diff --git a/examples/PlanarSLAMExample.cpp b/examples/PlanarSLAMExample.cpp index f05ffa9b2..c6f6b636f 100644 --- a/examples/PlanarSLAMExample.cpp +++ b/examples/PlanarSLAMExample.cpp @@ -11,22 +11,12 @@ /** * @file PlanarSLAMExample.cpp - * @brief Simple robotics example using the pre-built planar SLAM domain + * @brief Simple robotics example using odometry measurements and bearing-range (laser) measurements * @author Alex Cunningham */ -// pull in the planar SLAM domain with all typedefs and helper functions defined -#include - -// we will use Symbol keys -#include - -using namespace std; -using namespace gtsam; -using namespace gtsam::noiseModel; - /** - * Example of a simple 2D planar slam example with landmarls + * A simple 2D planar slam example with landmarks * - The robot and landmarks are on a 2 meter grid * - Robot poses are facing along the X axis (horizontal, to the right in 2D) * - The robot moves 2 meters each step @@ -34,29 +24,74 @@ using namespace gtsam::noiseModel; * - We have bearing and range information for measurements * - Landmarks are 2 meters away from the robot trajectory */ + +// As this is a planar SLAM example, we will use Pose2 variables (x, y, theta) to represent +// the robot positions and Point2 variables (x, y) to represent the landmark coordinates. +#include +#include + +// Each variable in the system (poses and landmarks) must be identified with a unique key. +// We can either use simple integer keys (1, 2, 3, ...) or symbols (X1, X2, L1). +// Here we will use Symbols +#include + +// In GTSAM, measurement functions are represented as 'factors'. Several common factors +// have been provided with the library for solving robotics/SLAM/Bundle Adjustment problems. +// Here we will use a RangeBearing factor for the range-bearing measurements to identified +// landmarks, and Between factors for the relative motion described by odometry measurements. +// Also, we will initialize the robot at the origin using a Prior factor. +#include +#include +#include + +// When the factors are created, we will add them to a Factor Graph. As the factors we are using +// are nonlinear factors, we will need a Nonlinear Factor Graph. +#include + +// Finally, once all of the factors have been added to our factor graph, we will want to +// solve/optimize to graph to find the best (Maximum A Posteriori) set of variable values. +// GTSAM includes several nonlinear optimizers to perform this step. Here we will use the +// common Levenberg-Marquardt solver +#include + +// Once the optimized values have been calculated, we can also calculate the marginal covariance +// of desired variables +#include + +// The nonlinear solvers within GTSAM are iterative solvers, meaning they linearize the +// nonlinear functions around an initial linearization point, then solve the linear system +// to update the linearization point. This happens repeatedly until the solver converges +// to a consistent set of variable values. This requires us to specify an initial guess +// for each variable, held in a Values container. +#include + + +using namespace std; +using namespace gtsam; + int main(int argc, char** argv) { - // create the graph (defined in planarSlam.h, derived from NonlinearFactorGraph) - planarSLAM::Graph graph; + // Create a factor graph + NonlinearFactorGraph graph; - // Create some keys - static Symbol i1('x',1), i2('x',2), i3('x',3); - static Symbol j1('l',1), j2('l',2); + // Create the keys we need for this simple example + static Symbol x1('x',1), x2('x',2), x3('x',3); + static Symbol l1('l',1), l2('l',2); - // add a Gaussian prior on pose x_1 - Pose2 priorMean(0.0, 0.0, 0.0); // prior mean is at origin - SharedDiagonal priorNoise = Diagonal::Sigmas(Vector_(3, 0.3, 0.3, 0.1)); // 30cm std on x,y, 0.1 rad on theta - graph.addPosePrior(i1, priorMean, priorNoise); // add directly to graph + // Add a prior on pose x1 at the origin. A prior factor consists of a mean and a noise model (covariance matrix) + Pose2 prior(0.0, 0.0, 0.0); // prior mean is at origin + noiseModel::Diagonal::shared_ptr priorNoise = noiseModel::Diagonal::Sigmas(Vector_(3, 0.3, 0.3, 0.1)); // 30cm std on x,y, 0.1 rad on theta + graph.add(PriorFactor(x1, prior, priorNoise)); // add directly to graph - // add two odometry factors + // Add two odometry factors Pose2 odometry(2.0, 0.0, 0.0); // create a