| 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 testParticleFactor.cpp |
| 14 | * @brief Unit tests for the Particle factor |
| 15 | * @author Frank Dellaert |
| 16 | * @date Dec 9, 2013 |
| 17 | */ |
| 18 | |
| 19 | #include <gtsam/linear/NoiseModel.h> |
| 20 | |
| 21 | namespace gtsam { |
| 22 | |
| 23 | /** |
| 24 | * A factor approximating a density on a variable as a set of weighted particles |
| 25 | */ |
| 26 | template<class X> |
| 27 | class ParticleFactor { |
| 28 | |
| 29 | public: |
| 30 | typedef ParticleFactor This; ///< Typedef to this class |
| 31 | typedef std::shared_ptr<This> shared_ptr; ///< shared_ptr to this class |
| 32 | |
| 33 | }; |
| 34 | |
| 35 | /** |
| 36 | * A particle filter class, styled on functional KalmanFilter |
| 37 | * A state is created, which is functionally updates through [predict] and [update] |
| 38 | */ |
| 39 | template<class X> |
| 40 | class ParticleFilter { |
| 41 | |
| 42 | public: |
| 43 | |
| 44 | /** |
| 45 | * The Particle filter state is simply a ParticleFactor |
| 46 | */ |
| 47 | typedef typename ParticleFactor<X>::shared_ptr State; |
| 48 | |
| 49 | /** |
| 50 | * Create initial state, i.e., prior density at time k=0 |
| 51 | * In Bayes Filter notation, these are x_{0|0} and P_{0|0} |
| 52 | * @param x0 estimate at time 0 |
| 53 | * @param P0 covariance at time 0, given as a diagonal Gaussian 'model' |
| 54 | */ |
| 55 | State init(const Vector& x0, const SharedDiagonal& P0) { |
| 56 | return std::make_shared<ParticleFactor<X> >(); |
| 57 | } |
| 58 | |
| 59 | }; |
| 60 | // ParticleFilter |
| 61 | |
| 62 | }// namespace gtsam |
| 63 | |
| 64 | #include <gtsam/slam/BetweenFactor.h> |
| 65 | #include <gtsam/linear/KalmanFilter.h> |
| 66 | #include <gtsam/geometry/Pose2.h> |
| 67 | #include <CppUnitLite/TestHarness.h> |
| 68 | |
| 69 | using namespace std; |
| 70 | using namespace gtsam; |
| 71 | |
| 72 | //****************************************************************************** |
| 73 | |
| 74 | TEST( particleFactor, constructor ) { |
| 75 | // ParticleFactor<Pose2> pf; |
| 76 | //CHECK(assert_equal(expected, actual)); |
| 77 | } |
| 78 | |
| 79 | //****************************************************************************** |
| 80 | // Tests to do: |
| 81 | // Take two variables pf-x-*-y, eliminate x, multiply and sample then marginalize |
| 82 | TEST( particleFactor, eliminate) { |
| 83 | // ParticleFactor<Pose2> fx; |
| 84 | BetweenFactor<Pose2> fxy; |
| 85 | |
| 86 | } |
| 87 | |
| 88 | //****************************************************************************** |
| 89 | |
| 90 | /** Small 2D point class implemented as a Vector */ |
| 91 | struct State: Vector { |
| 92 | State(double x, double y) : |
| 93 | Vector((Vector(2) << x, y).finished()) { |
| 94 | } |
| 95 | }; |
| 96 | |
| 97 | //****************************************************************************** |
| 98 | TEST( ParticleFilter, constructor) { |
| 99 | |
| 100 | // Create a Kalman filter of dimension 2 |
| 101 | ParticleFilter<Pose2> pf1; |
| 102 | |
| 103 | // Create inital mean/covariance |
| 104 | State x_initial(0.0, 0.0); |
| 105 | SharedDiagonal P1 = noiseModel::Isotropic::Sigma(dim: 2, sigma: 0.1); |
| 106 | |
| 107 | // Get initial state by passing initial mean/covariance to the filter |
| 108 | ParticleFilter<Pose2>::State p1 = pf1.init(x0: x_initial, P0: P1); |
| 109 | |
| 110 | // // Assert it has the correct mean, covariance and information |
| 111 | // EXPECT(assert_equal(x_initial, p1->mean())); |
| 112 | // Matrix Sigma = (Mat(2, 2) << 0.01, 0.0, 0.0, 0.01); |
| 113 | // EXPECT(assert_equal(Sigma, p1->covariance())); |
