bestColIdx);
            measurements.erase(measurements.begin() + bestRowIdx);
            positions.erase(positions.begin() + bestRowIdx);

            Eigen::Index numRowsAfterResize = scores.rows() - 1;
            Eigen::Index numColsAfterResize = scores.cols() - 1;
            Eigen::Index numRowsAfterBestRow = numRowsAfterResize - bestRowIdx;
            Eigen::Index numColsAfterBestCol = numColsAfterResize - bestColIdx;

            // delete the row by moving all rows after the best row one row up and resize the matrix accordingly
            if(bestRowIdx < numRowsAfterResize)
                scores.block(bestRowIdx, 0, numRowsAfterBestRow, scores.cols()) = scores.bottomRows(numRowsAfterBestRow).eval();
            scores.conservativeResize(numRowsAfterResize, scores.cols());

            // delete the column by moving all columns right of the best column one column to the left and resize the matrix accordingly
            if(bestColIdx < numColsAfterResize)
                scores.block(0, bestColIdx, numRowsAfterResize, numColsAfterBestCol) = scores.rightCols(numColsAfterBestCol).eval();
            scores.conservativeResize(numRowsAfterResize, numColsAfterResize);
        }
    }

    // phase 4: create new filter for each percept which couldn't be associated with a filter
    // Diskussion: vielleicht filter erstellen wenn keine da ist und dann die zweite beobachtung dazu matchen ...
    for(const Vector2d& p : positions){
        Eigen::Vector4d newState;
        newState << p.x, 0, p.y, 0;
        // add BallHypothesis to filter vector
        filter.emplace_back(getFrameInfo(), newState, processNoiseStdSingleDimension, measurementNoiseCovariances, initialStateStdSingleDimension);
    }

    // TODO maybe add possibility to merge filters
    // TODO maybe add possibility to apply update of observation to every filter in a bayesian fashion
}

void MultiKalmanBallLocator::updateByPerceptsCool()
{
  // update by percepts
  // A ~ goal posts
  // B ~ hypotheses
  Assoziation sensorAssoziation(static_cast<int>(getMultiBallPercept().getPercepts().size()), static_cast<int>(filter.size()));

  int i = 0;
  for(MultiBallPercept::ConstABPIterator iter = getMultiBallPercept().begin(); iter != getMultiBallPercept().end(); iter++)
  {
      // set correct camera matrix and info in functional
      if((*iter).cameraId == CameraInfo::Bottom) {
          h.camMat  = getCameraMatrix();
          h.camInfo = getCameraInfo();
      } else {
          h.camMat  = getCameraMatrixTop();
          h.camInfo = getCameraInfoTop();
      }

      // tansform measurement into angles
      Vector2d angles = CameraGeometry::pixelToAngles(h.camInfo,(*iter).centerInImage.x,(*iter).centerInImage.y);
      Eigen::Vector2d z;
      z << angles.x, angles.y;

      int x = 0;
      for(Filters::iterator iter = filter.begin(); iter != filter.end(); ++iter)
      {
        double confidence = updateAssociationFunction->associationScore(*iter, z, h);

        // store best association for measurement
        if(confidence > updateAssociationFunction->getThreshold() &&
          confidence > sensorAssoziation.getW4A(i) &&
          confidence > sensorAssoziation.getW4B(x))
        {
          sensorAssoziation.addAssociation(i, x, confidence);
        }
        x++;
      }
      i++;
  }

  i = 0;
  for(MultiBallPercept::ConstABPIterator iter = getMultiBallPercept().begin(); iter != getMultiBallPercept().end(); iter++)
  {
    int x = sensorAssoziation.getB4A(i); // get hypothesis for measurement
    if (x != -1)
    {
      // set correct camera matrix and info in functional
      if((*iter).cameraId == CameraInfo::Bottom) {
          h.camMat  = getCameraMatrix();
          h.camInfo = getCameraInfo();
      } else {
          h.camMat  = getCameraMatrixTop();
          h.camInfo = getCameraInfoTop();
      }

      // tansform measurement into angles
      Vector2d angles = CameraGeometry::pixelToAngles(h.camInfo,(*iter).centerInImage.x,(*iter).centerInImage.y);
      Eigen::Vector2d z;
      z << angles.x, angles.y;

      filter[x].update(z,h,getFrameInfo());
    }
    else
    {
      Vector2d p = (*iter).positionOnField;
      Eigen::Vector4d newState;
      newState << p.x, 0, p.y, 0;
      filter.push_back(BallHypothesis(getFrameInfo(), newState, processNoiseStdSingleDimension, measurementNoiseCovariances, initialStateStdSingleDimension));
    }
    i++;
  }

