update accelerometer calibration to use Levenberg Marquardt least squares fit

- update accel cal to use Levenberg Marquardt least squares fit (same as magnetometer)
    - this is more robust to user positioning error as it only requires a nice distribution of points and doesn't actually depend on the user holding the vehicle in each orientation (although the calibration flow is still structured this way)
 - use online Welford mean to gather clean readings per orientation
 - reset if movement during per side data gathering
 - detect invalid rotations (and soon auto correct like mag)
 - add `CAL_ACC_SIDES` (analogous to CAL_MAG_SIDES) to limit accel calibration to a subset of orientations (when full set is difficult or impossible)
 - accel quick calibration (commander calibrate accel quick) assume vehicle is aligned with gravity and adjust offsets immediately
This commit is contained in:
Daniel Agar
2021-03-08 11:23:44 -05:00
parent 374d44f38f
commit 9955cb4cc8
9 changed files with 712 additions and 544 deletions
@@ -76,6 +76,7 @@ public:
const matrix::Dcmf &rotation() const { return _rotation; }
const Rotation &rotation_enum() const { return _rotation_enum; }
const matrix::Vector3f &scale() const { return _scale; }
const matrix::Vector3f &thermal_offset() const { return _thermal_offset; }
// apply offsets and scale
// rotate corrected measurements from sensor to body frame
@@ -1,6 +1,6 @@
/****************************************************************************
*
* Copyright (c) 2013-2020 PX4 Development Team. All rights reserved.
* Copyright (c) 2013-2021 PX4 Development Team. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
@@ -130,13 +130,17 @@
#include <px4_platform_common/posix.h>
#include <px4_platform_common/time.h>
#include "lm_fit.hpp"
#include <drivers/drv_hrt.h>
#include <include/containers/Bitset.hpp>
#include <lib/sensor_calibration/Accelerometer.hpp>
#include <lib/sensor_calibration/Utilities.hpp>
#include <lib/mathlib/mathlib.h>
#include <lib/geo/geo.h>
#include <matrix/math.hpp>
#include <lib/conversion/rotation.h>
#include <lib/mathlib/math/WelfordMean.hpp>
#include <lib/parameters/param.h>
#include <lib/systemlib/err.h>
#include <lib/systemlib/mavlink_log.h>
@@ -150,191 +154,174 @@ using namespace matrix;
using namespace time_literals;
static constexpr char sensor_name[] {"accel"};
static constexpr unsigned MAX_ACCEL_SENS = 4;
static constexpr uint8_t MAX_ACCEL_SENS = 4;
/// Data passed to calibration worker routine
struct accel_worker_data_s {
orb_advert_t *mavlink_log_pub{nullptr};
unsigned done_count{0};
float accel_ref[MAX_ACCEL_SENS][detect_orientation_side_count][3] {};
float accel_temperature_ref[MAX_ACCEL_SENS] {NAN, NAN, NAN, NAN};
orb_advert_t *mavlink_log_pub{nullptr};
unsigned done_count{0};
matrix::Vector3f accel_ref[MAX_ACCEL_SENS][detect_orientation_side_count] {};
float accel_temperature_ref[MAX_ACCEL_SENS] {NAN, NAN, NAN, NAN};
unsigned calibration_sides{0}; ///< The total number of sides
bool side_data_collected[detect_orientation_side_count] {};
};
// Read specified number of accelerometer samples, calculate average and dispersion.
static calibrate_return read_accelerometer_avg(float (&accel_avg)[MAX_ACCEL_SENS][detect_orientation_side_count][3],
float (&accel_temperature_avg)[MAX_ACCEL_SENS], unsigned orient, unsigned samples_num)
static calibrate_return read_accelerometer_avg(Vector3f(&accel_avg)[MAX_ACCEL_SENS][detect_orientation_side_count],
float (&accel_temperature_avg)[MAX_ACCEL_SENS], unsigned orient)
{
Vector3f accel_sum[MAX_ACCEL_SENS] {};
calibration::Accelerometer calibrations[MAX_ACCEL_SENS] {};
uORB::SubscriptionMultiArray<sensor_accel_s, MAX_ACCEL_SENS> accel_subs{ORB_ID::sensor_accel};
math::WelfordMean<Vector3f> mean[MAX_ACCEL_SENS] {};
Vector3f accel_prev[MAX_ACCEL_SENS] {};
float temperature_sum[MAX_ACCEL_SENS] {NAN, NAN, NAN, NAN};
unsigned counts[MAX_ACCEL_SENS] {};
unsigned errcount = 0;
// sensor thermal corrections
uORB::Subscription sensor_correction_sub{ORB_ID(sensor_correction)};
sensor_correction_s sensor_correction{};
sensor_correction_sub.copy(&sensor_correction);
uORB::SubscriptionBlocking<sensor_accel_s> accel_sub[MAX_ACCEL_SENS] {
{ORB_ID(sensor_accel), 0, 0},
{ORB_ID(sensor_accel), 0, 1},
{ORB_ID(sensor_accel), 0, 2},
{ORB_ID(sensor_accel), 0, 3},
};
unsigned temperature_count[MAX_ACCEL_SENS] {};
/* use the first sensor to pace the readout, but do per-sensor counts */
while (counts[0] < samples_num) {
if (accel_sub[0].updatedBlocking(100000)) {
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
sensor_accel_s arp;
bool done = false;
while (accel_sub[accel_index].update(&arp)) {
// fetch optional thermal offset corrections in sensor/board frame
Vector3f offset{0, 0, 0};
sensor_correction_sub.update(&sensor_correction);
const unsigned num_accels = math::min(accel_subs.advertised_count(), MAX_ACCEL_SENS);
if (sensor_correction.timestamp > 0 && arp.device_id != 0) {
for (uint8_t correction_index = 0; correction_index < MAX_ACCEL_SENS; correction_index++) {
if (sensor_correction.accel_device_ids[correction_index] == arp.device_id) {
switch (correction_index) {
case 0:
offset = Vector3f{sensor_correction.accel_offset_0};
break;
case 1:
offset = Vector3f{sensor_correction.accel_offset_1};
break;
case 2:
offset = Vector3f{sensor_correction.accel_offset_2};
break;
case 3:
offset = Vector3f{sensor_correction.accel_offset_3};
break;
}
}
}
}
const hrt_abstime timestamp_start = hrt_absolute_time();
accel_sum[accel_index] += Vector3f{arp.x, arp.y, arp.z} - offset;
while (!done && (hrt_elapsed_time(&timestamp_start) < 5_s)) {
unsigned good_count = 0;
counts[accel_index]++;
for (unsigned accel_index = 0; accel_index < num_accels; accel_index++) {
sensor_accel_s arp;
if (!PX4_ISFINITE(temperature_sum[accel_index])) {
// set first valid value
temperature_sum[accel_index] = (arp.temperature * counts[accel_index]);
while (accel_subs[accel_index].update(&arp)) {
calibrations[accel_index].set_device_id(arp.device_id);
calibrations[accel_index].SensorCorrectionsUpdate();
} else {
temperature_sum[accel_index] += arp.temperature;
}
// fetch optional thermal offset corrections in sensor/board frame
const Vector3f accel{Vector3f{arp.x, arp.y, arp.z} - calibrations[accel_index].thermal_offset()};
if ((accel - accel_prev[accel_index]).longerThan(0.5f)) {
mean[accel_index].reset();
accel_prev[accel_index] = accel;
} else {
mean[accel_index].update(accel);
}
if (!PX4_ISFINITE(temperature_sum[accel_index])) {
// set first valid value
temperature_sum[accel_index] = arp.temperature;
temperature_count[accel_index] = 1;
} else {
temperature_sum[accel_index] += arp.temperature;
temperature_count[accel_index]++;
}
}
if ((mean[accel_index].count() > 1000) && !mean[accel_index].variance().longerThan(0.01)) {
good_count++;
}
}
if ((good_count > 0) && (good_count == num_accels)) {
PX4_INFO("finished early: %d", mean[0].count());
done = true;
} else {
errcount++;
continue;
}
if (errcount > samples_num / 10) {
return calibrate_return_error;
px4_usleep(2000);
}
}
// rotate sensor measurements from sensor to body frame using board rotation matrix
const Dcmf board_rotation = calibration::GetBoardRotationMatrix();
unsigned good_count = 0;
for (unsigned s = 0; s < MAX_ACCEL_SENS; s++) {
accel_sum[s] = board_rotation * accel_sum[s];
for (unsigned s = 0; s < num_accels; s++) {
if (mean[s].valid()) {
const Vector3f avg = mean[s].mean();
const Vector3f var = mean[s].variance();
PX4_INFO("O: %d, Accel: %d, Mean: [%.6f, %.6f, %.6f], Variance: [%.6f, %.6f, %.6f]", orient, s,
(double)avg(0), (double)avg(1), (double)avg(2),
(double)var(0), (double)var(1), (double)var(2)
);
if (!var.longerThan(0.01f)) {
good_count++;
} else {
break;
}
accel_avg[s][orient] = mean[s].mean();
} else {
accel_avg[s][orient].zero();
}
}
for (unsigned s = 0; s < MAX_ACCEL_SENS; s++) {
const Vector3f avg{accel_sum[s] / counts[s]};
avg.copyTo(accel_avg[s][orient]);
if (temperature_count[s] > 0) {
accel_temperature_avg[s] = temperature_sum[s] / temperature_count[s];
accel_temperature_avg[s] = temperature_sum[s] / counts[s];
} else {
accel_temperature_avg[s] = NAN;
}
}
return calibrate_return_ok;
return (good_count == num_accels) ? calibrate_return_ok : calibrate_return_error;
}
static calibrate_return accel_calibration_worker(detect_orientation_return orientation, void *data)
{
static constexpr unsigned samples_num = 750;
accel_worker_data_s *worker_data = (accel_worker_data_s *)(data);
calibration_log_info(worker_data->mavlink_log_pub, "[cal] Hold still, measuring %s side",
detect_orientation_str(orientation));
int attempt = 0;