| 114 | // EXPECT(assert_equal(inverse(Sigma), p1->information())); |
| 115 | // |
| 116 | // // Create one with a sharedGaussian |
| 117 | // KalmanFilter::State p2 = pf1.init(x_initial, Sigma); |
| 118 | // EXPECT(assert_equal(Sigma, p2->covariance())); |
| 119 | // |
| 120 | // // Now make sure both agree |
| 121 | // EXPECT(assert_equal(p1->covariance(), p2->covariance())); |
| 122 | } |
| 123 | |
| 124 | //****************************************************************************** |
| 125 | TEST( ParticleFilter, linear1 ) { |
| 126 | |
| 127 | // Create the controls and measurement properties for our example |
| 128 | Matrix F = I_2x2; |
| 129 | Matrix B = I_2x2; |
| 130 | Vector u = Vector2(1.0, 0.0); |
| 131 | SharedDiagonal modelQ = noiseModel::Isotropic::Sigma(dim: 2, sigma: 0.1); |
| 132 | Matrix Q = 0.01 * I_2x2; |
| 133 | Matrix H = I_2x2; |
| 134 | State z1(1.0, 0.0); |
| 135 | State z2(2.0, 0.0); |
| 136 | State z3(3.0, 0.0); |
| 137 | SharedDiagonal modelR = noiseModel::Isotropic::Sigma(dim: 2, sigma: 0.1); |
| 138 | Matrix R = 0.01 * I_2x2; |
| 139 | |
| 140 | // Create the set of expected output TestValues |
| 141 | State expected0(0.0, 0.0); |
| 142 | Matrix P00 = 0.01 * I_2x2; |
| 143 | |
| 144 | State expected1(1.0, 0.0); |
| 145 | Matrix P01 = P00 + Q; |
| 146 | Matrix I11 = P01.inverse() + R.inverse(); |
| 147 | |
| 148 | State expected2(2.0, 0.0); |
| 149 | Matrix P12 = I11.inverse() + Q; |
| 150 | Matrix I22 = P12.inverse() + R.inverse(); |
| 151 | |
| 152 | State expected3(3.0, 0.0); |
| 153 | Matrix P23 = I22.inverse() + Q; |
| 154 | Matrix I33 = P23.inverse() + R.inverse(); |
| 155 | |
| 156 | // Create a Kalman filter of dimension 2 |
| 157 | KalmanFilter kf(2); |
| 158 | |
| 159 | // Create the Kalman Filter initialization point |
| 160 | State x_initial(0.0, 0.0); |
| 161 | SharedDiagonal P_initial = noiseModel::Isotropic::Sigma(dim: 2, sigma: 0.1); |
| 162 | |
| 163 | // Create initial KalmanFilter object |
| 164 | KalmanFilter::State p0 = kf.init(x0: x_initial, P0: P_initial); |
| 165 | EXPECT(assert_equal(expected0, p0->mean())); |
| 166 | EXPECT(assert_equal(P00, p0->covariance())); |
| 167 | |
| 168 | // Run iteration 1 |
| 169 | KalmanFilter::State p1p = kf.predict(p: p0, F, B, u, modelQ); |
| 170 | EXPECT(assert_equal(expected1, p1p->mean())); |
| 171 | EXPECT(assert_equal(P01, p1p->covariance())); |
| 172 | |
| 173 | KalmanFilter::State p1 = kf.update(p: p1p, H, z: z1, model: modelR); |
| 174 | EXPECT(assert_equal(expected1, p1->mean())); |
| 175 | EXPECT(assert_equal(I11, p1->information())); |
| 176 | |
| 177 | // Run iteration 2 (with full covariance) |
| 178 | KalmanFilter::State p2p = kf.predictQ(p: p1, F, B, u, Q); |
| 179 | EXPECT(assert_equal(expected2, p2p->mean())); |
| 180 | |
| 181 | KalmanFilter::State p2 = kf.update(p: p2p, H, z: z2, model: modelR); |
| 182 | EXPECT(assert_equal(expected2, p2->mean())); |
| 183 | |
| 184 | // Run iteration 3 |
| 185 | KalmanFilter::State p3p = kf.predict(p: p2, F, B, u, modelQ); |
| 186 | EXPECT(assert_equal(expected3, p3p->mean())); |
| 187 | LONGS_EQUAL(3, (long)KalmanFilter::step(p3p)); |
| 188 | |
| 189 | KalmanFilter::State p3 = kf.update(p: p3p, H, z: z3, model: modelR); |
| 190 | EXPECT(assert_equal(expected3, p3->mean())); |
| 191 | LONGS_EQUAL(3, (long)KalmanFilter::step(p3)); |
| 192 | } |
| 193 | |
| 194 | //****************************************************************************** |
| 195 | int main() { |
| 196 | TestResult tr; |
| 197 | return TestRegistry::runAllTests(result&: tr); |
| 198 | } |
| 199 | //****************************************************************************** |
| 200 | |
| 201 | |