px4-firmware/apps/px4/attitude_estimator_bm/attitude_bm.c

315 lines
8.3 KiB
C

/*
* attitude_bm.c
*
* Created on: 21.12.2010
* Author: Laurens Mackay, Tobias Naegeli
*/
#include <math.h>
#include "attitude_bm.h"
#include "kalman.h"
#define TIME_STEP (1.0f / 500.0f)
static kalman_t attitude_blackmagic_kal;
void vect_norm(float_vect3 *vect)
{
float length = sqrtf(
vect->x * vect->x + vect->y * vect->y + vect->z * vect->z);
if (length != 0) {
vect->x /= length;
vect->y /= length;
vect->z /= length;
}
}
void vect_cross_product(const float_vect3 *a, const float_vect3 *b,
float_vect3 *c)
{
c->x = a->y * b->z - a->z * b->y;
c->y = a->z * b->x - a->x * b->z;
c->z = a->x * b->y - a->y * b->x;
}
void attitude_blackmagic_update_a(void)
{
// for acc
// Idendity matrix already in A.
M(attitude_blackmagic_kal.a, 0, 1) = TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 11);
M(attitude_blackmagic_kal.a, 0, 2) = -TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 10);
M(attitude_blackmagic_kal.a, 1, 0) = -TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 11);
M(attitude_blackmagic_kal.a, 1, 2) = TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 9);
M(attitude_blackmagic_kal.a, 2, 0) = TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 10);
M(attitude_blackmagic_kal.a, 2, 1) = -TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 9);
// for mag
// Idendity matrix already in A.
M(attitude_blackmagic_kal.a, 3, 4) = TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 11);
M(attitude_blackmagic_kal.a, 3, 5) = -TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 10);
M(attitude_blackmagic_kal.a, 4, 3) = -TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 11);
M(attitude_blackmagic_kal.a, 4, 5) = TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 9);
M(attitude_blackmagic_kal.a, 5, 3) = TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 10);
M(attitude_blackmagic_kal.a, 5, 4) = -TIME_STEP * kalman_get_state(
&attitude_blackmagic_kal, 9);
}
void attitude_blackmagic_init(void)
{
//X Kalmanfilter
//initalize matrices
static m_elem kal_a[12 * 12] = {
1.0f, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 1.0f, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 1.0f, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 1.0f, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 1.0f, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 1.0f, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 1.0f, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 1.0f, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 1.0f, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0f, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0f, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1.0f
};
static m_elem kal_c[9 * 12] = {
1.0f, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 1.0f, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 1.0f, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 1.0f, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 1.0f, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 1.0f, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 1.0f, 0, 0, 1.0f, 0, 0,
0, 0, 0, 0, 0, 0, 0, 1.0f, 0, 0, 1.0f, 0,
0, 0, 0, 0, 0, 0, 0, 0, 1.0f, 0, 0, 1.0f
};
#define FACTOR 0.5
#define FACTORstart 1
// static m_elem kal_gain[12 * 9] =
// { 0.004 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0,
// 0 , 0.004 , 0 , 0 , 0 , 0 , 0 , 0 , 0,
// 0 , 0 , 0.004 , 0 , 0 , 0 , 0 , 0 , 0,
// 0 , 0 , 0 , 0.015, 0 , 0 , 0 , 0 , 0,
// 0 , 0 , 0 , 0 , 0.015, 0 , 0 , 0 , 0,
// 0 , 0 , 0 , 0 , 0 , 0.015, 0 , 0 , 0,
// 0.0000 , +0.00002,0 , 0 , 0, 0, 0, 0 , 0,
// -0.00002,0 , 0 , 0 , 0, 0, 0, 0, 0,
// 0, 0 , 0 , 0, 0, 0, 0, 0, 0,
// 0 , 0 , 0 , 0 , 0 , 0 , 0.4 , 0 , 0,
// 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0.4 , 0,
// 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0.4
// };
static m_elem kal_gain[12 * 9] = {
0.0006f , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0,
0 , 0.0006f , 0 , 0 , 0 , 0 , 0 , 0 , 0,
0 , 0 , 0.0006f , 0 , 0 , 0 , 0 , 0 , 0,
0 , 0 , 0 , 0.015f, 0 , 0 , 0 , 0 , 0,
0 , 0 , 0 , 0 , 0.015f, 0 , 0 , 0 , 0,
0 , 0 , 0 , 0 , 0 , 0.015f, 0 , 0 , 0,
0.0000f , +0.00002f, 0 , 0 , 0, 0, 0, 0 , 0,
-0.00002f, 0 , 0 , 0 , 0, 0, 0, 0, 0,
0, 0 , 0 , 0, 0, 0, 0, 0, 0,
0 , 0 , 0 , 0 , 0 , 0 , 0.6f , 0 , 0,
0 , 0 , 0 , 0 , 0 , 0 , 0 , 0.6f , 0,
0 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0.6f
};
//offset update only correct if not upside down.