read_accelerometer_avg(worker_data->accel_ref, worker_data->accel_temperature_ref, orientation, samples_num);
while (attempt < 10) {
calibration_log_info(worker_data->mavlink_log_pub, "[cal] Hold still, measuring %s side",
detect_orientation_str(orientation));
// check accel
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
switch (orientation) {
case ORIENTATION_TAIL_DOWN: // [ g, 0, 0 ]
if (worker_data->accel_ref[accel_index][ORIENTATION_TAIL_DOWN][0] < 0.f) {
calibration_log_emergency(worker_data->mavlink_log_pub, "[cal] accel %d invalid X-axis, check rotation", accel_index);
return calibrate_return_error;
}
if (read_accelerometer_avg(worker_data->accel_ref, worker_data->accel_temperature_ref,
orientation) == calibrate_return_ok) {
break;
calibration_log_info(worker_data->mavlink_log_pub, "[cal] %s side result: [%.3f %.3f %.3f]",
detect_orientation_str(orientation),
(double)worker_data->accel_ref[0][orientation](0),
(double)worker_data->accel_ref[0][orientation](1),
(double)worker_data->accel_ref[0][orientation](2));
case ORIENTATION_NOSE_DOWN: // [ -g, 0, 0 ]
if (worker_data->accel_ref[accel_index][ORIENTATION_NOSE_DOWN][0] > 0.f) {
calibration_log_emergency(worker_data->mavlink_log_pub, "[cal] accel %d invalid X-axis, check rotation", accel_index);
return calibrate_return_error;
}
worker_data->done_count++;
calibration_log_info(worker_data->mavlink_log_pub, CAL_QGC_PROGRESS_MSG, 17 * worker_data->done_count);
break;
case ORIENTATION_LEFT: // [ 0, g, 0 ]
if (worker_data->accel_ref[accel_index][ORIENTATION_LEFT][1] < 0.f) {
calibration_log_emergency(worker_data->mavlink_log_pub, "[cal] accel %d invalid Y-axis, check rotation", accel_index);
return calibrate_return_error;
}
break;
case ORIENTATION_RIGHT: // [ 0, -g, 0 ]
if (worker_data->accel_ref[accel_index][ORIENTATION_RIGHT][1] > 0.f) {
calibration_log_emergency(worker_data->mavlink_log_pub, "[cal] accel %d invalid Y-axis, check rotation", accel_index);
return calibrate_return_error;
}
break;
case ORIENTATION_UPSIDE_DOWN: // [ 0, 0, g ]
if (worker_data->accel_ref[accel_index][ORIENTATION_UPSIDE_DOWN][2] < 0.f) {
calibration_log_emergency(worker_data->mavlink_log_pub, "[cal] accel %d invalid Z-axis, check rotation", accel_index);
return calibrate_return_error;
}
break;
case ORIENTATION_RIGHTSIDE_UP: // [ 0, 0, -g ]
if (worker_data->accel_ref[accel_index][ORIENTATION_RIGHTSIDE_UP][2] > 0.f) {
calibration_log_emergency(worker_data->mavlink_log_pub, "[cal] accel %d invalid Z-axis, check rotation", accel_index);
return calibrate_return_error;
}
break;
default:
break;
return calibrate_return_ok;
}
attempt++;
}
calibration_log_info(worker_data->mavlink_log_pub, "[cal] %s side result: [%.3f %.3f %.3f]",
detect_orientation_str(orientation),
(double)worker_data->accel_ref[0][orientation][0],
(double)worker_data->accel_ref[0][orientation][1],
(double)worker_data->accel_ref[0][orientation][2]);
worker_data->done_count++;
calibration_log_info(worker_data->mavlink_log_pub, CAL_QGC_PROGRESS_MSG, 17 * worker_data->done_count);
return calibrate_return_ok;
return calibrate_return_error;
}
int do_accel_calibration(orb_advert_t *mavlink_log_pub)
{
// We should not try to subscribe if the topic doesn't actually exist and can be counted.
const unsigned orb_accel_count = orb_group_count(ORB_ID(sensor_accel));
// Warn that we will not calibrate more than MAX_GYROS gyroscopes
if (orb_accel_count > MAX_ACCEL_SENS) {
calibration_log_critical(mavlink_log_pub, "Detected %u accels, but will calibrate only %u", orb_accel_count,
MAX_ACCEL_SENS);
} else if (orb_accel_count < 1) {
calibration_log_critical(mavlink_log_pub, "No accels found");
return PX4_ERROR;
}
// Collect: As defined by configuration
// start with a full mask, all six bits set
int32_t cal_mask = (1 << 6) - 1;
param_get(param_find("CAL_ACC_SIDES"), &cal_mask);
calibration_log_info(mavlink_log_pub, CAL_QGC_STARTED_MSG, sensor_name);
calibration::Accelerometer calibrations[MAX_ACCEL_SENS] {};
@@ -370,83 +357,222 @@ int do_accel_calibration(orb_advert_t *mavlink_log_pub)
/* measure and calculate offsets & scales */
accel_worker_data_s worker_data{};
worker_data.mavlink_log_pub = mavlink_log_pub;
bool data_collected[detect_orientation_side_count] {};
if (calibrate_from_orientation(mavlink_log_pub, data_collected, accel_calibration_worker, &worker_data,
false) == calibrate_return_ok) {
for (unsigned i = 0; i < (sizeof(worker_data.side_data_collected) / sizeof(worker_data.side_data_collected[0])); i++) {
if ((cal_mask & (1 << i)) > 0) {
// mark as missing
worker_data.side_data_collected[i] = false;
worker_data.calibration_sides++;
const Dcmf board_rotation = calibration::GetBoardRotationMatrix();
const Dcmf board_rotation_t = board_rotation.transpose();
} else {
// mark as completed from the beginning
worker_data.side_data_collected[i] = true;
calibration_log_info(mavlink_log_pub,
"[cal] %s side done, rotate to a different side",
detect_orientation_str(static_cast<enum detect_orientation_return>(i)));
px4_usleep(100000);
}
}
if (calibrate_from_orientation(mavlink_log_pub, worker_data.side_data_collected, accel_calibration_worker,
&worker_data) == calibrate_return_ok) {
const Rotation board_rotation = calibration::GetBoardRotation();
bool param_save = false;
bool failed = true;
for (unsigned i = 0; i < MAX_ACCEL_SENS; i++) {
if (i < active_sensors) {
// calculate offsets
Vector3f offset{};
int internal_rotation_count{0};
Rotation internal_rotations[MAX_ACCEL_SENS] {};
// X offset: average X from TAIL_DOWN + NOSE_DOWN
const Vector3f accel_tail_down{worker_data.accel_ref[i][ORIENTATION_TAIL_DOWN]};
const Vector3f accel_nose_down{worker_data.accel_ref[i][ORIENTATION_NOSE_DOWN]};
offset(0) = (accel_tail_down(0) + accel_nose_down(0)) * 0.5f;
for (uint8_t cur_accel = 0; cur_accel < active_sensors; cur_accel++) {
int accel_data_count = 0;
// Y offset: average Y from LEFT + RIGHT
const Vector3f accel_left{worker_data.accel_ref[i][ORIENTATION_LEFT]};
const Vector3f accel_right{worker_data.accel_ref[i][ORIENTATION_RIGHT]};
offset(1) = (accel_left(1) + accel_right(1)) * 0.5f;
matrix::Vector3f accel_data[6] {};
// Z offset: average Z from UPSIDE_DOWN + RIGHTSIDE_UP
const Vector3f accel_upside_down{worker_data.accel_ref[i][ORIENTATION_UPSIDE_DOWN]};
const Vector3f accel_rightside_up{worker_data.accel_ref[i][ORIENTATION_RIGHTSIDE_UP]};
offset(2) = (accel_upside_down(2) + accel_rightside_up(2)) * 0.5f;
for (int orientation = 0; orientation < 6; orientation++) {
if ((cal_mask & (1 << orientation)) > 0) {
accel_data[accel_data_count] = worker_data.accel_ref[cur_accel][orientation];
accel_data_count++;
}
}
// transform matrix
Matrix3f mat_A;
mat_A.row(0) = accel_tail_down - offset;
mat_A.row(1) = accel_left - offset;
mat_A.row(2) = accel_upside_down - offset;
bool sphere_fit_only = false;
bool sphere_fit_success = false;
bool ellipsoid_fit_success = false;
// calculate inverse matrix for A: simplify matrices mult because b has only one non-zero element == g at index i
const Matrix3f accel_T = mat_A.I() * CONSTANTS_ONE_G;
sphere_params sphere_data{};
sphere_data.radius = CONSTANTS_ONE_G;
int ret = lm_fit(worker_data.accel_ref[cur_accel], 6, sphere_data, false);
// update calibration
const Vector3f accel_offs_rotated{board_rotation_t *offset};
calibrations[i].set_offset(accel_offs_rotated);
if (ret == PX4_OK) {
sphere_fit_success = true;
const Matrix3f accel_T_rotated{board_rotation_t *accel_T * board_rotation};
calibrations[i].set_scale(accel_T_rotated.diag());
if (!sphere_fit_only) {
int ellipsoid_ret = lm_fit(accel_data, accel_data_count, sphere_data, true);
calibrations[i].set_temperature(worker_data.accel_temperature_ref[i]);
if (ellipsoid_ret == PX4_OK) {
ellipsoid_fit_success = true;
}
}
PX4_INFO("Accel: %" PRIu8 " sphere radius: %.4f", cur_accel, (double)sphere_data.radius);
}
static constexpr float scale_min = 0.9f;
static constexpr float scale_max = 1.1f;
for (int axis = 0; axis < 3; axis++) {
if ((sphere_data.diag(axis) < scale_min) || (sphere_data.diag(axis) > scale_max)) {
PX4_WARN("bad scale factor"); // TODO, abort and report failure
}
}
bool calibration_valid = sphere_fit_success || ellipsoid_fit_success; // TODO: scale range, offset range, etc
#if defined(DEBUD_BUILD)
PX4_INFO("accel %d: offset", i);
offset.print();
PX4_INFO("accel %d: bT * offset", i);
accel_offs_rotated.print();
PX4_INFO("accel %d offset", cur_accel);
sphere_data.offset.print();
PX4_INFO("accel %d: mat_A", i);
mat_A.print();
PX4_INFO("accel %d: accel_T", i);
accel_T.print();
PX4_INFO("accel %d: bT * accel_T * b", i);
accel_T_rotated.print();
PX4_INFO("accel %d scale", cur_accel);
sphere_data.diag.print();
PX4_INFO("accel %d offdiagonal", cur_accel);
