eradicate ukf
This commit is contained in:
ary
parent
0180ec7580
commit
38952fc102
@ -1,44 +0,0 @@
|
||||
#pragma once
|
||||
|
||||
#include <Eigen/Dense>
|
||||
#include <Eigen/src/Cholesky/LLT.h>
|
||||
#include <Eigen/src/Core/Matrix.h>
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
class UnscentedKalman
|
||||
{
|
||||
public:
|
||||
void predict();
|
||||
void update(const Eigen::Vector<double, measure_dim> measurements);
|
||||
Eigen::Vector<double, state_dim> predict_function_f(Eigen::Matrix<double, 1, state_dim> item); //build this last
|
||||
Eigen::Vector<double, measure_dim> update_function_h(Eigen::Matrix<double, 1, state_dim> item); //build this last
|
||||
void unscented_transform_f(Eigen::Matrix<double, 2 * state_dim + 1, state_dim> sigmas); //Add spot for Q and R matricies
|
||||
void unscented_transform_h(Eigen::Matrix<double, 2 * state_dim + 1, measure_dim> sigmas);
|
||||
const Eigen::Vector<double, state_dim> get_state() { return state_vector; }
|
||||
const Eigen::Matrix<double, state_dim, state_dim> get_covar() { return covariance_matrix; }
|
||||
const Eigen::Vector<double, state_dim> get_state_p() { return state_vector_p; }
|
||||
const Eigen::Matrix<double, state_dim, state_dim> get_covar_p() { return covariance_matrix_p; }
|
||||
private:
|
||||
static const int sigma_dim = 2 * state_dim + 1;
|
||||
const double kappa = 3 - state_dim;
|
||||
const double alpha = 1;
|
||||
const double beta = 2;
|
||||
const double lambda_ = pow(alpha, 2) * (state_dim - kappa) - state_dim;
|
||||
Eigen::Matrix<double, 1, sigma_dim> weights_m;
|
||||
Eigen::Matrix<double, 1, sigma_dim> weights_c;
|
||||
Eigen::Vector<double, state_dim> state_vector = Eigen::Vector<double, state_dim>::Ones();
|
||||
Eigen::Matrix<double, state_dim, state_dim> covariance_matrix = Eigen::Matrix<double, state_dim, state_dim>::Identity();
|
||||
Eigen::Vector<double, state_dim> state_vector_p = Eigen::Vector<double, state_dim>::Ones();
|
||||
Eigen::Matrix<double, state_dim, state_dim> covariance_matrix_p = Eigen::Matrix<double, state_dim, state_dim>::Identity();
|
||||
Eigen::Matrix<double, sigma_dim, state_dim> sigma_points = Eigen::Matrix<double, sigma_dim, state_dim>::Ones();
|
||||
Eigen::Matrix<double, sigma_dim, state_dim> sigma_points_f = Eigen::Matrix<double, sigma_dim, state_dim>::Ones();
|
||||
Eigen::Matrix<double, sigma_dim, measure_dim> sigma_points_h = Eigen::Matrix<double, sigma_dim, measure_dim>::Ones();
|
||||
Eigen::Vector<double, state_dim> unscented_x_p = Eigen::Vector<double, state_dim>::Ones();
|
||||
Eigen::Matrix<double, state_dim, state_dim> unscented_P_p = Eigen::Matrix<double, state_dim, state_dim>::Ones();
|
||||
Eigen::Vector<double, measure_dim> unscented_x_z = Eigen::Vector<double, measure_dim>::Ones();
|
||||
Eigen::Matrix<double, measure_dim, measure_dim> unscented_P_z = Eigen::Matrix<double, measure_dim, measure_dim>::Ones();
|
||||
Eigen::Vector<double, measure_dim> measure_vector = Eigen::Vector<double, measure_dim>::Ones();
|
||||
Eigen::Matrix<double, state_dim, state_dim> Q = Eigen::Matrix<double, state_dim, state_dim>::Random();
|
||||
Eigen::Matrix<double, measure_dim, measure_dim> R = Eigen::Matrix<double, measure_dim, measure_dim>::Random();
|
||||
|
||||
};
|
||||
@ -1,120 +0,0 @@
|
||||
// Kalman_Filter_Test.cpp : This file contains the 'main' function. Program execution begins and ends there.