}

void MultiKalmanBallLocator::predict(ExtendedKalmanFilter4d& filter, double dt) const
{
    /*
        rolling resistance = rolling resitance coefficient * normal force
        F_R = d / R * F_N

        F_N = g * weight
        g = 9.81
        weight = ballMass

        deceleration = F_R/ballMass
    */

    const Eigen::Vector4d& x = filter.getState();
    Eigen::Vector2d vel; // control vector
    vel <<  x(1), x(3);
    double abs_velocity = vel.norm();

    double time_until_vel_zero = 0;

    if(abs_velocity > epsilon){
        time_until_vel_zero = -abs_velocity/getFieldInfo().ballDeceleration;
    }

    if(time_until_vel_zero < epsilon)
    {
        filter.predict(Eigen::Vector2d::Zero(),dt);
    } else {
        // ballDeceleration is negative so the deceleration will be in opposite direction of current velocity
        Eigen::Vector2d u = vel.normalized() * getFieldInfo().ballDeceleration;
        if(time_until_vel_zero >= dt) {
            filter.predict(u,dt);
        } else {
            filter.predict(u, time_until_vel_zero);
            dt -= time_until_vel_zero;
            filter.predict(Eigen::Vector2d::Zero(), dt);
        }
    }

    return;
}

void MultiKalmanBallLocator::applyOdometryOnFilterState(ExtendedKalmanFilter4d& filter)
{
    const Eigen::Vector4d& x = filter.getState();
    Pose2D odometryDelta = lastRobotOdometry - getOdometryData();

    Eigen::Vector4d newStateX;
    newStateX << x;

    // construct rotation matrix
    double s = sin(odometryDelta.getAngle());
    double c = cos(odometryDelta.getAngle());

    Eigen::Matrix4d rotation;
    rotation << c, 0, -s,  0,
                0, c,  0, -s,
                s, 0,  c,  0,
                0, s,  0,  c;

    newStateX = rotation*newStateX;

    // translate the location of the filters state (velocity is relative so translating it doesn't make sense)
    newStateX(0) = newStateX(0) + odometryDelta.translation.x;
    newStateX(2) = newStateX(2) + odometryDelta.translation.y;

    // rotate P (translation doesn't affect the covariances)
    Eigen::Matrix4d P = filter.getProcessCovariance();

    Eigen::Matrix4d new_P = rotation * P * rotation.transpose();

    filter.setState(newStateX);
    filter.setCovarianceOfState(new_P);
}

// TODO: returns the first model as best model even if it is "not seen"
//		 it might be better to return an invalid iterator and handle this case outside
//		 handling this better might make last_known_ball symbol obsolete
MultiKalmanBallLocator::Filters::const_iterator MultiKalmanBallLocator::selectBestModel() const
{
  // find the best model for the ball: closest hypothesis that is "known"
  Filters::const_iterator bestModel = filter.end();
  double minDistance = 0;

  for(Filters::const_iterator iter = filter.begin(); iter != filter.end(); ++iter) {
    double distance = Vector2d(iter->getState()(0), iter->getState()(2)).abs();
    if( bestModel == filter.end() ||
       (iter->ballSeenFilter.value() && !bestModel->ballSeenFilter.value()) ||
       (iter->ballSeenFilter.value() == bestModel->ballSeenFilter.value() && distance < minDistance))
    {
      bestModel = iter;
      minDistance = distance;
    }
  }

  return bestModel;
}

MultiKalmanBallLocator::Filters::const_iterator MultiKalmanBallLocator::selectBestModelBasedOnCovariance() const
{
  Filters::const_iterator bestModel = filter.end();
  double value = 0;

  // find the best seen model for the ball based on covariance
  for(Filters::const_iterator iter = filter.begin(); iter != filter.end(); ++iter) {
      double temp = iter->getEllipseLocation().major * iter->getEllipseLocation().minor * Math::pi;
      if(bestModel == filter.end()
         || (iter->ballSeenFilter.value() && !bestModel->ballSeenFilter.value())
         || (iter->ballSeenFilter.value() == bestModel->ballSeenFilter.value() && temp < value)) {
          bestModel = iter;
          value = temp;
      }
  }

  return bestModel;
}

void MultiKalmanBallLocator::provideBallModel(const BallHypothesis& model)
{
  getBallModel().valid = true;
  getBallModel().knows = model.ballSeenFilter.value();
  getBallModel().setFrameInfoWhenBallWasSeen(model.getLastUpdateFrame());

