| 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 LocalizationExample.cpp |
| 14 | * @brief Simple robot localization example, with three "GPS-like" measurements |
| 15 | * @author Frank Dellaert |
| 16 | */ |
| 17 | |
| 18 | /** |
| 19 | * A simple 2D pose slam example with "GPS" measurements |
| 20 | * - The robot moves forward 2 meter each iteration |
| 21 | * - The robot initially faces along the X axis (horizontal, to the right in 2D) |
| 22 | * - We have full odometry between pose |
| 23 | * - We have "GPS-like" measurements implemented with a custom factor |
| 24 | */ |
| 25 | |
| 26 | // We will use Pose2 variables (x, y, theta) to represent the robot positions |
| 27 | #include <gtsam/geometry/Pose2.h> |
| 28 | |
| 29 | // We will use simple integer Keys to refer to the robot poses. |
| 30 | #include <gtsam/inference/Key.h> |
| 31 | |
| 32 | // As in OdometryExample.cpp, we use a BetweenFactor to model odometry measurements. |
| 33 | #include <gtsam/slam/BetweenFactor.h> |
| 34 | |
| 35 | // We add all facors to a Nonlinear Factor Graph, as our factors are nonlinear. |
| 36 | #include <gtsam/nonlinear/NonlinearFactorGraph.h> |
| 37 | |
| 38 | // The nonlinear solvers within GTSAM are iterative solvers, meaning they linearize the |
| 39 | // nonlinear functions around an initial linearization point, then solve the linear system |
| 40 | // to update the linearization point. This happens repeatedly until the solver converges |
| 41 | // to a consistent set of variable values. This requires us to specify an initial guess |
| 42 | // for each variable, held in a Values container. |
| 43 | #include <gtsam/nonlinear/Values.h> |
| 44 | |
| 45 | // Finally, once all of the factors have been added to our factor graph, we will want to |
| 46 | // solve/optimize to graph to find the best (Maximum A Posteriori) set of variable values. |
| 47 | // GTSAM includes several nonlinear optimizers to perform this step. Here we will use the |
| 48 | // standard Levenberg-Marquardt solver |
| 49 | #include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h> |
| 50 | |
| 51 | // Once the optimized values have been calculated, we can also calculate the marginal covariance |
| 52 | // of desired variables |
| 53 | #include <gtsam/nonlinear/Marginals.h> |
| 54 | |
| 55 | using namespace std; |
| 56 | using namespace gtsam; |
| 57 | |
| 58 | // Before we begin the example, we must create a custom unary factor to implement a |
| 59 | // "GPS-like" functionality. Because standard GPS measurements provide information |
| 60 | // only on the position, and not on the orientation, we cannot use a simple prior to |
| 61 | // properly model this measurement. |
| 62 | // |
| 63 | // The factor will be a unary factor, affect only a single system variable. It will |
| 64 | // also use a standard Gaussian noise model. Hence, we will derive our new factor from |
| 65 | // the NoiseModelFactorN. |
| 66 | #include <gtsam/nonlinear/NonlinearFactor.h> |
| 67 | |
| 68 | class UnaryFactor: public NoiseModelFactorN<Pose2> { |
| 69 | // The factor will hold a measurement consisting of an (X,Y) location |
| 70 | // We could this with a Point2 but here we just use two doubles |
| 71 | double mx_, my_; |
| 72 | |
| 73 | public: |
| 74 | |
| 75 | // Provide access to Matrix& version of evaluateError: |
| 76 | using NoiseModelFactor1<Pose2>::evaluateError; |
| 77 | |
| 78 | /// shorthand for a smart pointer to a factor |
| 79 | typedef std::shared_ptr<UnaryFactor> shared_ptr; |
| 80 | |
| 81 | // The constructor requires the variable key, the (X, Y) measurement value, and the noise model |
| 82 | UnaryFactor(Key j, double x, double y, const SharedNoiseModel& model): |
| 83 | NoiseModelFactorN<Pose2>(model, j), mx_(x), my_(y) {} |
| 84 | |
| 85 | ~UnaryFactor() override {} |
| 86 | |
| 87 | // Using the NoiseModelFactorN base class there are two functions that must be overridden. |
| 88 | // The first is the 'evaluateError' function. This function implements the desired measurement |
| 89 | // function, returning a vector of errors when evaluated at the provided variable value. It |
| 90 | // must also calculate the Jacobians for this measurement function, if requested. |
| 91 | Vector evaluateError(const Pose2& q, OptionalMatrixType H) const override { |