sphere_data.offdiag.print();
#endif // DEBUD_BUILD
calibrations[i].PrintStatus();
if (!calibration_valid) {
failed = true;
break;
}
// determine best rotation
float calibration_metric[ROTATION_MAX] {};
float min_error = FLT_MAX;
Rotation best_rotation = ROTATION_NONE;
for (int r = ROTATION_NONE; r < ROTATION_MAX; r++) {
calibration_metric[r] = FLT_MAX;
// try all rotations
switch (r) {
case ROTATION_ROLL_90_PITCH_68_YAW_293: // skip
case ROTATION_PITCH_180_YAW_90: // skip 26, same as 14 ROTATION_ROLL_180_YAW_270
case ROTATION_PITCH_180_YAW_270: // skip 27, same as 10 ROTATION_ROLL_180_YAW_90
break;
default: {
matrix::Matrix<float, 6, 3> Y;
static constexpr float g = CONSTANTS_ONE_G;
Y.row(0) = Vector3f{ g, 0, 0}; // ORIENTATION_TAIL_DOWN
Y.row(1) = Vector3f{-g, 0, 0}; // ORIENTATION_NOSE_DOWN,
Y.row(2) = Vector3f{ 0, g, 0}; // ORIENTATION_LEFT,
Y.row(3) = Vector3f{ 0, -g, 0}; // ORIENTATION_RIGHT,
Y.row(4) = Vector3f{ 0, 0, g}; // ORIENTATION_UPSIDE_DOWN,
Y.row(5) = Vector3f{ 0, 0, -g}; // ORIENTATION_RIGHTSIDE_UP
// apply calibration and compare data with expected direction
for (int orientation = 0; orientation < 6; orientation++) {
if ((cal_mask & (1 << orientation)) > 0) {
Vector3f calibrated_accel{sphere_data.diag.emult(worker_data.accel_ref[cur_accel][orientation] - sphere_data.offset)};
const Vector3f rotated_data{get_rot_matrix((Rotation)r) *calibrated_accel};
calibration_metric[r] += (rotated_data - Y.row(orientation)).norm();
}
}
}
}
if (calibrations[i].ParametersSave()) {
param_save = true;
failed = false;
if (calibration_metric[r] < min_error) {
min_error = calibration_metric[r];
best_rotation = (Rotation)r;
}
} else {
failed = true;
calibration_log_critical(mavlink_log_pub, "calibration save failed");
}
PX4_INFO("best rotation: %d", best_rotation);
if (!calibrations[cur_accel].external()) {
internal_rotations[internal_rotation_count] = best_rotation;
internal_rotation_count++;
}
bool print_all_errors = true;
if (calibrations[cur_accel].external()) {
switch (calibrations[cur_accel].rotation_enum()) {
case ROTATION_ROLL_90_PITCH_68_YAW_293:
PX4_INFO("[cal] External Accel: %" PRIu8 " (%" PRIu32 "), keeping manually configured rotation %" PRIu8,
cur_accel, calibrations[cur_accel].device_id(), calibrations[cur_accel].rotation_enum());
continue;
default: {
if (best_rotation != calibrations[cur_accel].rotation_enum()) {
calibration_log_info(mavlink_log_pub, "[cal] External Accel: %" PRIu8 " (%" PRIu32 "), determined rotation: %" PRIu8,
cur_accel, calibrations[cur_accel].device_id(), best_rotation);
calibrations[cur_accel].set_rotation(best_rotation);
} else {
PX4_INFO("[cal] External Accel: %" PRIu8 " (%" PRIu32 "), no rotation change: %" PRIu8,
cur_accel, calibrations[cur_accel].device_id(), best_rotation);
}
}
break;
}
}
if (print_all_errors) {
for (int r = ROTATION_NONE; r < ROTATION_MAX; r++) {
if (calibration_metric[r] < FLT_MAX) {
PX4_ERR("Accel: %" PRIu8 " (%" PRIu32 "), rotation: %" PRIu32 ", error: %.3f",
cur_accel, calibrations[cur_accel].device_id(), (uint32_t)r, (double)calibration_metric[r]);
}
}
}
// update calibration
calibrations[cur_accel].set_offset(sphere_data.offset);
calibrations[cur_accel].set_scale(sphere_data.diag);
calibrations[cur_accel].set_temperature(worker_data.accel_temperature_ref[cur_accel]);
calibrations[cur_accel].PrintStatus();
// save all calibrations including empty slots
if (calibrations[cur_accel].ParametersSave()) {
param_save = true;
failed = false;
} else {
failed = true;
calibration_log_critical(mavlink_log_pub, "calibration save failed");
break;
}
}
// review SENS_BOARD_ROT, do all internal accels agree on SENS_BOARD_ROT?
if (!failed && (internal_rotation_count > 0)) {
Rotation best_rotation = internal_rotations[0];
bool same = true;
for (unsigned i = 0; i < MAX_ACCEL_SENS; i++) {
if (internal_rotations[0] != internal_rotations[i]) {
same = false;
break;
}
}
if (same && (best_rotation != board_rotation)) {
PX4_ERR("Incorrect board rotation: %d", board_rotation);
int32_t sens_board_rot = best_rotation;
param_set_no_notification(param_find("SENS_BOARD_ROT"), &sens_board_rot);
}
}
if (!failed && factory_storage.store() != PX4_OK) {
failed = true;
}
@@ -472,7 +598,6 @@ int do_accel_calibration_quick(orb_advert_t *mavlink_log_pub)
#if !defined(CONSTRAINED_FLASH)
PX4_INFO("Accelerometer quick calibration");
bool param_save = false;
bool failed = true;
FactoryCalibrationStorage factory_storage;
@@ -482,135 +607,113 @@ int do_accel_calibration_quick(orb_advert_t *mavlink_log_pub)
return PX4_ERROR;
}
// sensor thermal corrections (optional)
uORB::Subscription sensor_correction_sub{ORB_ID(sensor_correction)};
sensor_correction_s sensor_correction{};
sensor_correction_sub.copy(&sensor_correction);
calibration::Accelerometer calibrations[MAX_ACCEL_SENS] {};
math::WelfordMean<Vector3f> mean[MAX_ACCEL_SENS] {};
uORB::SubscriptionMultiArray<sensor_accel_s, MAX_ACCEL_SENS> accel_subs{ORB_ID::sensor_accel};
px4::Bitset<MAX_ACCEL_SENS> valid_cal;
/* use the first sensor to pace the readout, but do per-sensor counts */
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
sensor_accel_s arp{};
Vector3f accel_sum{};
float temperature_sum{NAN};
unsigned count = 0;
const hrt_abstime start_time = hrt_absolute_time();
while (accel_subs[accel_index].update(&arp)) {
// fetch optional thermal offset corrections in sensor/board frame
if ((arp.timestamp > 0) && (arp.device_id != 0)) {
Vector3f offset{0, 0, 0};
Vector3f accel_prev[MAX_ACCEL_SENS] {};
if (sensor_correction.timestamp > 0) {
for (uint8_t correction_index = 0; correction_index < MAX_ACCEL_SENS; correction_index++) {
if (sensor_correction.accel_device_ids[correction_index] == arp.device_id) {
switch (correction_index) {
case 0:
offset = Vector3f{sensor_correction.accel_offset_0};
break;
case 1:
offset = Vector3f{sensor_correction.accel_offset_1};
break;
case 2:
offset = Vector3f{sensor_correction.accel_offset_2};
break;
case 3:
offset = Vector3f{sensor_correction.accel_offset_3};
break;
}
}
}
}
while ((hrt_elapsed_time(&start_time) < 3_s) && (valid_cal.count() < accel_subs.advertised_count())) {
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
sensor_accel_s sensor_accel;
const Vector3f accel{Vector3f{arp.x, arp.y, arp.z} - offset};
while (accel_subs[accel_index].update(&sensor_accel)) {
if ((sensor_accel.timestamp > 0) && (sensor_accel.device_id != 0)) {
calibrations[accel_index].set_device_id(sensor_accel.device_id);
calibrations[accel_index].SensorCorrectionsUpdate();
if (count > 0) {
const Vector3f diff{accel - (accel_sum / count)};
const Vector3f accel{Vector3f{sensor_accel.x, sensor_accel.y, sensor_accel.z} - calibrations[accel_index].thermal_offset()};
if (diff.norm() < 1.f) {
accel_sum += Vector3f{arp.x, arp.y, arp.z} - offset;
if ((accel - accel_prev[accel_index]).longerThan(0.5f)) {
mean[accel_index].reset();
accel_prev[accel_index] = accel;
count++;
if (!PX4_ISFINITE(temperature_sum)) {
// set first valid value
temperature_sum = (arp.temperature * count);
} else {
temperature_sum += arp.temperature;
}
} else {
mean[accel_index].update(accel);
}
} else {
accel_sum = accel;
temperature_sum = arp.temperature;
count = 1;
if (mean[accel_index].count() > 300 && !mean[accel_index].variance().longerThan(0.01f)) {
valid_cal.set(accel_index, true);
}
}
}
}
if ((count > 0) && (arp.device_id != 0)) {
px4_usleep(1000);
bool calibrated = false;
const Vector3f accel_avg = accel_sum / count;
const float temperature_avg = temperature_sum / count;
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
if (mean[accel_index].count() > 0) {
PX4_DEBUG("accel %d/%d valid %d var: %.6f", accel_index, accel_subs.advertised_count(), mean[accel_index].count(),
(double)mean[accel_index].variance().length());
}
}
}
bool calibration_updated = false;
px4::Bitset<MAX_ACCEL_SENS> calibrated;
static constexpr float g = CONSTANTS_ONE_G;
matrix::Matrix<float, 6, 3> Y;
Y.row(0) = Vector3f{ g, 0, 0}; // ORIENTATION_TAIL_DOWN
Y.row(1) = Vector3f{-g, 0, 0}; // ORIENTATION_NOSE_DOWN,
Y.row(2) = Vector3f{ 0, g, 0}; // ORIENTATION_LEFT,
Y.row(3) = Vector3f{ 0, -g, 0}; // ORIENTATION_RIGHT,
Y.row(4) = Vector3f{ 0, 0, g}; // ORIENTATION_UPSIDE_DOWN,
Y.row(5) = Vector3f{ 0, 0, -g}; // ORIENTATION_RIGHTSIDE_UP
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
if (valid_cal[accel_index]) {
const Vector3f accel_avg = mean[accel_index].mean();
int closest_orientation = -1;
float smallest_orientation_error = INFINITY;
for (int i = 0; i < 6; i++) {
// assume sensor/board is in one of these orientations
float accel_error = (Vector3f{Y.row(i)} - accel_avg).length();
if (accel_error < smallest_orientation_error) {
