| 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 RangeISAMExample_plaza2.cpp |
| 14 | * @brief A 2D Range SLAM example |
| 15 | * @date June 20, 2013 |
| 16 | * @author Frank Dellaert |
| 17 | */ |
| 18 | |
| 19 | // Both relative poses and recovered trajectory poses will be stored as Pose2. |
| 20 | #include <gtsam/geometry/Pose2.h> |
| 21 | using gtsam::Pose2; |
| 22 | |
| 23 | // gtsam::Vectors are dynamic Eigen vectors, Vector3 is statically sized. |
| 24 | #include <gtsam/base/Vector.h> |
| 25 | using gtsam::Vector; |
| 26 | using gtsam::Vector3; |
| 27 | |
| 28 | // Unknown landmarks are of type Point2 (which is just a 2D Eigen vector). |
| 29 | #include <gtsam/geometry/Point2.h> |
| 30 | using gtsam::Point2; |
| 31 | |
| 32 | // Each variable in the system (poses and landmarks) must be identified with a |
| 33 | // unique key. We can either use simple integer keys (1, 2, 3, ...) or symbols |
| 34 | // (X1, X2, L1). Here we will use Symbols. |
| 35 | #include <gtsam/inference/Symbol.h> |
| 36 | using gtsam::Symbol; |
| 37 | |
| 38 | // We want to use iSAM2 to solve the range-SLAM problem incrementally. |
| 39 | #include <gtsam/nonlinear/ISAM2.h> |
| 40 | |
| 41 | // iSAM2 requires as input a set set of new factors to be added stored in a |
| 42 | // factor graph, and initial guesses for any new variables in the added factors. |
| 43 | #include <gtsam/nonlinear/NonlinearFactorGraph.h> |
| 44 | #include <gtsam/nonlinear/Values.h> |
| 45 | |
| 46 | // We will use a non-linear solver to batch-initialize from the first 150 frames |
| 47 | #include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h> |
| 48 | |
| 49 | // Measurement functions are represented as 'factors'. Several common factors |
| 50 | // have been provided with the library for solving robotics SLAM problems: |
| 51 | #include <gtsam/sam/RangeFactor.h> |
| 52 | #include <gtsam/slam/BetweenFactor.h> |
| 53 | #include <gtsam/slam/dataset.h> |
| 54 | |
| 55 | // Timing, with functions below, provides nice facilities to benchmark. |
| 56 | #include <gtsam/base/timing.h> |
| 57 | using gtsam::tictoc_print_; |
| 58 | |
| 59 | // Standard headers, added last, so we know headers above work on their own. |
| 60 | #include <fstream> |
| 61 | #include <iostream> |
| 62 | #include <random> |
| 63 | #include <set> |
| 64 | #include <string> |
| 65 | #include <utility> |
| 66 | #include <vector> |
| 67 | |
| 68 | namespace NM = gtsam::noiseModel; |
| 69 | |
| 70 | // Data is second UWB ranging dataset, B2 or "plaza 2", from |
| 71 | // "Navigating with Ranging Radios: Five Data Sets with Ground Truth" |
| 72 | // by Joseph Djugash, Bradley Hamner, and Stephan Roth |
| 73 | // https://www.ri.cmu.edu/pub_files/2009/9/Final_5datasetsRangingRadios.pdf |
| 74 | |
| 75 | // load the odometry |
| 76 | // DR: Odometry Input (delta distance traveled and delta heading change) |
| 77 | // Time (sec) Delta Distance Traveled (m) Delta Heading (rad) |
| 78 | using TimedOdometry = std::pair<double, Pose2>; |
| 79 | std::list<TimedOdometry> readOdometry() { |
| 80 | std::list<TimedOdometry> odometryList; |
| 81 | std::string data_file = gtsam::findExampleDataFile(name: "Plaza2_DR.txt" ); |
| 82 | std::ifstream is(data_file.c_str()); |
| 83 | |
| 84 | while (is) { |
| 85 | double t, distance_traveled, delta_heading; |
| 86 | is >> t >> distance_traveled >> delta_heading; |
| 87 | odometryList.emplace_back(args&: t, args: Pose2(distance_traveled, 0, delta_heading)); |
| 88 | } |
| 89 | is.clear(); /* clears the end-of-file and error flags */ |
| 90 | return odometryList; |
| 91 | } |
| 92 | |
| 93 | // load the ranges from TD |
| 94 | // Time (sec) Sender / Antenna ID Receiver Node ID Range (m) |
| 95 | using RangeTriple = std::tuple<double, size_t, double>; |
| 96 | std::vector<RangeTriple> readTriples() { |
| 97 | std::vector<RangeTriple> triples; |
| 98 | std::string data_file = gtsam::findExampleDataFile(name: "Plaza2_TD.txt" ); |
| 99 | std::ifstream is(data_file.c_str()); |
| 100 | |
| 101 | while (is) { |
| 102 | double t, range, sender, receiver; |
| 103 | is >> t >> sender >> receiver >> range; |
| 104 | triples.emplace_back(args&: t, args&: receiver, args&: range); |
| 105 | } |
