| 1 | /* ---------------------------------------------------------------------------- |
| 2 | |
| 3 | * GTSAM Copyright 2010, Georgia Tech Research Corporation, |
| 4 | * Atlanta, Georgia 30332-0415 |
| 5 | * All Rights Reserved |
| 6 | * Authors: Frank Dellaert, et al. (see THANKS for the full author list) |
| 7 | |
| 8 | * See LICENSE for the license information |
| 9 | |
| 10 | * -------------------------------------------------------------------------- */ |
| 11 | |
| 12 | /** |
| 13 | * @file GaussMarkov1stOrderFactor.h |
| 14 | * @author Vadim Indelman, Stephen Williams, Luca Carlone |
| 15 | * @date Jan 17, 2012 |
| 16 | **/ |
| 17 | #pragma once |
| 18 | |
| 19 | #include <gtsam/nonlinear/NonlinearFactor.h> |
| 20 | #include <gtsam/linear/GaussianFactor.h> |
| 21 | #include <gtsam/linear/NoiseModel.h> |
| 22 | #include <gtsam/base/Testable.h> |
| 23 | #include <gtsam/base/Lie.h> |
| 24 | |
| 25 | #include <ostream> |
| 26 | |
| 27 | namespace gtsam { |
| 28 | |
| 29 | /* |
| 30 | * - The 1st order GaussMarkov factor relates two keys of the same type. This relation is given via |
| 31 | * key_2 = exp(-1/tau*delta_t) * key1 + w_d |
| 32 | * where tau is the time constant and delta_t is the time difference between the two keys. |
| 33 | * w_d is the equivalent discrete noise, whose covariance is calculated from the continuous noise model and delta_t. |
| 34 | * - w_d is approximated as a Gaussian noise. |
| 35 | * - In the multi-dimensional case, tau is a vector, and the above equation is applied on each element |
| 36 | * in the state (represented by keys), using the appropriate time constant in the vector tau. |
| 37 | */ |
| 38 | |
| 39 | /* |
| 40 | * A class for a measurement predicted by "GaussMarkov1stOrderFactor(config[key1],config[key2])" |
| 41 | * KEY1::Value is the Lie Group type |
| 42 | * T is the measurement type, by default the same |
| 43 | */ |
| 44 | template<class VALUE> |
| 45 | class GaussMarkov1stOrderFactor: public NoiseModelFactorN<VALUE, VALUE> { |
| 46 | |
| 47 | private: |
| 48 | |
| 49 | typedef GaussMarkov1stOrderFactor<VALUE> This; |
| 50 | typedef NoiseModelFactorN<VALUE, VALUE> Base; |
| 51 | |
| 52 | double dt_; |
| 53 | Vector tau_; |
| 54 | |
| 55 | public: |
| 56 | |
| 57 | // Provide access to the Matrix& version of evaluateError: |
| 58 | using Base::evaluateError; |
| 59 | |
| 60 | // shorthand for a smart pointer to a factor |
| 61 | typedef typename std::shared_ptr<GaussMarkov1stOrderFactor> shared_ptr; |
| 62 | |
| 63 | /** default constructor - only use for serialization */ |
| 64 | GaussMarkov1stOrderFactor() {} |
| 65 | |
| 66 | /** Constructor */ |
| 67 | GaussMarkov1stOrderFactor(const Key& key1, const Key& key2, double delta_t, Vector tau, |
| 68 | const SharedGaussian& model) : |
| 69 | Base(calcDiscreteNoiseModel(model, delta_t), key1, key2), dt_(delta_t), tau_(tau) { |
| 70 | } |
| 71 | |
| 72 | ~GaussMarkov1stOrderFactor() override {} |
| 73 | |
| 74 | /** implement functions needed for Testable */ |
| 75 | |
| 76 | /** print */ |
| 77 | void print(const std::string& s, const KeyFormatter& keyFormatter = DefaultKeyFormatter) const override { |
| 78 | std::cout << s << "GaussMarkov1stOrderFactor(" |
| 79 | << keyFormatter(this->key1()) << "," |
| 80 | << keyFormatter(this->key2()) << ")\n" ; |
| 81 | this->noiseModel_->print(" noise model" ); |
| 82 | } |
| 83 | |
| 84 | /** equals */ |
| 85 | bool equals(const NonlinearFactor& expected, double tol=1e-9) const override { |
| 86 | const This *e = dynamic_cast<const This*> (&expected); |
| 87 | return e != nullptr && Base::equals(*e, tol); |
| 88 | } |
| 89 | |
| 90 | /** implement functions needed to derive from Factor */ |
| 91 | |
| 92 | /** vector of errors */ |
| 93 | Vector evaluateError(const VALUE& p1, const VALUE& p2, |
| 94 | OptionalMatrixType H1, OptionalMatrixType H2) const override { |
| 95 | |
| 96 | Vector v1( traits<VALUE>::Logmap(p1) ); |
| 97 | Vector v2( traits<VALUE>::Logmap(p2) ); |
| 98 | |
| 99 | Vector alpha(tau_.size()); |
| 100 | Vector alpha_v1(tau_.size()); |
| 101 | for(int i=0; i<tau_.size(); i++){ |
| 102 | alpha(i) = exp(x: - 1/tau_(i)*dt_ ); |
| 103 | alpha_v1(i) = alpha(i) * v1(i); |
| 104 | } |
| 105 | |
| 106 | Vector hx(v2 - alpha_v1); |
| 107 | |
| 108 | if(H1) *H1 = -1 * alpha.asDiagonal(); |
| 109 | if(H2) *H2 = Matrix::Identity(rows: v2.size(),cols: v2.size()); |
| 110 | |
| 111 | return hx; |
| 112 | } |
| 113 | |
| 114 | private: |
| 115 | |
| 116 | #if GTSAM_ENABLE_BOOST_SERIALIZATION |
| 117 | /** Serialization function */ |
| 118 | friend class boost::serialization::access; |
| 119 | template<class ARCHIVE> |
| 120 | void serialize(ARCHIVE & ar, const unsigned int /*version*/) { |
| 121 | ar & BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base); |
| 122 | ar & BOOST_SERIALIZATION_NVP(dt_); |
| 123 | ar & BOOST_SERIALIZATION_NVP(tau_); |
| 124 | } |
| 125 | #endif |
| 126 | |
| 127 | SharedGaussian calcDiscreteNoiseModel(const SharedGaussian& model, double delta_t){ |
| 128 | /* Q_d (approx)= Q * delta_t */ |
| 129 | /* In practice, square root of the information matrix is represented, so that: |
| 130 | * R_d (approx)= R / sqrt(delta_t) |
| 131 | * */ |
| 132 | noiseModel::Gaussian::shared_ptr gaussian_model = std::dynamic_pointer_cast<noiseModel::Gaussian>(r: model); |
| 133 | SharedGaussian model_d(noiseModel::Gaussian::SqrtInformation(R: gaussian_model->R()/sqrt(x: delta_t))); |
| 134 | return model_d; |
| 135 | } |
| 136 | |
| 137 | }; // \class GaussMarkov1stOrderFactor |
| 138 | |
| 139 | /// traits |
| 140 | template<class VALUE> struct traits<GaussMarkov1stOrderFactor<VALUE> > : |
| 141 | public Testable<GaussMarkov1stOrderFactor<VALUE> > { |
| 142 | }; |
| 143 | |
| 144 | } /// namespace gtsam |
| 145 | |