closest_orientation = i;
smallest_orientation_error = accel_error;
}
}
Vector3f offset{0.f, 0.f, 0.f};
uORB::SubscriptionData<vehicle_attitude_s> attitude_sub{ORB_ID(vehicle_attitude)};
attitude_sub.update();
if (closest_orientation != -1) {
PX4_INFO("assuming accel %d is orientation %s (%d)", accel_index,
detect_orientation_str((enum detect_orientation_return)closest_orientation),
closest_orientation);
if (attitude_sub.advertised() && attitude_sub.get().timestamp != 0) {
// use vehicle_attitude if available
const vehicle_attitude_s &att = attitude_sub.get();
const matrix::Quatf q{att.q};
const Vector3f accel_ref = q.conjugate_inversed(Vector3f{0.f, 0.f, -CONSTANTS_ONE_G});
// sanity check angle between acceleration vectors
const float angle = AxisAnglef(Quatf(accel_avg, accel_ref)).angle();
if (angle <= math::radians(10.f)) {
offset = accel_avg - accel_ref;
calibrated = true;
}
offset = accel_avg - (calibrations[accel_index].scale().emult(Vector3f{Y.row(closest_orientation)}));
calibrated.set(accel_index, true);
}
if (!calibrated) {
// otherwise simply normalize to gravity and remove offset
Vector3f accel{accel_avg};
accel.normalize();
accel = accel * CONSTANTS_ONE_G;
const Vector3f accel_avg_calibrated = calibrations[accel_index].scale().emult(accel_avg - offset);
offset = accel_avg - accel;
calibrated = true;
}
calibration::Accelerometer calibration{arp.device_id};
// reset cal index to uORB
calibration.set_calibration_index(accel_index);
if (!calibrated || (offset.norm() > CONSTANTS_ONE_G)
|| !PX4_ISFINITE(offset(0))
|| !PX4_ISFINITE(offset(1))
|| !PX4_ISFINITE(offset(2))) {
if (!calibrated[accel_index] || !accel_avg_calibrated.longerThan(CONSTANTS_ONE_G * 0.9f)
|| accel_avg_calibrated.longerThan(CONSTANTS_ONE_G * 1.1f)
|| !PX4_ISFINITE(offset(0)) || !PX4_ISFINITE(offset(1)) || !PX4_ISFINITE(offset(2))) {
PX4_ERR("accel %d quick calibrate failed", accel_index);
failed = true;
break;
} else {
calibration.set_offset(offset);
calibration.set_temperature(temperature_avg);
// reset cal index to uORB
calibrations[accel_index].set_calibration_index(accel_index);
if (calibration.ParametersSave()) {
calibration.PrintStatus();
param_save = true;
failed = false;
} else {
failed = true;
calibration_log_critical(mavlink_log_pub, CAL_QGC_FAILED_MSG, "calibration save failed");
break;
if (calibrations[accel_index].set_offset(offset)) {
calibrations[accel_index].PrintStatus();
calibration_updated = true;
}
failed = false;
}
}
}
@@ -619,15 +722,29 @@ int do_accel_calibration_quick(orb_advert_t *mavlink_log_pub)
failed = true;
}
if (param_save) {
param_notify_changes();
}
if (!failed) {
if (calibration_updated) {
bool param_save = false;
for (unsigned accel_index = 0; accel_index < MAX_ACCEL_SENS; accel_index++) {
if (calibrated[accel_index]) {
if (calibrations[accel_index].ParametersSave()) {
param_save = true;
}
}
}
if (param_save) {
param_notify_changes();
}
}
calibration_log_info(mavlink_log_pub, CAL_QGC_DONE_MSG, sensor_name);
return PX4_OK;
}
#endif // !CONSTRAINED_FLASH
calibration_log_critical(mavlink_log_pub, CAL_QGC_FAILED_MSG, sensor_name);
return PX4_ERROR;
}
@@ -63,17 +63,16 @@
using namespace time_literals;
enum detect_orientation_return detect_orientation(orb_advert_t *mavlink_log_pub, bool lenient_still_position)
enum detect_orientation_return detect_orientation(orb_advert_t *mavlink_log_pub)
{
static constexpr unsigned ndim = 3;
float accel_ema[ndim] {}; // exponential moving average of accel
float accel_disp[3] {}; // max-hold dispersion of accel
static constexpr float ema_len = 0.5f; // EMA time constant in seconds
static constexpr float normal_still_thr = 0.25; // normal still threshold
float still_thr2 = powf(lenient_still_position ? (normal_still_thr * 3) : normal_still_thr, 2);
float still_thr2 = powf(0.25f * 3, 2);
static constexpr float accel_err_thr = 5.0f; // set accel error threshold to 5m/s^2
const hrt_abstime still_time = lenient_still_position ? 500000 : 1300000; // still time required in us
static constexpr hrt_abstime still_time = 500000;
/* set timeout to 90s */
static constexpr hrt_abstime timeout = 90_s;
@@ -219,7 +218,7 @@ const char *detect_orientation_str(enum detect_orientation_return orientation)
calibrate_return calibrate_from_orientation(orb_advert_t *mavlink_log_pub,
bool side_data_collected[detect_orientation_side_count], calibration_from_orientation_worker_t calibration_worker,
void *worker_data, bool lenient_still_position)
void *worker_data)
{
const hrt_abstime calibration_started = hrt_absolute_time();
calibrate_return result = calibrate_return_ok;
@@ -268,7 +267,7 @@ calibrate_return calibrate_from_orientation(orb_advert_t *mavlink_log_pub,
px4_usleep(20000);
calibration_log_info(mavlink_log_pub, "[cal] hold vehicle still on a pending side");
px4_usleep(20000);
enum detect_orientation_return orient = detect_orientation(mavlink_log_pub, lenient_still_position);
enum detect_orientation_return orient = detect_orientation(mavlink_log_pub);
if (orient == ORIENTATION_ERROR) {
orientation_failures++;
+4 -7
View File
@@ -54,8 +54,7 @@ static constexpr unsigned detect_orientation_side_count = 6;
/// Wait for vehicle to become still and detect it's orientation
/// @return Returns detect_orientation_return according to orientation when vehicle
/// and ready for measurements
enum detect_orientation_return detect_orientation(orb_advert_t *mavlink_log_pub, ///< uORB handle to write output to
bool lenient_still_detection); ///< true: Use more lenient still position detection
enum detect_orientation_return detect_orientation(orb_advert_t *mavlink_log_pub);
/// Returns the human readable string representation of the orientation
/// @param orientation Orientation to return string for, "error" if buffer is too small
@@ -67,17 +66,15 @@ enum calibrate_return {
calibrate_return_cancelled
};
typedef calibrate_return(*calibration_from_orientation_worker_t)(detect_orientation_return
orientation, ///< Orientation which was detected
void *worker_data); ///< Opaque worker data
typedef calibrate_return(*calibration_from_orientation_worker_t)(detect_orientation_return orientation,
void *worker_data);
/// Perform calibration sequence which require a rest orientation detection prior to calibration.
/// @return OK: Calibration succeeded, ERROR: Calibration failed
calibrate_return calibrate_from_orientation(orb_advert_t *mavlink_log_pub, ///< uORB handle to write output to
bool side_data_collected[detect_orientation_side_count], ///< Sides for which data still needs calibration
calibration_from_orientation_worker_t calibration_worker, ///< Worker routine which performs the actual calibration
void *worker_data, ///< Opaque data passed to worker routine
bool lenient_still_detection); ///< true: Use more lenient still position detection
void *worker_data); ///< Opaque data passed to worker routine
/// Used to periodically check for a cancel command
bool calibrate_cancel_check(orb_advert_t *mavlink_log_pub, const hrt_abstime &calibration_started);
+141 -133
View File
@@ -41,43 +41,43 @@ struct iteration_result {
} result = STATUS::SUCCESS;
};
void lm_sphere_fit_iteration(const float x[], const float y[], const float z[],
unsigned int samples_collected, sphere_params &params, iteration_result &result)
static void lm_sphere_fit_iteration(const matrix::Vector3f data[], unsigned samples_collected, sphere_params &p,
iteration_result &result)
{
// Run Sphere Fit using Levenberg Marquardt LSq Fit
const float lma_damping = 10.f;
float fitness = result.cost;
float fit1 = 0.f;
float fit2 = 0.f;
matrix::SquareMatrix<float, 4> JTJ;
float JTFI[4] {};
float residual = 0.0f;
// Gauss Newton Part common for all kind of extensions including LM
for (uint16_t k = 0; k < samples_collected; k++) {
for (unsigned k = 0; k < samples_collected; k++) {
float x = data[k](0);
float y = data[k](1);
float z = data[k](2);
// Calculate Jacobian
float A = (p.diag(0) * (x - p.offset(0))) + (p.offdiag(0) * (y - p.offset(1))) + (p.offdiag(1) * (z - p.offset(2)));
float B = (p.offdiag(0) * (x - p.offset(0))) + (p.diag(1) * (y - p.offset(1))) + (p.offdiag(2) * (z - p.offset(2)));
float C = (p.offdiag(1) * (x - p.offset(0))) + (p.offdiag(2) * (y - p.offset(1))) + (p.diag(2) * (z - p.offset(2)));
float sphere_jacob[4];
//Calculate Jacobian
float A = (params.diag(0) * (x[k] - params.offset(0))) + (params.offdiag(0) * (y[k] - params.offset(1))) +
(params.offdiag(1) * (z[k] - params.offset(2)));
float B = (params.offdiag(0) * (x[k] - params.offset(0))) + (params.diag(1) * (y[k] - params.offset(1))) +
(params.offdiag(2) * (z[k] - params.offset(2)));
float C = (params.offdiag(1) * (x[k] - params.offset(0))) + (params.offdiag(2) * (y[k] - params.offset(1))) +