measurement for both factors (the same in this case) - SharedDiagonal odometryNoise = Diagonal::Sigmas(Vector_(3, 0.2, 0.2, 0.1)); // 20cm std on x,y, 0.1 rad on theta - graph.addRelativePose(i1, i2, odometry, odometryNoise); - graph.addRelativePose(i2, i3, odometry, odometryNoise); + noiseModel::Diagonal::shared_ptr odometryNoise = noiseModel::Diagonal::Sigmas(Vector_(3, 0.2, 0.2, 0.1)); // 20cm std on x,y, 0.1 rad on theta + graph.add(BetweenFactor(x1, x2, odometry, odometryNoise)); + graph.add(BetweenFactor(x2, x3, odometry, odometryNoise)); + // Add Range-Bearing measurements to two different landmarks // create a noise model for the landmark measurements - SharedDiagonal measurementNoise = Diagonal::Sigmas(Vector_(2, 0.1, 0.2)); // 0.1 rad std on bearing, 20cm on range - + noiseModel::Diagonal::shared_ptr measurementNoise = noiseModel::Diagonal::Sigmas(Vector_(2, 0.1, 0.2)); // 0.1 rad std on bearing, 20cm on range // create the measurement values - indices are (pose id, landmark id) Rot2 bearing11 = Rot2::fromDegrees(45), bearing21 = Rot2::fromDegrees(90), @@ -65,27 +100,42 @@ int main(int argc, char** argv) { range21 = 2.0, range32 = 2.0; - // add bearing/range factors (created by "addBearingRange") - graph.addBearingRange(i1, j1, bearing11, range11, measurementNoise); - graph.addBearingRange(i2, j1, bearing21, range21, measurementNoise); - graph.addBearingRange(i3, j2, bearing32, range32, measurementNoise); + // Add Bearing-Range factors + graph.add(BearingRangeFactor(x1, l1, bearing11, range11, measurementNoise)); + graph.add(BearingRangeFactor(x2, l1, bearing21, range21, measurementNoise)); + graph.add(BearingRangeFactor(x3, l2, bearing32, range32, measurementNoise)); - // print - graph.print("Factor graph"); + // Print + graph.print("Factor Graph:\n"); - // create (deliberatly inaccurate) initial estimate - planarSLAM::Values initialEstimate; - initialEstimate.insertPose(i1, Pose2(0.5, 0.0, 0.2)); - initialEstimate.insertPose(i2, Pose2(2.3, 0.1,-0.2)); - initialEstimate.insertPose(i3, Pose2(4.1, 0.1, 0.1)); - initialEstimate.insertPoint(j1, Point2(1.8, 2.1)); - initialEstimate.insertPoint(j2, Point2(4.1, 1.8)); + // Create (deliberately inaccurate) initial estimate + Values initialEstimate; + initialEstimate.insert(x1, Pose2(0.5, 0.0, 0.2)); + initialEstimate.insert(x2, Pose2(2.3, 0.1,-0.2)); + initialEstimate.insert(x3, Pose2(4.1, 0.1, 0.1)); + initialEstimate.insert(l1, Point2(1.8, 2.1)); + initialEstimate.insert(l2, Point2(4.1, 1.8)); - initialEstimate.print("Initial estimate:\n "); + // Print + initialEstimate.print("Initial Estimate:\n"); - // optimize using Levenberg-Marquardt optimization with an ordering from colamd - planarSLAM::Values result = graph.optimize(initialEstimate); - result.print("Final result:\n "); + // Optimize using Levenberg-Marquardt optimization. The optimizer + // accepts an optional set of configuration parameters, controlling + // things like convergence criteria, the type of linear system solver + // to use, and the amount of information displayed during optimization. + // Here we will use the default set of parameters. See the + // documentation for the full set of parameters. + LevenbergMarquardtOptimizer optimizer(graph, initialEstimate); + Values result = optimizer.optimize(); + result.print("Final Result:\n"); + + // Calculate and print marginal covariances for all variables + Marginals marginals(graph, result); + print(marginals.marginalCovariance(x1), "x1 covariance"); + print(marginals.marginalCovariance(x2), "x2 covariance"); + print(marginals.marginalCovariance(x3), "x3 covariance"); + print(marginals.marginalCovariance(l1), "l1 covariance"); + print(marginals.marginalCovariance(l2), "l2 covariance"); return 0; } diff --git a/examples/PlanarSLAMExample_selfcontained.cpp b/examples/PlanarSLAMExample_selfcontained.cpp deleted file mode 100644 index 7c8ccc134..000000000 --- a/examples/PlanarSLAMExample_selfcontained.cpp +++ /dev/null @@ -1,137 +0,0 @@ -/* ---------------------------------------------------------------------------- - - * GTSAM Copyright 2010, Georgia Tech Research Corporation, - * Atlanta, Georgia 30332-0415 - * All Rights Reserved - * Authors: Frank Dellaert, et al. (see THANKS for the full author list) - - * See LICENSE for the license information - - * -------------------------------------------------------------------------- */ - -/** - * @file PlanarSLAMExample_selfcontained.cpp - * @brief Simple robotics example with all typedefs internal to this script. - * @author Alex Cunningham - */ - -// add in headers for specific factors -#include -#include -#include - -// for all nonlinear keys -#include - -// implementations for structures - needed if self-contained, and these should be included last -#include -#include -#include - -// for modeling measurement uncertainty - all models included here -#include - -// for points and poses -#include -#include - -#include -#include - -using namespace std; -using namespace gtsam; - -/** - * In this version of the system we make the following assumptions: - * - All values are axis aligned - * - Robot poses are facing along the X axis (horizontal, to the right in images) - * - We have bearing and range information for measurements - * - We have full odometry for measurements - * - The robot and landmarks are on a grid, moving 2 meters each step - * - Landmarks are 2 meters away from the robot trajectory - */ -int main(int argc, char** argv) { - // create keys for variables - Symbol i1('x',1), i2('x',2), i3('x',3); - Symbol j1('l',1), j2('l',2); - - // create graph container and add factors to it - NonlinearFactorGraph graph; - - /* add prior */ - // gaussian for prior - SharedDiagonal priorNoise = noiseModel::Diagonal::Sigmas(Vector_(3, 0.3, 0.3, 0.1)); - Pose2 priorMean(0.0, 0.0, 0.0); // prior at origin - PriorFactor posePrior(i1, priorMean, priorNoise); // create the factor - graph.add(posePrior); // add the factor to the graph - - /* add odometry */ - // general noisemodel for odometry - SharedDiagonal odometryNoise = noiseModel::Diagonal::Sigmas(Vector_(3, 0.2, 0.2, 0.1)); - Pose2 odometry(2.0, 0.0, 0.0); // create a measurement for both factors (the same in this case) - // create between factors to represent odometry - BetweenFactor odom12(i1, i2, odometry, odometryNoise); - BetweenFactor odom23(i2, i3, odometry, odometryNoise); - graph.add(odom12); // add both to graph - graph.add(odom23); - - /* add measurements */ - // general noisemodel for measurements - SharedDiagonal measurementNoise = noiseModel::Diagonal::Sigmas(Vector_(2, 0.1, 0.2)); - - // create the measurement values - indices are (pose id, landmark id) - Rot2 bearing11 = Rot2::fromDegrees(45), - bearing21 = Rot2::fromDegrees(90), - bearing32 = Rot2::fromDegrees(90); - double range11 = sqrt(4.0+4.0), - range21 = 2.0, - range32 = 2.0; - - // create bearing/range factors - BearingRangeFactor meas11(i1, j1, bearing11, range11, measurementNoise); - BearingRangeFactor meas21(i2, j1, bearing21, range21, measurementNoise); - BearingRangeFactor meas32(i3, j2, bearing32, range32, measurementNoise); - - // add the factors - graph.add(meas11); - graph.add(meas21); - graph.add(meas32); - - graph.print("Full Graph"); - - // initialize to noisy points - Values initial; - initial.insert(i1, Pose2(0.5, 0.0, 0.2)); - initial.insert(i2, Pose2(2.3, 0.1,-0.2)); - initial.insert(i3, Pose2(4.1, 0.1, 0.1)); - initial.insert(j1, Point2(1.8, 2.1)); - initial.insert(j2, Point2(4.1, 1.8)); - - initial.print("initial estimate"); - - // optimize using Levenberg-Marquardt optimization with an ordering from colamd - - // first using sequential elimination - LevenbergMarquardtParams lmParams; - lmParams.linearSolverType = LevenbergMarquardtParams::SEQUENTIAL_CHOLESKY; - Values resultSequential = LevenbergMarquardtOptimizer(graph, initial, lmParams).optimize(); - resultSequential.print("final result (solved with a sequential solver)"); - - // then using multifrontal, advanced interface - // Note that we keep the original optimizer object so we can use the COLAMD - // ordering it computes. - LevenbergMarquardtOptimizer optimizer(graph, initial); - Values resultMultifrontal = optimizer.optimize(); - resultMultifrontal.print("final result (solved with a multifrontal solver)"); - - // Print marginals covariances for all variables - Marginals marginals(graph, resultMultifrontal, Marginals::CHOLESKY); - print(marginals.marginalCovariance(i1), "i1 covariance"); - print(marginals.marginalCovariance(i2), "i2 covariance"); - print(marginals.marginalCovariance(i3), "i3 covariance"); - print(marginals.marginalCovariance(j1), "j1 covariance"); - print(marginals.marginalCovariance(j2), "j2 covariance"); - - return 0; -} -