#define K (10.0f*TIME_STEP)
static m_elem kal_gain_start[12 * 9] = {
K, 0, 0, 0, 0, 0, 0, 0, 0,
0, K, 0, 0, 0, 0, 0, 0, 0,
0, 0, K, 0, 0, 0, 0, 0, 0,
0, 0, 0, K, 0, 0, 0, 0, 0,
0, 0, 0, 0, K, 0, 0, 0, 0,
0, 0, 0, 0, 0, K, 0, 0, 0,
0, 0, 0, 0, 0, 0, K, 0, 0,
0, 0, 0, 0, 0, 0, 0, K, 0,
0, 0, 0, 0, 0, 0, 0, 0, K,
0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0
};
static m_elem kal_x_apriori[12 * 1] =
{ };
//---> initial states sind aposteriori!? ---> fehler
static m_elem kal_x_aposteriori[12 * 1] =
{ 0.0f, 0.0f, -1.0f, 0.6f, 0.0f, 0.8f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f };
kalman_init(&attitude_blackmagic_kal, 12, 9, kal_a, kal_c,
kal_gain_start, kal_gain, kal_x_apriori, kal_x_aposteriori, 1000);
}
void attitude_blackmagic(const float_vect3 *accel, const float_vect3 *mag, const float_vect3 *gyro)
{
//Transform accelerometer used in all directions
// float_vect3 acc_nav;
//body2navi(&global_data.accel_si, &global_data.attitude, &acc_nav);
// Kalman Filter
//Calculate new linearized A matrix
attitude_blackmagic_update_a();
kalman_predict(&attitude_blackmagic_kal);
//correction update
m_elem measurement[9] =
{ };
m_elem mask[9] =
{ 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f };
measurement[0] = accel->x;
measurement[1] = accel->y;
measurement[2] = accel->z;
measurement[3] = mag->x;
measurement[4] = mag->y;
measurement[5] = mag->z;
measurement[6] = gyro->x;
measurement[7] = gyro->y;
measurement[8] = gyro->z;
//Put measurements into filter
// static int j = 0;
// if (j >= 3)
// {
// j = 0;
//
// mask[3]=1;
// mask[4]=1;
// mask[5]=1;
// j=0;
//
// }else{
// j++;}
kalman_correct(&attitude_blackmagic_kal, measurement, mask);
}
void attitude_blackmagic_get_all(float_vect3 *euler, float_vect3 *rates, float_vect3 *x_n_b, float_vect3 *y_n_b, float_vect3 *z_n_b)
{
//debug
// save outputs
float_vect3 kal_acc;
float_vect3 kal_mag;
// float_vect3 kal_w0, kal_w;
kal_acc.x = kalman_get_state(&attitude_blackmagic_kal, 0);
kal_acc.y = kalman_get_state(&attitude_blackmagic_kal, 1);
kal_acc.z = kalman_get_state(&attitude_blackmagic_kal, 2);
kal_mag.x = kalman_get_state(&attitude_blackmagic_kal, 3);
kal_mag.y = kalman_get_state(&attitude_blackmagic_kal, 4);
kal_mag.z = kalman_get_state(&attitude_blackmagic_kal, 5);
// kal_w0.x = kalman_get_state(&attitude_blackmagic_kal, 6);
// kal_w0.y = kalman_get_state(&attitude_blackmagic_kal, 7);
// kal_w0.z = kalman_get_state(&attitude_blackmagic_kal, 8);
//
// kal_w.x = kalman_get_state(&attitude_blackmagic_kal, 9);
// kal_w.y = kalman_get_state(&attitude_blackmagic_kal, 10);
// kal_w.z = kalman_get_state(&attitude_blackmagic_kal, 11);
rates->x = kalman_get_state(&attitude_blackmagic_kal, 9);
rates->y = kalman_get_state(&attitude_blackmagic_kal, 10);
rates->z = kalman_get_state(&attitude_blackmagic_kal, 11);
// kal_w = kal_w; // XXX hack to silence compiler warning
// kal_w0 = kal_w0; // XXX hack to silence compiler warning
//debug_vect("magn", mag);
//float_vect3 x_n_b, y_n_b, z_n_b;
z_n_b->x = -kal_acc.x;
z_n_b->y = -kal_acc.y;
z_n_b->z = -kal_acc.z;
vect_norm(z_n_b);
vect_cross_product(z_n_b, &kal_mag, y_n_b);
vect_norm(y_n_b);
vect_cross_product(y_n_b, z_n_b, x_n_b);
//save euler angles
euler->x = atan2f(z_n_b->y, z_n_b->z);
euler->y = -asinf(z_n_b->x);
euler->z = atan2f(y_n_b->x, x_n_b->x);
}