(params.diag(2) * (z[k] - params.offset(2)));
float length = sqrtf(A * A + B * B + C * C);
float sphere_jacob[4];
// 0: partial derivative (radius wrt fitness fn) fn operated on sample
sphere_jacob[0] = 1.0f;
sphere_jacob[0] = 1.f;
// 1-3: partial derivative (offsets wrt fitness fn) fn operated on sample
sphere_jacob[1] = 1.0f * (((params.diag(0) * A) + (params.offdiag(0) * B) + (params.offdiag(1) * C)) / length);
sphere_jacob[2] = 1.0f * (((params.offdiag(0) * A) + (params.diag(1) * B) + (params.offdiag(2) * C)) / length);
sphere_jacob[3] = 1.0f * (((params.offdiag(1) * A) + (params.offdiag(2) * B) + (params.diag(2) * C)) / length);
residual = params.radius - length;
sphere_jacob[1] = 1.f * (((p.diag(0) * A) + (p.offdiag(0) * B) + (p.offdiag(1) * C)) / length);
sphere_jacob[2] = 1.f * (((p.offdiag(0) * A) + (p.diag(1) * B) + (p.offdiag(2) * C)) / length);
sphere_jacob[3] = 1.f * (((p.offdiag(1) * A) + (p.offdiag(2) * B) + (p.diag(2) * C)) / length);
for (uint8_t i = 0; i < 4; i++) {
float residual = p.radius - length;
for (int i = 0; i < 4; i++) {
// compute JTJ
for (uint8_t j = 0; j < 4; j++) {
for (int j = 0; j < 4; j++) {
JTJ(i, j) += sphere_jacob[i] * sphere_jacob[j];
}
@@ -87,12 +87,12 @@ void lm_sphere_fit_iteration(const float x[], const float y[], const float z[],
//------------------------Levenberg-Marquardt-part-starts-here---------------------------------//
// refer: http://en.wikipedia.org/wiki/Levenberg%E2%80%93Marquardt_algorithm#Choice_of_damping_parameter
float fit1_params[4] = {params.radius, params.offset(0), params.offset(1), params.offset(2)};
float fit2_params[4];
memcpy(fit2_params, fit1_params, sizeof(fit1_params));
// refer: https://en.wikipedia.org/wiki/Levenberg%E2%80%93Marquardt_algorithm#Choice_of_damping_parameter
float fit1_p[4] {p.radius, p.offset(0), p.offset(1), p.offset(2)};
float fit2_p[4] {p.radius, p.offset(0), p.offset(1), p.offset(2)};
JTJ = (JTJ + JTJ.transpose()) * 0.5f; // fix numerical issues
matrix::SquareMatrix<float, 4> JTJ2 = JTJ;
matrix::SquareMatrix<float, 4> JTJ2{JTJ};
for (uint8_t i = 0; i < 4; i++) {
JTJ(i, i) += result.gradient_damping;
@@ -107,42 +107,39 @@ void lm_sphere_fit_iteration(const float x[], const float y[], const float z[],
if (!JTJ2.I(JTJ2)) {
result.result = iteration_result::STATUS::FAILURE;
return;
}
for (uint8_t row = 0; row < 4; row++) {
for (uint8_t col = 0; col < 4; col++) {
fit1_params[row] -= JTFI[col] * JTJ(row, col);
fit2_params[row] -= JTFI[col] * JTJ2(row, col);
for (int row = 0; row < 4; row++) {
for (int col = 0; col < 4; col++) {
fit1_p[row] -= JTFI[col] * JTJ(row, col);
fit2_p[row] -= JTFI[col] * JTJ2(row, col);
}
}
// Calculate mean squared residuals
for (uint16_t k = 0; k < samples_collected; k++) {
float A = (params.diag(0) * (x[k] - fit1_params[1])) + (params.offdiag(0) * (y[k] - fit1_params[2])) +
(params.offdiag(1) *
(z[k] + fit1_params[3]));
float B = (params.offdiag(0) * (x[k] - fit1_params[1])) + (params.diag(1) * (y[k] - fit1_params[2])) +
(params.offdiag(2) *
(z[k] + fit1_params[3]));
float C = (params.offdiag(1) * (x[k] - fit1_params[1])) + (params.offdiag(2) * (y[k] - fit1_params[2])) + (params.diag(
2) *
(z[k] - fit1_params[3]));
for (unsigned k = 0; k < samples_collected; k++) {
float x = data[k](0);
float y = data[k](1);
float z = data[k](2);
float A = (p.diag(0) * (x - fit1_p[1])) + (p.offdiag(0) * (y - fit1_p[2])) + (p.offdiag(1) * (z + fit1_p[3]));
float B = (p.offdiag(0) * (x - fit1_p[1])) + (p.diag(1) * (y - fit1_p[2])) + (p.offdiag(2) * (z + fit1_p[3]));
float C = (p.offdiag(1) * (x - fit1_p[1])) + (p.offdiag(2) * (y - fit1_p[2])) + (p.diag(2) * (z - fit1_p[3]));
float length = sqrtf(A * A + B * B + C * C);
residual = fit1_params[0] - length;
float residual = fit1_p[0] - length;
fit1 += residual * residual;
A = (params.diag(0) * (x[k] - fit2_params[1])) + (params.offdiag(0) * (y[k] - fit2_params[2])) + (params.offdiag(1) *
(z[k] - fit2_params[3]));
B = (params.offdiag(0) * (x[k] - fit2_params[1])) + (params.diag(1) * (y[k] - fit2_params[2])) + (params.offdiag(2) *
(z[k] - fit2_params[3]));
C = (params.offdiag(1) * (x[k] - fit2_params[1])) + (params.offdiag(2) * (y[k] - fit2_params[2])) + (params.diag(2) *
(z[k] - fit2_params[3]));
A = (p.diag(0) * (x - fit2_p[1])) + (p.offdiag(0) * (y - fit2_p[2])) + (p.offdiag(1) * (z - fit2_p[3]));
B = (p.offdiag(0) * (x - fit2_p[1])) + (p.diag(1) * (y - fit2_p[2])) + (p.offdiag(2) * (z - fit2_p[3]));
C = (p.offdiag(1) * (x - fit2_p[1])) + (p.offdiag(2) * (y - fit2_p[2])) + (p.diag(2) * (z - fit2_p[3]));
length = sqrtf(A * A + B * B + C * C);
residual = fit2_params[0] - length;
residual = fit2_p[0] - length;
fit2 += residual * residual;
}
float fitness = result.cost;
fit1 = sqrtf(fit1) / samples_collected;
fit2 = sqrtf(fit2) / samples_collected;
@@ -151,7 +148,7 @@ void lm_sphere_fit_iteration(const float x[], const float y[], const float z[],
} else if (fit2 < result.cost && fit2 < fit1) {
result.gradient_damping /= lma_damping;
memcpy(fit1_params, fit2_params, sizeof(fit1_params));
memcpy(fit1_p, fit2_p, sizeof(fit1_p));
fitness = fit2;
} else if (fit1 < result.cost) {
@@ -159,13 +156,15 @@ void lm_sphere_fit_iteration(const float x[], const float y[], const float z[],
}
//--------------------Levenberg-Marquardt-part-ends-here--------------------------------//
if (PX4_ISFINITE(fitness) && fitness <= result.cost) {
result.cost = fitness;
params.radius = fit1_params[0];
params.offset(0) = fit1_params[1];
params.offset(1) = fit1_params[2];
params.offset(2) = fit1_params[3];
p.radius = fit1_p[0];
p.offset(0) = fit1_p[1];
p.offset(1) = fit1_p[2];
p.offset(2) = fit1_p[3];
result.result = iteration_result::STATUS::SUCCESS;
} else {
@@ -173,49 +172,49 @@ void lm_sphere_fit_iteration(const float x[], const float y[], const float z[],
}
}
void lm_ellipsoid_fit_iteration(const float x[], const float y[], const float z[],
unsigned int samples_collected, sphere_params &params, iteration_result &result)
static void lm_ellipsoid_fit_iteration(const matrix::Vector3f data[], unsigned samples_collected, sphere_params &p,
iteration_result &result)
{
// Run Sphere Fit using Levenberg Marquardt LSq Fit
const float lma_damping = 10.0f;
float fitness = result.cost;
float fit1 = 0.0f;
float fit2 = 0.0f;
const float lma_damping = 10.f;
matrix::SquareMatrix<float, 9> JTJ;
float fit1 = 0.f;
float fit2 = 0.f;
matrix::SquareMatrix<float, 9> JTJ{};
float JTFI[9] {};
float residual = 0.0f;
float ellipsoid_jacob[9];
// Gauss Newton Part common for all kind of extensions including LM
for (uint16_t k = 0; k < samples_collected; k++) {
for (unsigned k = 0; k < samples_collected; k++) {
float x = data[k](0);
float y = data[k](1);
float z = data[k](2);
// Calculate Jacobian
float A = (params.diag(0) * (x[k] - params.offset(0))) + (params.offdiag(0) * (y[k] - params.offset(1))) +
(params.offdiag(1) * (z[k] - params.offset(2)));
float B = (params.offdiag(0) * (x[k] - params.offset(0))) + (params.diag(1) * (y[k] - params.offset(1))) +
(params.offdiag(2) * (z[k] - params.offset(2)));
float C = (params.offdiag(1) * (x[k] - params.offset(0))) + (params.offdiag(2) * (y[k] - params.offset(1))) +
(params.diag(2) * (z[k] - params.offset(2)));
float A = (p.diag(0) * (x - p.offset(0))) + (p.offdiag(0) * (y - p.offset(1))) + (p.offdiag(1) * (z - p.offset(2)));
float B = (p.offdiag(0) * (x - p.offset(0))) + (p.diag(1) * (y - p.offset(1))) + (p.offdiag(2) * (z - p.offset(2)));
float C = (p.offdiag(1) * (x - p.offset(0))) + (p.offdiag(2) * (y - p.offset(1))) + (p.diag(2) * (z - p.offset(2)));
float length = sqrtf(A * A + B * B + C * C);
residual = params.radius - length;
float residual = p.radius - length;
fit1 += residual * residual;
// 0-2: partial derivative (offset wrt fitness fn) fn operated on sample
ellipsoid_jacob[0] = 1.0f * (((params.diag(0) * A) + (params.offdiag(0) * B) + (params.offdiag(1) * C)) / length);
ellipsoid_jacob[1] = 1.0f * (((params.offdiag(0) * A) + (params.diag(1) * B) + (params.offdiag(2) * C)) / length);
ellipsoid_jacob[2] = 1.0f * (((params.offdiag(1) * A) + (params.offdiag(2) * B) + (params.diag(2) * C)) / length);
// 3-5: partial derivative (diag offset wrt fitness fn) fn operated on sample
ellipsoid_jacob[3] = -1.0f * ((x[k] - params.offset(0)) * A) / length;
ellipsoid_jacob[4] = -1.0f * ((y[k] - params.offset(1)) * B) / length;
ellipsoid_jacob[5] = -1.0f * ((z[k] - params.offset(2)) * C) / length;
// 6-8: partial derivative (off-diag offset wrt fitness fn) fn operated on sample
ellipsoid_jacob[6] = -1.0f * (((y[k] - params.offset(1)) * A) + ((x[k] - params.offset(0)) * B)) / length;