| 92 | // The measurement function for a GPS-like measurement h(q) which predicts the measurement (m) is h(q) = q, q = [qx qy qtheta] |
| 93 | // The error is then simply calculated as E(q) = h(q) - m: |
| 94 | // error_x = q.x - mx |
| 95 | // error_y = q.y - my |
| 96 | // Node's orientation reflects in the Jacobian, in tangent space this is equal to the right-hand rule rotation matrix |
| 97 | // H = [ cos(q.theta) -sin(q.theta) 0 ] |
| 98 | // [ sin(q.theta) cos(q.theta) 0 ] |
| 99 | const Rot2& R = q.rotation(); |
| 100 | if (H) (*H) = (gtsam::Matrix(2, 3) << R.c(), -R.s(), 0.0, R.s(), R.c(), 0.0).finished(); |
| 101 | return (Vector(2) << q.x() - mx_, q.y() - my_).finished(); |
| 102 | } |
| 103 | |
| 104 | // The second is a 'clone' function that allows the factor to be copied. Under most |
| 105 | // circumstances, the following code that employs the default copy constructor should |
| 106 | // work fine. |
| 107 | gtsam::NonlinearFactor::shared_ptr clone() const override { |
| 108 | return std::static_pointer_cast<gtsam::NonlinearFactor>( |
| 109 | r: gtsam::NonlinearFactor::shared_ptr(new UnaryFactor(*this))); } |
| 110 | |
| 111 | // Additionally, we encourage you the use of unit testing your custom factors, |
| 112 | // (as all GTSAM factors are), in which you would need an equals and print, to satisfy the |
| 113 | // GTSAM_CONCEPT_TESTABLE_INST(T) defined in Testable.h, but these are not needed below. |
| 114 | }; // UnaryFactor |
| 115 | |
| 116 | |
| 117 | int main(int argc, char** argv) { |
| 118 | // 1. Create a factor graph container and add factors to it |
| 119 | NonlinearFactorGraph graph; |
| 120 | |
| 121 | // 2a. Add odometry factors |
| 122 | // For simplicity, we will use the same noise model for each odometry factor |
| 123 | auto odometryNoise = noiseModel::Diagonal::Sigmas(sigmas: Vector3(0.2, 0.2, 0.1)); |
| 124 | // Create odometry (Between) factors between consecutive poses |
| 125 | graph.emplace_shared<BetweenFactor<Pose2> >(args: 1, args: 2, args: Pose2(2.0, 0.0, 0.0), args&: odometryNoise); |
| 126 | graph.emplace_shared<BetweenFactor<Pose2> >(args: 2, args: 3, args: Pose2(2.0, 0.0, 0.0), args&: odometryNoise); |
| 127 | |
| 128 | // 2b. Add "GPS-like" measurements |
| 129 | // We will use our custom UnaryFactor for this. |
| 130 | auto unaryNoise = |
| 131 | noiseModel::Diagonal::Sigmas(sigmas: Vector2(0.1, 0.1)); // 10cm std on x,y |
| 132 | graph.emplace_shared<UnaryFactor>(args: 1, args: 0.0, args: 0.0, args&: unaryNoise); |
| 133 | graph.emplace_shared<UnaryFactor>(args: 2, args: 2.0, args: 0.0, args&: unaryNoise); |
| 134 | graph.emplace_shared<UnaryFactor>(args: 3, args: 4.0, args: 0.0, args&: unaryNoise); |
| 135 | graph.print(str: "\nFactor Graph:\n" ); // print |
| 136 | |
| 137 | // 3. Create the data structure to hold the initialEstimate estimate to the solution |
| 138 | // For illustrative purposes, these have been deliberately set to incorrect values |
| 139 | Values initialEstimate; |
| 140 | initialEstimate.insert(j: 1, val: Pose2(0.5, 0.0, 0.2)); |
| 141 | initialEstimate.insert(j: 2, val: Pose2(2.3, 0.1, -0.2)); |
| 142 | initialEstimate.insert(j: 3, val: Pose2(4.1, 0.1, 0.1)); |
| 143 | initialEstimate.print(str: "\nInitial Estimate:\n" ); // print |
| 144 | |
| 145 | // 4. Optimize using Levenberg-Marquardt optimization. The optimizer |
| 146 | // accepts an optional set of configuration parameters, controlling |
| 147 | // things like convergence criteria, the type of linear system solver |
| 148 | // to use, and the amount of information displayed during optimization. |
| 149 | // Here we will use the default set of parameters. See the |
| 150 | // documentation for the full set of parameters. |
| 151 | LevenbergMarquardtOptimizer optimizer(graph, initialEstimate); |
| 152 | Values result = optimizer.optimize(); |
| 153 | result.print(str: "Final Result:\n" ); |
| 154 | |
| 155 | // 5. Calculate and print marginal covariances for all variables |
| 156 | Marginals marginals(graph, result); |
| 157 | cout << "x1 covariance:\n" << marginals.marginalCovariance(variable: 1) << endl; |
| 158 | cout << "x2 covariance:\n" << marginals.marginalCovariance(variable: 2) << endl; |
| 159 | cout << "x3 covariance:\n" << marginals.marginalCovariance(variable: 3) << endl; |
| 160 | |
| 161 | return 0; |
| 162 | } |
| 163 | |