| Line |
Branch |
Exec |
Source |
| 1 |
|
|
/* |
| 2 |
|
|
* Copyright 2010, |
| 3 |
|
|
* Nicolas Mansard, |
| 4 |
|
|
* François Bleibel, |
| 5 |
|
|
* Olivier Stasse, |
| 6 |
|
|
* Florent Lamiraux |
| 7 |
|
|
* |
| 8 |
|
|
* CNRS/AIST |
| 9 |
|
|
* |
| 10 |
|
|
*/ |
| 11 |
|
|
|
| 12 |
|
|
/* --- SOT --- */ |
| 13 |
|
|
#include <dynamic-graph/command-setter.h> |
| 14 |
|
|
#include <dynamic-graph/factory.h> |
| 15 |
|
|
|
| 16 |
|
|
#include <sot/core/debug.hh> |
| 17 |
|
|
#include <sot/core/exception-tools.hh> |
| 18 |
|
|
#include <sot/core/factory.hh> |
| 19 |
|
|
#include <sot/core/kalman.hh> /* Header of the class implemented here. */ |
| 20 |
|
|
|
| 21 |
|
|
namespace dynamicgraph { |
| 22 |
|
|
using command::Setter; |
| 23 |
|
|
namespace sot { |
| 24 |
|
|
|
| 25 |
|
✗ |
DYNAMICGRAPH_FACTORY_ENTITY_PLUGIN(Kalman, "Kalman"); |
| 26 |
|
|
|
| 27 |
|
✗ |
Kalman::Kalman(const std::string &name) |
| 28 |
|
|
: Entity(name), |
| 29 |
|
✗ |
measureSIN(NULL, "Kalman(" + name + ")::input(vector)::y"), |
| 30 |
|
✗ |
modelTransitionSIN(NULL, "Kalman(" + name + ")::input(matrix)::F"), |
| 31 |
|
✗ |
modelMeasureSIN(NULL, "Kalman(" + name + ")::input(matrix)::H"), |
| 32 |
|
✗ |
noiseTransitionSIN(NULL, "Kalman(" + name + ")::input(matrix)::Q"), |
| 33 |
|
✗ |
noiseMeasureSIN(NULL, "Kalman(" + name + ")::input(matrix)::R") |
| 34 |
|
|
|
| 35 |
|
|
, |
| 36 |
|
✗ |
statePredictedSIN(0, "Kalman(" + name + ")::input(vector)::x_pred"), |
| 37 |
|
✗ |
observationPredictedSIN(0, "Kalman(" + name + ")::input(vector)::y_pred"), |
| 38 |
|
✗ |
varianceUpdateSOUT("Kalman(" + name + ")::output(vector)::P"), |
| 39 |
|
✗ |
stateUpdateSOUT("Kalman(" + name + ")::output(vector)::x_est"), |
| 40 |
|
✗ |
stateEstimation_(), |
| 41 |
|
✗ |
stateVariance_() { |
| 42 |
|
|
sotDEBUGIN(15); |
| 43 |
|
✗ |
varianceUpdateSOUT.setFunction( |
| 44 |
|
|
boost::bind(&Kalman::computeVarianceUpdate, this, _1, _2)); |
| 45 |
|
✗ |
stateUpdateSOUT.setFunction( |
| 46 |
|
|
boost::bind(&Kalman::computeStateUpdate, this, _1, _2)); |
| 47 |
|
|
|
| 48 |
|
✗ |
signalRegistration(measureSIN << observationPredictedSIN << modelTransitionSIN |
| 49 |
|
✗ |
<< modelMeasureSIN << noiseTransitionSIN |
| 50 |
|
✗ |
<< noiseMeasureSIN << statePredictedSIN |
| 51 |
|
✗ |
<< stateUpdateSOUT << varianceUpdateSOUT); |
| 52 |
|
|
|
| 53 |
|
|
std::string docstring = |
| 54 |
|
|
" Set initial state estimation\n" |
| 55 |
|
|
"\n" |
| 56 |
|
|
" input:\n" |
| 57 |
|
✗ |
" - a vector: initial state\n"; |
| 58 |
|
✗ |
addCommand("setInitialState", |
| 59 |
|
|
new Setter<Kalman, Vector>(*this, &Kalman::setStateEstimation, |
| 60 |
|
✗ |
docstring)); |
| 61 |
|
|
|
| 62 |