ellipsoid_jacob[7] = -1.0f * (((z[k] - params.offset(2)) * A) + ((x[k] - params.offset(0)) * C)) / length;
ellipsoid_jacob[8] = -1.0f * (((z[k] - params.offset(2)) * B) + ((y[k] - params.offset(1)) * C)) / length;
for (uint8_t i = 0; i < 9; i++) {
float ellipsoid_jacob[9];
// 0-2: partial derivative (offset wrt fitness fn) fn operated on sample
ellipsoid_jacob[0] = 1.f * (((p.diag(0) * A) + (p.offdiag(0) * B) + (p.offdiag(1) * C)) / length);
ellipsoid_jacob[1] = 1.f * (((p.offdiag(0) * A) + (p.diag(1) * B) + (p.offdiag(2) * C)) / length);
ellipsoid_jacob[2] = 1.f * (((p.offdiag(1) * A) + (p.offdiag(2) * B) + (p.diag(2) * C)) / length);
// 3-5: partial derivative (diag offset wrt fitness fn) fn operated on sample
ellipsoid_jacob[3] = -1.f * ((x - p.offset(0)) * A) / length;
ellipsoid_jacob[4] = -1.f * ((y - p.offset(1)) * B) / length;
ellipsoid_jacob[5] = -1.f * ((z - p.offset(2)) * C) / length;
// 6-8: partial derivative (off-diag offset wrt fitness fn) fn operated on sample
ellipsoid_jacob[6] = -1.f * (((y - p.offset(1)) * A) + ((x - p.offset(0)) * B)) / length;
ellipsoid_jacob[7] = -1.f * (((z - p.offset(2)) * A) + ((x - p.offset(0)) * C)) / length;
ellipsoid_jacob[8] = -1.f * (((z - p.offset(2)) * B) + ((y - p.offset(1)) * C)) / length;
for (int i = 0; i < 9; i++) {
// compute JTJ
for (uint8_t j = 0; j < 9; j++) {
for (int j = 0; j < 9; j++) {
JTJ(i, j) += ellipsoid_jacob[i] * ellipsoid_jacob[j];
}
@@ -226,12 +225,13 @@ void lm_ellipsoid_fit_iteration(const float x[], const float y[], const float z[
//------------------------Levenberg-Marquardt-part-starts-here---------------------------------//
// refer: http://en.wikipedia.org/wiki/Levenberg%E2%80%93Marquardt_algorithm#Choice_of_damping_parameter
float fit1_params[9] = {params.offset(0), params.offset(1), params.offset(2), params.diag(0), params.diag(1), params.diag(2), params.offdiag(0), params.offdiag(1), params.offdiag(2)};
float fit2_params[9];
memcpy(fit2_params, fit1_params, sizeof(fit1_params));
matrix::SquareMatrix<float, 9> JTJ2 = JTJ;
float fit1_p[9] {p.offset(0), p.offset(1), p.offset(2), p.diag(0), p.diag(1), p.diag(2), p.offdiag(0), p.offdiag(1), p.offdiag(2)};
float fit2_p[9];
memcpy(fit2_p, fit1_p, sizeof(fit1_p));
for (uint8_t i = 0; i < 9; i++) {
matrix::SquareMatrix<float, 9> JTJ2{JTJ};
for (int i = 0; i < 9; i++) {
JTJ(i, i) += result.gradient_damping;
JTJ2(i, i) += result.gradient_damping / lma_damping;
}
@@ -247,36 +247,36 @@ void lm_ellipsoid_fit_iteration(const float x[], const float y[], const float z[
return;
}
for (uint8_t row = 0; row < 9; row++) {
for (uint8_t col = 0; col < 9; col++) {
fit1_params[row] -= JTFI[col] * JTJ(row, col);
fit2_params[row] -= JTFI[col] * JTJ2(row, col);
for (int row = 0; row < 9; row++) {
for (int col = 0; col < 9; col++) {
fit1_p[row] -= JTFI[col] * JTJ(row, col);
fit2_p[row] -= JTFI[col] * JTJ2(row, col);
}
}
// Calculate mean squared residuals
for (uint16_t k = 0; k < samples_collected; k++) {
float A = (fit1_params[3] * (x[k] - fit1_params[0])) + (fit1_params[6] * (y[k] - fit1_params[1])) + (fit1_params[7] *
(z[k] - fit1_params[2]));
float B = (fit1_params[6] * (x[k] - fit1_params[0])) + (fit1_params[4] * (y[k] - fit1_params[1])) + (fit1_params[8] *
(z[k] - fit1_params[2]));
float C = (fit1_params[7] * (x[k] - fit1_params[0])) + (fit1_params[8] * (y[k] - fit1_params[1])) + (fit1_params[5] *
(z[k] - fit1_params[2]));
for (unsigned k = 0; k < samples_collected; k++) {
float x = data[k](0);
float y = data[k](1);
float z = data[k](2);
float A = (fit1_p[3] * (x - fit1_p[0])) + (fit1_p[6] * (y - fit1_p[1])) + (fit1_p[7] * (z - fit1_p[2]));
float B = (fit1_p[6] * (x - fit1_p[0])) + (fit1_p[4] * (y - fit1_p[1])) + (fit1_p[8] * (z - fit1_p[2]));
float C = (fit1_p[7] * (x - fit1_p[0])) + (fit1_p[8] * (y - fit1_p[1])) + (fit1_p[5] * (z - fit1_p[2]));
float length = sqrtf(A * A + B * B + C * C);
residual = params.radius - length;
float residual = p.radius - length;
fit1 += residual * residual;
A = (fit2_params[3] * (x[k] - fit2_params[0])) + (fit2_params[6] * (y[k] - fit2_params[1])) + (fit2_params[7] *
(z[k] - fit2_params[2]));
B = (fit2_params[6] * (x[k] - fit2_params[0])) + (fit2_params[4] * (y[k] - fit2_params[1])) + (fit2_params[8] *
(z[k] - fit2_params[2]));
C = (fit2_params[7] * (x[k] - fit2_params[0])) + (fit2_params[8] * (y[k] - fit2_params[1])) + (fit2_params[5] *
(z[k] - fit2_params[2]));
A = (fit2_p[3] * (x - fit2_p[0])) + (fit2_p[6] * (y - fit2_p[1])) + (fit2_p[7] * (z - fit2_p[2]));
B = (fit2_p[6] * (x - fit2_p[0])) + (fit2_p[4] * (y - fit2_p[1])) + (fit2_p[8] * (z - fit2_p[2]));
C = (fit2_p[7] * (x - fit2_p[0])) + (fit2_p[8] * (y - fit2_p[1])) + (fit2_p[5] * (z - fit2_p[2]));
length = sqrtf(A * A + B * B + C * C);
residual = params.radius - length;
residual = p.radius - length;
fit2 += residual * residual;
}
float fitness = result.cost;
fit1 = sqrtf(fit1) / samples_collected;
fit2 = sqrtf(fit2) / samples_collected;
@@ -285,7 +285,7 @@ void lm_ellipsoid_fit_iteration(const float x[], const float y[], const float z[
} else if (fit2 < result.cost && fit2 < fit1) {
result.gradient_damping /= lma_damping;
memcpy(fit1_params, fit2_params, sizeof(fit1_params));
memcpy(fit1_p, fit2_p, sizeof(fit1_p));
fitness = fit2;
} else if (fit1 < result.cost) {
@@ -295,15 +295,17 @@ void lm_ellipsoid_fit_iteration(const float x[], const float y[], const float z[
//--------------------Levenberg-Marquardt-part-ends-here--------------------------------//
if (PX4_ISFINITE(fitness) && fitness <= result.cost) {
result.cost = fitness;
params.offset(0) = fit1_params[0];
params.offset(1) = fit1_params[1];
params.offset(2) = fit1_params[2];
params.diag(0) = fit1_params[3];
params.diag(1) = fit1_params[4];
params.diag(2) = fit1_params[5];
params.offdiag(0) = fit1_params[6];
params.offdiag(1) = fit1_params[7];
params.offdiag(2) = fit1_params[8];
p.offset(0) = fit1_p[0];
p.offset(1) = fit1_p[1];
p.offset(2) = fit1_p[2];
p.diag(0) = fit1_p[3];
p.diag(1) = fit1_p[4];
p.diag(2) = fit1_p[5];
p.offdiag(0) = fit1_p[6];
p.offdiag(1) = fit1_p[7];
p.offdiag(2) = fit1_p[8];
result.result = iteration_result::STATUS::SUCCESS;
} else {
@@ -311,19 +313,13 @@ void lm_ellipsoid_fit_iteration(const float x[], const float y[], const float z[
}
}
int lm_mag_fit(const float x[], const float y[], const float z[], unsigned int samples_collected, sphere_params &params,
bool full_ellipsoid)
int lm_fit(const matrix::Vector3f data[], unsigned samples_collected, sphere_params &params, bool full_ellipsoid)
{
static constexpr int min_iterations = 10;
static constexpr int max_iterations = 100;
const int max_iterations = 100;
const int min_iterations = 10;
const float cost_threshold = 0.01;
const float step_threshold = 0.001;
const float min_radius = 0.2;
const float max_radius = 0.7;
static constexpr float cost_threshold = 0.02f;
static constexpr float step_threshold = 0.001f;
iteration_result iter;
iter.cost = 1e30f;
@@ -333,15 +329,27 @@ int lm_mag_fit(const float x[], const float y[], const float z[], unsigned int s
for (int i = 0; i < max_iterations; i++) {
if (full_ellipsoid) {
lm_ellipsoid_fit_iteration(x, y, z, samples_collected, params, iter);
lm_ellipsoid_fit_iteration(data, samples_collected, params, iter);
PX4_INFO("[%d] O:[%.5f, %.5f, %.5f], S:[%.3f, %.3f, %.3f], Result: %d, Cost: %.6f, Damping: %.6f\n",
i,
(double)params.offset(0), (double)params.offset(1), (double)params.offset(2),
(double)params.diag(0), (double)params.diag(1), (double)params.diag(2),
iter.result == iteration_result::STATUS::SUCCESS, (double)iter.cost, (double)iter.gradient_damping);
} else {
lm_sphere_fit_iteration(x, y, z, samples_collected, params, iter);
lm_sphere_fit_iteration(data, samples_collected, params, iter);
PX4_INFO("[%d] O:[%.5f, %.5f, %.5f], Result: %d, Cost: %.6f, Damping: %.6f, Radius: %.4f\n",
i,
(double)params.offset(0), (double)params.offset(1), (double)params.offset(2),
iter.result == iteration_result::STATUS::SUCCESS, (double)iter.cost, (double)iter.gradient_damping,
(double)params.radius);
}
if (iter.result == iteration_result::STATUS::SUCCESS
&& min_radius < params.radius && params.radius < max_radius
&& i > min_iterations && (iter.cost < cost_threshold || iter.gradient_damping < step_threshold)) {
success = true;
break;
}
+5 -6
View File
@@ -37,7 +37,9 @@
#include <px4_platform_common/defines.h>
struct sphere_params {
matrix::Vector3f offset, diag{1.f, 1.f, 1.f}, offdiag;
matrix::Vector3f offset{0.f, 0.f, 0.f};
matrix::Vector3f diag{1.f, 1.f, 1.f};
matrix::Vector3f offdiag{0.f, 0.f, 0.f};
float radius{0.2f};
};
@@ -47,9 +49,7 @@ struct sphere_params {
*
* Fits a sphere to a set of points on the sphere surface.