|
||||
//
|
||||
|
||||
#include <iostream>
|
||||
#include <chrono>
|
||||
#include "UnscentedKalman.h"
|
||||
#include "SystemSims.h"
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
void UnscentedKalman<state_dim, measure_dim>::unscented_transform_f(Eigen::Matrix<double, 2 * state_dim + 1, state_dim> sigmas)
|
||||
{
|
||||
for (int i = 0; i < state_dim; ++i)
|
||||
{
|
||||
unscented_x_p[i] = weights_m.dot(sigmas.col(i));
|
||||
}
|
||||
Eigen::Matrix<double, state_dim, state_dim> temp;
|
||||
for (int i = 0; i < sigma_dim; ++i)
|
||||
{
|
||||
Eigen::Matrix<double, 1, state_dim> Y = sigmas.row(i) - unscented_x_p.transpose();
|
||||
auto scalar = (weights_c[i] * (Y * Y.transpose())[0]);
|
||||
temp.fill(scalar);
|
||||
unscented_P_p = (unscented_P_p + temp);
|
||||
}
|
||||
unscented_P_p += Q;
|
||||
}
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
void UnscentedKalman<state_dim, measure_dim>::unscented_transform_h(Eigen::Matrix<double, 2 * state_dim + 1, measure_dim> sigmas)
|
||||
{
|
||||
for (int i = 0; i < measure_dim; ++i)
|
||||
{
|
||||
unscented_x_z[i] = weights_m.dot(sigmas.col(i));
|
||||
}
|
||||
Eigen::Matrix<double, measure_dim, measure_dim> temp;
|
||||
for (int i = 0; i < sigma_dim; ++i)
|
||||
{
|
||||
Eigen::Matrix<double, 1, measure_dim> Y = sigmas.row(i) - unscented_x_z.transpose();
|
||||
auto scalar = (weights_c[i] * (Y * Y.transpose())[0]);
|
||||
temp.fill(scalar);
|
||||
unscented_P_z += temp;
|
||||
}
|
||||
unscented_P_z += R;
|
||||
}
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
void UnscentedKalman<state_dim, measure_dim>::predict()
|
||||
{
|
||||
weights_m.fill(1 / (2 * (state_dim + lambda_)));
|
||||
weights_c.fill(1 / (2 * (state_dim + lambda_)));
|
||||
weights_m(0) = lambda_ / (state_dim + lambda_) + (1 - pow(alpha, 2) + beta);
|
||||
weights_c(0) = lambda_ / (state_dim + lambda_);
|
||||
Eigen::LLT<Eigen::Matrix<double, state_dim, state_dim>> U((state_dim + lambda_) * covariance_matrix);
|
||||
Eigen::Matrix<double, state_dim, state_dim> U_decomposed = U.matrixU();
|
||||
sigma_points.row(0) = state_vector;
|
||||
for (int i = 0; i < state_dim; ++i)
|
||||
{
|
||||
sigma_points.row(i + 1) = state_vector.transpose() + U_decomposed.row(i);
|
||||
sigma_points.row(state_dim + i + 1) = state_vector.transpose() - U_decomposed.row(i);
|
||||
}
|
||||
for (int i = 0; i < sigma_dim; ++i)
|
||||
{
|
||||
sigma_points_f.row(i) = this->predict_function_f(sigma_points.row(i));
|
||||
}
|
||||
this->unscented_transform_f(sigma_points_f);
|
||||
state_vector_p = unscented_x_p;
|
||||
covariance_matrix_p = unscented_P_p;
|
||||
}
|
||||
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
void UnscentedKalman<state_dim, measure_dim>::update(const Eigen::Vector<double, measure_dim> measurements)
|
||||
{
|
||||
for (int i = 0; i < sigma_dim; ++i)
|
||||
{
|
||||
sigma_points_h.row(i) = this->update_function_h(sigma_points_f.row(i));
|
||||
}
|
||||
this->unscented_transform_h(sigma_points_h);
|
||||
Eigen::Matrix<double, state_dim, measure_dim> covariance_x_z;
|
||||
for (int i = 0; i < sigma_dim; ++i)
|
||||
{
|
||||
auto inner1 = (sigma_points_h.row(i) - unscented_x_z.transpose());
|
||||
auto inner2 = (sigma_points_f.row(i) - state_vector_p.transpose());
|
||||
auto comp = inner1.transpose() * inner2;
|
||||
auto temp = weights_c[i] * comp.transpose();
|
||||
covariance_x_z += temp;
|
||||
}
|
||||
auto K = covariance_x_z * unscented_P_z.inverse();
|
||||
state_vector = state_vector_p + K * (measurements - unscented_x_z);
|
||||
covariance_matrix = covariance_matrix_p - (K * unscented_P_z * K.transpose());
|
||||
}
|
||||
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
Eigen::Vector<double, state_dim> UnscentedKalman<state_dim, measure_dim>::predict_function_f(Eigen::Matrix<double, 1, state_dim> item)
|
||||
{
|
||||
Eigen::Matrix<double, state_dim, state_dim> F = Eigen::Matrix<double, state_dim, state_dim>::Ones();
|
||||
return F * item.transpose();
|
||||
}
|
||||
|
||||
|
||||
template<int state_dim, int measure_dim>
|
||||
Eigen::Vector<double, measure_dim> UnscentedKalman<state_dim, measure_dim>::update_function_h(Eigen::Matrix<double, 1, state_dim> item)
|
||||
{
|
||||
return { 1, 2, 3 };
|
||||
}
|
||||
|
||||
|
||||
|
||||
UnscentedKalman<3, 3> filter;
|
||||
|
||||
Eigen::Vector<double, 3> test2 = Eigen::Matrix<double, 1, 3>::Random();
|
||||
Eigen::Vector<double, 3> test3 = Eigen::Matrix<double, 1, 3>::Random();
|
||||
Eigen::Vector<double, 3> test4 = Eigen::Matrix<double, 1, 3>::Random();
|
||||
|
||||
int main()
|
||||
{
|
||||
|
||||
}
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user