|
|
docstring = |
| 63 |
|
|
" Set variance of initial state estimation\n" |
| 64 |
|
|
"\n" |
| 65 |
|
|
" input:\n" |
| 66 |
|
✗ |
" - a matrix: variance covariance matrix\n"; |
| 67 |
|
✗ |
addCommand( |
| 68 |
|
|
"setInitialVariance", |
| 69 |
|
✗ |
new Setter<Kalman, Matrix>(*this, &Kalman::setStateVariance, docstring)); |
| 70 |
|
|
sotDEBUGOUT(15); |
| 71 |
|
|
} |
| 72 |
|
|
|
| 73 |
|
✗ |
Matrix &Kalman::computeVarianceUpdate(Matrix &Pk_k, const int &time) { |
| 74 |
|
|
sotDEBUGIN(15); |
| 75 |
|
✗ |
if (time == 0) { |
| 76 |
|
|
// First time return variance initial state |
| 77 |
|
✗ |
Pk_k = stateVariance_; |
| 78 |
|
|
// Set dependency to input signals for latter computations |
| 79 |
|
✗ |
varianceUpdateSOUT.addDependency(noiseTransitionSIN); |
| 80 |
|
✗ |
varianceUpdateSOUT.addDependency(modelTransitionSIN); |
| 81 |
|
|
} else { |
| 82 |
|
✗ |
const Matrix &Q = noiseTransitionSIN(time); |
| 83 |
|
✗ |
const Matrix &R = noiseMeasureSIN(time); |
| 84 |
|
✗ |
const Matrix &F = modelTransitionSIN(time); |
| 85 |
|
✗ |
const Matrix &H = modelMeasureSIN(time); |
| 86 |
|
✗ |
const Matrix &Pk_1_k_1 = stateVariance_; |
| 87 |
|
|
|
| 88 |
|
|
sotDEBUG(15) << "Q=" << Q << std::endl; |
| 89 |
|
|
sotDEBUG(15) << "R=" << R << std::endl; |
| 90 |
|
|
sotDEBUG(15) << "F=" << F << std::endl; |
| 91 |
|
|
sotDEBUG(15) << "H=" << H << std::endl; |
| 92 |
|
|
sotDEBUG(15) << "Pk_1_k_1=" << Pk_1_k_1 << std::endl; |
| 93 |
|
|
|
| 94 |
|
✗ |
FP_ = F * Pk_1_k_1; |
| 95 |
|
✗ |
Pk_k_1_ = FP_ * F.transpose(); |
| 96 |
|
✗ |
Pk_k_1_ += Q; |
| 97 |
|
|
|
| 98 |
|
|
sotDEBUG(15) << "F " << std::endl << F << std::endl; |
| 99 |
|
|
sotDEBUG(15) << "P_{k-1|k-1} " << std::endl << Pk_1_k_1 << std::endl; |
| 100 |
|
|
sotDEBUG(15) << "F^T " << std::endl << F.transpose() << std::endl; |
| 101 |
|
|
sotDEBUG(15) << "P_{k|k-1} " << std::endl << Pk_k_1_ << std::endl; |
| 102 |
|
|
|
| 103 |
|
✗ |
S_ = H * Pk_k_1_ * H.transpose() + R; |
| 104 |
|
✗ |
K_ = Pk_k_1_ * H.transpose() * S_.inverse(); |
| 105 |
|
✗ |
Pk_k = Pk_k_1_ - K_ * H * Pk_k_1_; |
| 106 |
|
|
|
| 107 |
|
|
sotDEBUG(15) << "S_{k} " << std::endl << S_ << std::endl; |
| 108 |
|
|
sotDEBUG(15) << "K_{k} " << std::endl << K_ << std::endl; |
| 109 |
|
|
sotDEBUG(15) << "P_{k|k} " << std::endl << Pk_k << std::endl; |
| 110 |
|
|
|
| 111 |
|
|
sotDEBUGOUT(15); |
| 112 |
|
|
|
| 113 |
|
✗ |
stateVariance_ = Pk_k; |
| 114 |
|
|
} |
| 115 |
|
✗ |
return Pk_k; |
| 116 |
|
|
} |
| 117 |
|
|
|
| 118 |
|
|
// ^ |
| 119 |
|
|
// z = y - h (x ) |
| 120 |
|
|
// k k k|k-1 |
| 121 |
|
|
// |
| 122 |
|
|
// ^ ^ |
| 123 |
|
|
// x = x + K z |
| 124 |
|
|
// k|k k|k-1 k k |
| 125 |
|
|
// |
| 126 |