  // set ball model representation
  const Eigen::Vector4d& x = model.getState();
  getBallModel().position.x = x(0);
  getBallModel().position.y = x(2);
  getBallModel().speed.x = x(1);
  getBallModel().speed.y = x(3);

  // transform ball model into feet coordinates
  const Pose3D& lFoot = getKinematicChain().theLinks[KinematicChain::LFoot].M;
  const Pose3D& rFoot = getKinematicChain().theLinks[KinematicChain::RFoot].M;
  Vector2d ballLeftFoot  = lFoot.projectXY()/getBallModel().position;
  Vector2d ballRightFoot = rFoot.projectXY()/getBallModel().position;

  //set preview ball model representation
  getBallModel().positionPreview = getMotionStatus().plannedMotion.hip / getBallModel().position;
  getBallModel().positionPreviewInLFoot = getMotionStatus().plannedMotion.lFoot / ballLeftFoot;
  getBallModel().positionPreviewInRFoot = getMotionStatus().plannedMotion.rFoot / ballRightFoot;

  // determine rest position
  Eigen::Vector2d vel;
  vel <<  x(1), x(3);
  double abs_velocity = vel.norm();

  BallHypothesis modelCopy(model);
  if(abs_velocity > epsilon){
      double time_until_vel_zero = -abs_velocity / getFieldInfo().ballDeceleration;
      Eigen::Vector2d u = vel.normalized() * getFieldInfo().ballDeceleration;
      modelCopy.predict(u, time_until_vel_zero);

      DEBUG_REQUEST("MultiKalmanBallLocator:draw_final_ball_postion_at_rest",
          FIELD_DRAWING_CONTEXT;
          PEN(Color(Color::black), 20);
          CIRCLE(modelCopy.getState()(0), modelCopy.getState()(2), getFieldInfo().ballRadius-10);
      );
  }

  // set position at rest
  getBallModel().position_at_rest.x = modelCopy.getState()(0);
  getBallModel().position_at_rest.y = modelCopy.getState()(2);

  // update last known ball
  if(getBallModel().knows) {
    getBallModel().last_known_ball = getBallModel().position;
  } else {
    // need to update odometry by hand ...
    Pose2D odometryDelta = lastRobotOdometry - getOdometryData();
    getBallModel().last_known_ball = odometryDelta * getBallModel().last_known_ball;
  }

  // some final debug stuff
  PLOT("MultiKalmanBallLocator:ballSeenFilter", model.ballSeenFilter.floatValue());
}

void MultiKalmanBallLocator::doDebugRequestBeforPredictionAndUpdate()
{
    DEBUG_REQUEST("MultiKalmanBallLocator:draw_ball_on_field_before",
        drawFiltersOnField();
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:UpdateAssociationFunction:useEuclid",
        updateAssociationFunction = &euclid;
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:UpdateAssociationFunction:useMahalanobis",
        updateAssociationFunction = &mahalanobis;
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:UpdateAssociationFunction:useMaximumLikelihood",
        updateAssociationFunction = &likelihood;
    );
}

void MultiKalmanBallLocator::doDebugRequestBeforUpdate()
{
    DEBUG_REQUEST("MultiKalmanBallLocator:draw_ball_on_field",
        drawFiltersOnField();
    );
}

void MultiKalmanBallLocator::doDebugRequest()
{
    //PLOT("MultiKalmanBallLocator:ModelIsValid", getBallModel().valid);

    //to check correctness of the prediction
    DEBUG_REQUEST("MultiKalmanBallLocator:draw_real_ball_percept",
      if(getMultiBallPercept().wasSeen()) {
        FIELD_DRAWING_CONTEXT;
        PEN("FF0000", 10);
        for(MultiBallPercept::ConstABPIterator iter = getMultiBallPercept().begin(); iter != getMultiBallPercept().end(); iter++) {
            CIRCLE((*iter).positionOnField.x, (*iter).positionOnField.y, getFieldInfo().ballRadius-5);
        }
      }
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:draw_ball_on_field_after",
        drawFiltersOnField();
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:draw_final_ball",
        FIELD_DRAWING_CONTEXT;
        PEN("FF0000", 10);
        CIRCLE( getBallModel().position.x, getBallModel().position.y, getFieldInfo().ballRadius-10);
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:draw_last_known_ball",
        FIELD_DRAWING_CONTEXT;
        PEN("EF871E", 10);
        CIRCLE(getBallModel().last_known_ball.x, getBallModel().last_known_ball.y, getFieldInfo().ballRadius-10);
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:reloadParameters",
        reloadParameters();
    );