*
* @param x point coordinates on the X axis
* @param y point coordinates on the Y axis
* @param z point coordinates on the Z axis
* @param data x,y,z point coordinates
* @param samples_collected number of points
* @param max_iterations abort if maximum number of iterations have been reached. If unsure, set to 100.
* @param params the values to be optimized
@@ -59,5 +59,4 @@ struct sphere_params {
*
* @return 0 on success, 1 on failure
*/
int lm_mag_fit(const float x[], const float y[], const float z[], unsigned int samples_collected, sphere_params &params,
bool full_ellipsoid);
int lm_fit(const matrix::Vector3f data[], unsigned samples_collected, sphere_params &params, bool full_ellipsoid);
+39 -57
View File
@@ -90,9 +90,7 @@ struct mag_worker_data_t {
uint64_t calibration_interval_perside_us;
unsigned int calibration_counter_total[MAX_MAGS];
float *x[MAX_MAGS];
float *y[MAX_MAGS];
float *z[MAX_MAGS];
matrix::Vector3f *data[MAX_MAGS] {nullptr, nullptr, nullptr, nullptr};
float temperature[MAX_MAGS] {NAN, NAN, NAN, NAN};
@@ -141,19 +139,20 @@ int do_mag_calibration(orb_advert_t *mavlink_log_pub)
return result;
}
static bool reject_sample(float sx, float sy, float sz, float x[], float y[], float z[], unsigned count,
static bool reject_sample(Vector3f s, Vector3f data[], unsigned count,
unsigned max_count, float mag_sphere_radius)
{
float min_sample_dist = fabsf(5.4f * mag_sphere_radius / sqrtf(max_count)) / 3.0f;
for (size_t i = 0; i < count; i++) {
float dx = sx - x[i];
float dy = sy - y[i];
float dz = sz - z[i];
float dx = s(0) - data[i](0);
float dy = s(1) - data[i](1);
float dz = s(2) - data[i](2);
float dist = sqrtf(dx * dx + dy * dy + dz * dz);
if (dist < min_sample_dist) {
PX4_DEBUG("rejected X: %.3f Y: %.3f Z: %.3f (%.3f < %.3f) (%u/%u) ", (double)sx, (double)sy, (double)sz, (double)dist,
PX4_DEBUG("rejected X: %.3f Y: %.3f Z: %.3f (%.3f < %.3f) (%u/%u) ", (double)s(0), (double)s(1), (double)s(2),
(double)dist,
(double)min_sample_dist, count, max_count);
return true;
@@ -363,8 +362,7 @@ static calibrate_return mag_calibration_worker(detect_orientation_return orienta
}
// Check if this measurement is good to go in
bool reject = reject_sample(mag.x, mag.y, mag.z,
worker_data->x[cur_mag], worker_data->y[cur_mag], worker_data->z[cur_mag],
bool reject = reject_sample(Vector3f{mag.x, mag.y, mag.z}, worker_data->data[cur_mag],
worker_data->calibration_counter_total[cur_mag],
worker_data->calibration_sides * worker_data->calibration_points_perside,
mag_sphere_radius);
@@ -388,9 +386,9 @@ static calibrate_return mag_calibration_worker(detect_orientation_return orienta
if (!rejected) {
for (uint8_t cur_mag = 0; cur_mag < MAX_MAGS; cur_mag++) {
if (worker_data->calibration[cur_mag].device_id() != 0) {
worker_data->x[cur_mag][worker_data->calibration_counter_total[cur_mag]] = new_samples[cur_mag](0);
worker_data->y[cur_mag][worker_data->calibration_counter_total[cur_mag]] = new_samples[cur_mag](1);
worker_data->z[cur_mag][worker_data->calibration_counter_total[cur_mag]] = new_samples[cur_mag](2);
worker_data->data[cur_mag][worker_data->calibration_counter_total[cur_mag]] = new_samples[cur_mag];
worker_data->calibration_counter_total[cur_mag]++;
if (!PX4_ISFINITE(worker_data->temperature[cur_mag])) {
// set first valid value
@@ -399,13 +397,11 @@ static calibrate_return mag_calibration_worker(detect_orientation_return orienta
} else {
worker_data->temperature[cur_mag] = 0.5f * (worker_data->temperature[cur_mag] + new_temperature[cur_mag]);
}
worker_data->calibration_counter_total[cur_mag]++;
}
}
hrt_abstime now = hrt_absolute_time();
mag_worker_data_s status;
mag_worker_data_s status{};
status.timestamp = now;
status.timestamp_sample = now;
status.done_count = worker_data->done_count;
@@ -418,9 +414,9 @@ static calibrate_return mag_calibration_worker(detect_orientation_return orienta
if (worker_data->calibration[cur_mag].device_id() != 0) {
const unsigned int sample = worker_data->calibration_counter_total[cur_mag] - 1;
status.x[cur_mag] = worker_data->x[cur_mag][sample];
status.y[cur_mag] = worker_data->y[cur_mag][sample];
status.z[cur_mag] = worker_data->z[cur_mag][sample];
status.x[cur_mag] = worker_data->data[cur_mag][sample](0);
status.y[cur_mag] = worker_data->data[cur_mag][sample](1);
status.z[cur_mag] = worker_data->data[cur_mag][sample](2);
} else {
status.x[cur_mag] = 0.f;
@@ -527,9 +523,6 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
for (size_t cur_mag = 0; cur_mag < MAX_MAGS; cur_mag++) {
// Initialize to no memory allocated
worker_data.x[cur_mag] = nullptr;
worker_data.y[cur_mag] = nullptr;
worker_data.z[cur_mag] = nullptr;
worker_data.calibration_counter_total[cur_mag] = 0;
}
@@ -547,11 +540,10 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
worker_data.calibration[cur_mag].set_calibration_index(cur_mag);
if (worker_data.calibration[cur_mag].device_id() != 0) {
worker_data.x[cur_mag] = static_cast<float *>(malloc(sizeof(float) * calibration_points_maxcount));
worker_data.y[cur_mag] = static_cast<float *>(malloc(sizeof(float) * calibration_points_maxcount));
worker_data.z[cur_mag] = static_cast<float *>(malloc(sizeof(float) * calibration_points_maxcount));
if (worker_data.x[cur_mag] == nullptr || worker_data.y[cur_mag] == nullptr || worker_data.z[cur_mag] == nullptr) {
worker_data.data[cur_mag] = new matrix::Vector3f[calibration_points_maxcount];
if (worker_data.data[cur_mag] == nullptr) {
calibration_log_critical(mavlink_log_pub, "ERROR: out of memory");
result = calibrate_return_error;
break;
@@ -566,8 +558,7 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
result = calibrate_from_orientation(mavlink_log_pub, // uORB handle to write output
worker_data.side_data_collected, // Sides to calibrate
mag_calibration_worker, // Calibration worker
&worker_data, // Opaque data for calibration worked
true); // true: lenient still detection
&worker_data); // Opaque data for calibration worked
}
// calibration values for each mag
@@ -594,26 +585,26 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
// Estimate only the offsets if two-sided calibration is selected, as the problem is not constrained
// enough to reliably estimate both scales and offsets with 2 sides only (even if the existing calibration
// is already close)
bool sphere_fit_only = worker_data.calibration_sides <= 2;
const bool sphere_fit_only = worker_data.calibration_sides <= 2;
sphere_params sphere_data;
sphere_params sphere_data{};
sphere_data.radius = sphere_radius[cur_mag];
sphere_data.offset = matrix::Vector3f(sphere[cur_mag](0), sphere[cur_mag](1), sphere[cur_mag](2));
sphere_data.diag = matrix::Vector3f(diag[cur_mag](0), diag[cur_mag](1), diag[cur_mag](2));
sphere_data.offdiag = matrix::Vector3f(offdiag[cur_mag](0), offdiag[cur_mag](1), offdiag[cur_mag](2));
sphere_data.offset = sphere[cur_mag];
sphere_data.diag = diag[cur_mag];
sphere_data.offdiag = offdiag[cur_mag];
bool sphere_fit_success = false;
bool ellipsoid_fit_success = false;
int ret = lm_mag_fit(worker_data.x[cur_mag], worker_data.y[cur_mag], worker_data.z[cur_mag],
worker_data.calibration_counter_total[cur_mag], sphere_data, false);
int ret = lm_fit(worker_data.data[cur_mag], worker_data.calibration_counter_total[cur_mag], sphere_data, false);
if (ret == PX4_OK) {
sphere_fit_success = true;
PX4_INFO("Mag: %" PRIu8 " sphere radius: %.4f", cur_mag, (double)sphere_data.radius);
if (!sphere_fit_only) {
int ellipsoid_ret = lm_mag_fit(worker_data.x[cur_mag], worker_data.y[cur_mag], worker_data.z[cur_mag],
worker_data.calibration_counter_total[cur_mag], sphere_data, true);
int ellipsoid_ret = lm_fit(worker_data.data[cur_mag], worker_data.calibration_counter_total[cur_mag], sphere_data,
true);
if (ellipsoid_ret == PX4_OK) {
ellipsoid_fit_success = true;
@@ -621,13 +612,14 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
}
}
// TODO: verify sphere radius
sphere_radius[cur_mag] = sphere_data.radius;
for (int i = 0; i < 3; i++) {
sphere[cur_mag](i) = sphere_data.offset(i);
diag[cur_mag](i) = sphere_data.diag(i);
offdiag[cur_mag](i) = sphere_data.offdiag(i);
}
sphere[cur_mag] = sphere_data.offset;
diag[cur_mag] = sphere_data.diag;
offdiag[cur_mag] = sphere_data.offdiag;
if (!sphere_fit_success && !ellipsoid_fit_success) {
calibration_log_emergency(mavlink_log_pub, "Retry calibration (unable to fit mag %" PRIu8 ")", cur_mag);
@@ -728,7 +720,7 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
for (unsigned i = 0; i < worker_data.calibration_counter_total[cur_mag]; i++) {
const Vector3f sample{worker_data.x[cur_mag][i], worker_data.y[cur_mag][i], worker_data.z[cur_mag][i]};
const Vector3f sample{worker_data.data[cur_mag][i]};
// apply calibration
Vector3f m{scale *(sample - offset)};
@@ -739,9 +731,7 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
}
// store back in worker_data