|
|
// T |
| 127 |
|
|
// S = H P H + R |
| 128 |
|
|
// k k k|k-1 k |
| 129 |
|
|
// |
| 130 |
|
|
// T -1 |
| 131 |
|
|
// K = P H S |
| 132 |
|
|
// k k|k-1 k k |
| 133 |
|
|
// |
| 134 |
|
|
// P = (I - K H ) P |
| 135 |
|
|
// k|k k k k|k-1 |
| 136 |
|
|
|
| 137 |
|
✗ |
Vector &Kalman::computeStateUpdate(Vector &x_est, const int &time) { |
| 138 |
|
|
sotDEBUGIN(15); |
| 139 |
|
✗ |
if (time == 0) { |
| 140 |
|
|
// First time return variance initial state |
| 141 |
|
✗ |
x_est = stateEstimation_; |
| 142 |
|
|
// Set dependency to input signals for latter computations |
| 143 |
|
✗ |
stateUpdateSOUT.addDependency(measureSIN); |
| 144 |
|
✗ |
stateUpdateSOUT.addDependency(observationPredictedSIN); |
| 145 |
|
✗ |
stateUpdateSOUT.addDependency(modelMeasureSIN); |
| 146 |
|
✗ |
stateUpdateSOUT.addDependency(noiseTransitionSIN); |
| 147 |
|
✗ |
stateUpdateSOUT.addDependency(noiseMeasureSIN); |
| 148 |
|
✗ |
stateUpdateSOUT.addDependency(statePredictedSIN); |
| 149 |
|
✗ |
stateUpdateSOUT.addDependency(varianceUpdateSOUT); |
| 150 |
|
|
} else { |
| 151 |
|
✗ |
varianceUpdateSOUT.recompute(time); |
| 152 |
|
✗ |
const Vector &x_pred = statePredictedSIN(time); |
| 153 |
|
✗ |
const Vector &y_pred = observationPredictedSIN(time); |
| 154 |
|
✗ |
const Vector &y = measureSIN(time); |
| 155 |
|
|
|
| 156 |
|
|
sotDEBUG(25) << "K_{k} = " << std::endl << K_ << std::endl; |
| 157 |
|
|
sotDEBUG(25) << "y = " << y << std::endl; |
| 158 |
|
|
sotDEBUG(25) << "h (\\hat{x}_{k|k-1}) = " << y_pred << std::endl; |
| 159 |
|
|
sotDEBUG(25) << "y = " << y << std::endl; |
| 160 |
|
|
|
| 161 |
|
|
// Innovation: z_ = y - Hx |
| 162 |
|
✗ |
z_ = y - y_pred; |
| 163 |
|
|
// x_est = x_pred + (K*(y-(H*x_pred))); |
| 164 |
|
✗ |
x_est = x_pred + K_ * z_; |
| 165 |
|
|
|
| 166 |
|
|
sotDEBUG(25) << "z_{k} = " << z_ << std::endl; |
| 167 |
|
|
sotDEBUG(25) << "x_{k|k} = " << x_est << std::endl; |
| 168 |
|
|
|
| 169 |
|
✗ |
stateEstimation_ = x_est; |
| 170 |
|
|
} |
| 171 |
|
|
sotDEBUGOUT(15); |
| 172 |
|
✗ |
return x_est; |
| 173 |
|
|
} |
| 174 |
|
|
|
| 175 |
|
|
/* -------------------------------------------------------------------------- */ |
| 176 |
|
|
/* --- MODELE --------------------------------------------------------------- */ |
| 177 |
|
|
/* -------------------------------------------------------------------------- */ |
| 178 |
|
|
|
| 179 |
|
✗ |
void Kalman::display(std::ostream &) const {} |
| 180 |
|
|
|
| 181 |
|
|
} // namespace sot |
| 182 |
|
|
} // namespace dynamicgraph |
| 183 |
|
|
|
| 184 |
|
|
/*! |
| 185 |
|
|
\file Kalman.cpp |
| 186 |
|
|
\brief Generic kalman filtering implementation |
| 187 |
|
|
*/ |
| 188 |
|
|
|