    DEBUG_REQUEST("MultiKalmanBallLocator:draw_trust_the_ball",
        FIELD_DRAWING_CONTEXT;
        for(Filters::const_iterator iter = filter.begin(); iter != filter.end(); iter++)
        {
          if((*iter).ballSeenFilter.value()) {
            PEN("00FF00", 10);
          } else {
            PEN("FF0000", 10);
          }

          const Eigen::Vector4d& state = (*iter).getState();
          CIRCLE( state(0), state(2), (*iter).ballSeenFilter.value()*1000);
        }
    );
}

void MultiKalmanBallLocator::drawFilter(const BallHypothesis& bh, const Color& model_color, Color cov_loc_color, Color cov_vel_color) const
{
    bool draw_covariances = false;
    DEBUG_REQUEST("MultiKalmanBallLocator:draw_covariance_ellipse",
        draw_covariances = true;
    );

    if(!draw_covariances) {
        cov_loc_color[cov_loc_color.Alpha] = 0;
        cov_vel_color[cov_vel_color.Alpha] = 0;
    }

    PEN(model_color.toString(),20);

    const Eigen::Vector4d& state = bh.getState();

    CIRCLE( state(0), state(2), getFieldInfo().ballRadius-10);
    ARROW( state(0), state(2),
           state(0)+state(1),
           state(2)+state(3));

    PEN(cov_loc_color.toString(), 20);
    const Ellipse2d& ellipse_loc = bh.getEllipseLocation();
    OVAL_ROTATED(state(0),
                 state(2),
                 ellipse_loc.minor,
                 ellipse_loc.major,
                 ellipse_loc.angle);

    PEN(cov_vel_color, 20);
    const Ellipse2d& ellipse_vel = bh.getEllipseVelocity();
    OVAL_ROTATED(state(0)+state(1),
                 state(2)+state(3),
                 ellipse_vel.minor,
                 ellipse_vel.major,
                 ellipse_vel.angle);

    if (!draw_covariances) {
      cov_loc_color[cov_loc_color.Alpha] = 1;
      cov_vel_color[cov_vel_color.Alpha] = 1;
    }
}

void MultiKalmanBallLocator::drawFiltersOnField() const
{
    FIELD_DRAWING_CONTEXT;

    Color cov_loc_color("00FFFF");  // cyan
    Color cov_vel_color("FF00FF");  // pink
    Color model_color;

    for(Filters::const_iterator iter = filter.begin(); iter != filter.end(); iter++) {
        if(getBallModel().valid) {
            if((*iter).getLastUpdateFrame().getTime() == getFrameInfo().getTime()) {
                if(bestModel == iter)
                    model_color = "99FF00";  // green
                else
                    model_color = "FF9900";  // orange
            } else {
                    model_color = "0099FF";  // light blue
            }
        } else {
            model_color = "999999";  // grey
        }

        drawFilter(*iter, model_color, cov_loc_color, cov_vel_color);
    }
}

void MultiKalmanBallLocator::reloadParameters()
{
    // parameters for initializing new filters
    processNoiseStdSingleDimension << params.processNoiseStdQ00, params.processNoiseStdQ01,
                                      params.processNoiseStdQ10, params.processNoiseStdQ11;

    measurementNoiseCovariances << params.measurementNoiseR00, params.measurementNoiseR10,
                                   params.measurementNoiseR10, params.measurementNoiseR11;

    initialStateStdSingleDimension << params.initialStateStdP00, params.initialStateStdP01,
                                      params.initialStateStdP10, params.initialStateStdP11;

    // UAF thresholds
    euclid.setThreshold(params.euclidThreshold);
    mahalanobis.setThreshold(params.mahalanobisThreshold);
    likelihood.setThreshold(params.maximumLikelihoodThreshold);

    // update existing filters with new process and measurement noise
    Eigen::Matrix2d processNoiseCovariancesSingleDimension;
    processNoiseCovariancesSingleDimension = processNoiseStdSingleDimension.cwiseProduct(processNoiseStdSingleDimension);

    for(Filters::iterator iter = filter.begin(); iter != filter.end(); iter++){
        (*iter).setCovarianceOfProcessNoise(processNoiseCovariancesSingleDimension);
        (*iter).setCovarianceOfMeasurementNoise(measurementNoiseCovariances);
    }
}