worker_data.x[cur_mag][i] = m(0);
worker_data.y[cur_mag][i] = m(1);
worker_data.z[cur_mag][i] = m(2);
worker_data.data[cur_mag][i] = m;
}
}
}
@@ -775,14 +765,8 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
float diff_sum = 0.f;
for (int i = 0; i < last_sample_index; i++) {
float x = worker_data.x[cur_mag][i];
float y = worker_data.y[cur_mag][i];
float z = worker_data.z[cur_mag][i];
rotate_3f((enum Rotation)r, x, y, z);
Vector3f diff = Vector3f{x, y, z} - Vector3f{worker_data.x[internal_index][i], worker_data.y[internal_index][i], worker_data.z[internal_index][i]};
diff_sum += diff.norm_squared();
const Vector3f &m{worker_data.data[cur_mag][i]};
diff_sum += Vector3f((get_rot_matrix((enum Rotation)r) * m) - m).norm_squared();
}
// compute mean squared error
@@ -878,9 +862,7 @@ calibrate_return mag_calibrate_all(orb_advert_t *mavlink_log_pub, int32_t cal_ma
// Data points are no longer needed
for (size_t cur_mag = 0; cur_mag < MAX_MAGS; cur_mag++) {
free(worker_data.x[cur_mag]);
free(worker_data.y[cur_mag]);
free(worker_data.z[cur_mag]);
delete[] worker_data.data[cur_mag];
}
FactoryCalibrationStorage factory_storage;
+39 -40
View File
@@ -50,38 +50,38 @@ using matrix::Vector3f;
class MagCalTest : public ::testing::Test
{
public:
void generate2SidesMagData(float *x, float *y, float *z, unsigned int n_samples, float mag_str);
void generate2SidesMagData(matrix::Vector3f mag_data[], unsigned int n_samples, float mag_str);
/* Generate regularly spaced data on a sphere
* Ref.: How to generate equidistributed points on the surface of a sphere, Markus Deserno, 2004
*/
void generateRegularData(float *x, float *y, float *z, unsigned int n_samples, float mag_str);
void generateRegularData(matrix::Vector3f mag_data[], unsigned int n_samples, float mag_str);
void modifyOffsetScale(float *x, float *y, float *z, unsigned int n_samples, Vector3f offsets, Vector3f scale_factors);
void modifyOffsetScale(matrix::Vector3f mag_data[], unsigned int n_samples, Vector3f offsets, Vector3f scale_factors);
};
void MagCalTest::generate2SidesMagData(float *x, float *y, float *z, unsigned int n_samples, float mag_str)
void MagCalTest::generate2SidesMagData(matrix::Vector3f data[], unsigned int n_samples, float mag_str)
{
float psi = 0.f;
float theta = 0.f;
const float d_angle = 2.f * M_PI_F / float(n_samples / 2);
for (int i = 0; i < int(n_samples / 2); i++) {
x[i] = mag_str * sinf(psi);
y[i] = mag_str * cosf(psi);
z[i] = 0.f;
data[i](0) = mag_str * sinf(psi);
data[i](1) = mag_str * cosf(psi);
data[i](2) = 0.f;
psi += d_angle;
}
for (int i = int(n_samples / 2); i < int(n_samples); i++) {
x[i] = mag_str * sinf(theta);
y[i] = 0.f;
z[i] = mag_str * cosf(theta);
data[i](0) = mag_str * sinf(theta);
data[i](1) = 0.f;
data[i](2) = mag_str * cosf(theta);
theta += d_angle;
}
}
void MagCalTest::generateRegularData(float *x, float *y, float *z, unsigned int n_samples, float mag_str)
void MagCalTest::generateRegularData(matrix::Vector3f mag_data[], unsigned int n_samples, float mag_str)
{
const float a = 4.f * M_PI_F * mag_str * mag_str / n_samples;
const float d = sqrtf(a);
@@ -97,9 +97,9 @@ void MagCalTest::generateRegularData(float *x, float *y, float *z, unsigned int
for (int n = 0; n < m_phi; n++) {
const float phi = 2.f * M_PI_F * n / static_cast<float>(m_phi);
x[n_count] = mag_str * sinf(theta) * cosf(phi);
y[n_count] = mag_str * sinf(theta) * sinf(phi);
z[n_count] = mag_str * cosf(theta);
mag_data[n_count](0) = mag_str * sinf(theta) * cosf(phi);
mag_data[n_count](1) = mag_str * sinf(theta) * sinf(phi);
mag_data[n_count](2) = mag_str * cosf(theta);
n_count++;
}
}
@@ -111,20 +111,16 @@ void MagCalTest::generateRegularData(float *x, float *y, float *z, unsigned int
// Padd with constant data
while (n_count < n_samples) {
x[n_count] = x[n_count - 1];
y[n_count] = y[n_count - 1];
z[n_count] = z[n_count - 1];
mag_data[n_count] = mag_data[n_count - 1];
n_count++;
}
}
void MagCalTest::modifyOffsetScale(float *x, float *y, float *z, unsigned int n_samples, Vector3f offsets,
void MagCalTest::modifyOffsetScale(matrix::Vector3f mag_data[], unsigned int n_samples, Vector3f offsets,
Vector3f scale_factors)
{
for (unsigned int k = 0; k < n_samples; k++) {
x[k] = x[k] * scale_factors(0) + offsets(0);
y[k] = y[k] * scale_factors(1) + offsets(1);
z[k] = z[k] * scale_factors(2) + offsets(2);
mag_data[k] = mag_data[k].emult(scale_factors) + offsets;
}
}
@@ -138,17 +134,13 @@ TEST_F(MagCalTest, sphere2Sides)
const Vector3f offset_true;
const Vector3f scale_true = {1.f, 1.f, 1.f};
float x[N_SAMPLES];
float y[N_SAMPLES];
float z[N_SAMPLES];
generate2SidesMagData(x, y, z, N_SAMPLES, mag_str_true);
Vector3f mag_data[N_SAMPLES];
generate2SidesMagData(mag_data, N_SAMPLES, mag_str_true);
// WHEN: fitting a sphere with the data and given a wrong initial radius
sphere_params sphere;
sphere.diag = {1.f, 1.f, 1.f};
sphere_params sphere{};
sphere.radius = 0.2;
int success = lm_mag_fit(x, y, z, N_SAMPLES, sphere, false);
int success = lm_fit(mag_data, N_SAMPLES, sphere, false);
// THEN: the algorithm should converge in a single step
EXPECT_EQ(success, PX4_OK);
@@ -171,17 +163,15 @@ TEST_F(MagCalTest, sphereRegularlySpaced)
const Vector3f offset_true = {-1.07f, 0.35f, -0.78f};
const Vector3f scale_true = {1.f, 1.f, 1.f};
float x[N_SAMPLES];
float y[N_SAMPLES];
float z[N_SAMPLES];
generateRegularData(x, y, z, N_SAMPLES, mag_str_true);
modifyOffsetScale(x, y, z, N_SAMPLES, offset_true, scale_true);
Vector3f mag_data[N_SAMPLES];
generateRegularData(mag_data, N_SAMPLES, mag_str_true);
modifyOffsetScale(mag_data, N_SAMPLES, offset_true, scale_true);
// WHEN: fitting a sphere to the data
sphere_params sphere;
sphere_params sphere{};
sphere.diag = {1.f, 1.f, 1.f};
sphere.radius = 0.2;
int success = lm_mag_fit(x, y, z, N_SAMPLES, sphere, false);
int success = lm_fit(mag_data, N_SAMPLES, sphere, false);
// THEN: the algorithm should converge in a few iterations and
// find the correct parameters
@@ -200,7 +190,7 @@ TEST_F(MagCalTest, replayTestData)
// GIVEN: a real test dataset with large offsets
// and where the two first iterations of the LM algorithm
// produces a negative radius and a constant fitness value
constexpr unsigned int N_SAMPLES = 231;
constexpr unsigned N_SAMPLES = 231;
const float mag_str_true = 0.4f;
const Vector3f offset_true = {-0.18f, 0.05f, -0.58f};
@@ -209,7 +199,16 @@ TEST_F(MagCalTest, replayTestData)
sphere_params sphere;
sphere.diag = {1.f, 1.f, 1.f};
sphere.radius = 0.2;
int sphere_success = lm_mag_fit(mag_data1_x, mag_data1_y, mag_data1_z, N_SAMPLES, sphere, false);
matrix::Vector3f mag_data1[231];
for (unsigned i = 0; i < N_SAMPLES; i++) {
mag_data1[i](0) = mag_data1_x[i];
mag_data1[i](1) = mag_data1_y[i];
mag_data1[i](2) = mag_data1_z[i];
}
int sphere_success = lm_fit(mag_data1, N_SAMPLES, sphere, false);
// THEN: the algorithm should converge and find the correct parameters
EXPECT_EQ(sphere_success, PX4_OK);
@@ -222,8 +221,8 @@ TEST_F(MagCalTest, replayTestData)
sphere_params ellipsoid;
ellipsoid.diag = {1.f, 1.f, 1.f};
ellipsoid.radius = 0.2;
int ellipsoid_step_1_success = lm_mag_fit(mag_data1_x, mag_data1_y, mag_data1_z, N_SAMPLES, ellipsoid, false);
int ellipsoid_success = lm_mag_fit(mag_data1_x, mag_data1_y, mag_data1_z, N_SAMPLES, ellipsoid, true);
int ellipsoid_step_1_success = lm_fit(mag_data1, N_SAMPLES, ellipsoid, false);
int ellipsoid_success = lm_fit(mag_data1, N_SAMPLES, ellipsoid, true);
const Vector3f scale_true = {1.f, 1.06f, 0.94f};
EXPECT_EQ(ellipsoid_step_1_success, PX4_OK);
+66
View File
@@ -0,0 +1,66 @@
/****************************************************************************
*
* Copyright (c) 2021 PX4 Development Team. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
* 3. Neither the name PX4 nor the names of its contributors may be
* used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
* OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
* AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
****************************************************************************/
/**
* Bitfield selecting mag sides for calibration
*
* If set to two side calibration, only the offsets are estimated, the scale
* calibration is left unchanged. Thus an initial six side calibration is
* recommended.
*
* Bits:
* ORIENTATION_TAIL_DOWN = 1
* ORIENTATION_NOSE_DOWN = 2
* ORIENTATION_LEFT = 4
* ORIENTATION_RIGHT = 8
* ORIENTATION_UPSIDE_DOWN = 16
* ORIENTATION_RIGHTSIDE_UP = 32
*
* @min 34
* @max 63
* @value 34 Two side calibration
* @value 38 Three side calibration
* @value 63 Six side calibration
* @group Sensors
*/
PARAM_DEFINE_INT32(CAL_ACC_SIDES, 63);
/**
* Automatically set external rotations.
*
* During calibration attempt to automatically determine the rotation of external magnetometers.
*
* @boolean
* @group Sensors
*/
PARAM_DEFINE_INT32(CAL_CAL_ROT_AUTO, 1);