| 106 | is.clear(); /* clears the end-of-file and error flags */ |
| 107 | return triples; |
| 108 | } |
| 109 | |
| 110 | // main |
| 111 | int main(int argc, char** argv) { |
| 112 | // load Plaza2 data |
| 113 | std::list<TimedOdometry> odometry = readOdometry(); |
| 114 | size_t M = odometry.size(); |
| 115 | std::cout << "Read " << M << " odometry entries." << std::endl; |
| 116 | |
| 117 | std::vector<RangeTriple> triples = readTriples(); |
| 118 | size_t K = triples.size(); |
| 119 | std::cout << "Read " << K << " range triples." << std::endl; |
| 120 | |
| 121 | // parameters |
| 122 | size_t minK = |
| 123 | 150; // minimum number of range measurements to process initially |
| 124 | size_t incK = 25; // minimum number of range measurements to process after |
| 125 | bool robust = true; |
| 126 | |
| 127 | // Set Noise parameters |
| 128 | Vector priorSigmas = Vector3(1, 1, M_PI); |
| 129 | Vector odoSigmas = Vector3(0.05, 0.01, 0.1); |
| 130 | double sigmaR = 100; // range standard deviation |
| 131 | const NM::Base::shared_ptr // all same type |
| 132 | priorNoise = NM::Diagonal::Sigmas(sigmas: priorSigmas), // prior |
| 133 | looseNoise = NM::Isotropic::Sigma(dim: 2, sigma: 1000), // loose LM prior |
| 134 | odoNoise = NM::Diagonal::Sigmas(sigmas: odoSigmas), // odometry |
| 135 | gaussian = NM::Isotropic::Sigma(dim: 1, sigma: sigmaR), // non-robust |
| 136 | tukey = NM::Robust::Create(robust: NM::mEstimator::Tukey::Create(k: 15), |
| 137 | noise: gaussian), // robust |
| 138 | rangeNoise = robust ? tukey : gaussian; |
| 139 | |
| 140 | // Initialize iSAM |
| 141 | gtsam::ISAM2 isam; |
| 142 | |
| 143 | // Add prior on first pose |
| 144 | Pose2 pose0 = Pose2(-34.2086489999201, 45.3007639991120, M_PI - 2.021089); |
| 145 | gtsam::NonlinearFactorGraph newFactors; |
| 146 | newFactors.addPrior(key: 0, prior: pose0, model: priorNoise); |
| 147 | gtsam::Values initial; |
| 148 | initial.insert(j: 0, val: pose0); |
| 149 | |
| 150 | // We will initialize landmarks randomly, and keep track of which landmarks we |
| 151 | // already added with a set. |
| 152 | std::mt19937_64 rng; |
| 153 | std::normal_distribution<double> normal(0.0, 100.0); |
| 154 | std::set<Symbol> initializedLandmarks; |
| 155 | |
| 156 | // set some loop variables |
| 157 | size_t i = 1; // step counter |
| 158 | size_t k = 0; // range measurement counter |
| 159 | bool initialized = false; |
| 160 | Pose2 lastPose = pose0; |
| 161 | size_t countK = 0; |
| 162 | |
| 163 | // Loop over odometry |
| 164 | gttic_(iSAM); |
| 165 | for (const TimedOdometry& timedOdometry : odometry) { |
| 166 | //--------------------------------- odometry loop -------------------------- |
| 167 | double t; |
| 168 | Pose2 odometry; |
| 169 | std::tie(args&: t, args&: odometry) = timedOdometry; |
| 170 | |
| 171 | // add odometry factor |
| 172 | newFactors.emplace_shared<gtsam::BetweenFactor<Pose2>>(args: i - 1, args&: i, args&: odometry, |
| 173 | args: odoNoise); |
| 174 | |
| 175 | // predict pose and add as initial estimate |
| 176 | Pose2 predictedPose = lastPose.compose(g: odometry); |
| 177 | lastPose = predictedPose; |
| 178 | initial.insert(j: i, val: predictedPose); |
| 179 | |
| 180 | // Check if there are range factors to be added |
| 181 | while (k < K && t >= std::get<0>(t&: triples[k])) { |
| 182 | size_t j = std::get<1>(t&: triples[k]); |
| 183 | Symbol landmark_key('L', j); |
| 184 | double range = std::get<2>(t&: triples[k]); |
| 185 | newFactors.emplace_shared<gtsam::RangeFactor<Pose2, Point2>>( |
| 186 | args&: i, args&: landmark_key, args&: range, args: rangeNoise); |
| 187 | if (initializedLandmarks.count(x: landmark_key) == 0) { |
| 188 | std::cout << "adding landmark " << j << std::endl; |
| 189 | double x = normal(rng), y = normal(rng); |
| 190 | initial.insert(j: landmark_key, val: Point2(x, y)); |
| 191 | initializedLandmarks.insert(x: landmark_key); |
| 192 | // We also add a very loose prior on the landmark in case there is only |
| 193 | // one sighting, which cannot fully determine the landmark. |
| 194 | newFactors.emplace_shared<gtsam::PriorFactor<Point2>>( |
| 195 | args&: landmark_key, args: Point2(0, 0), args: looseNoise); |
| 196 | } |
| 197 | k = k + 1; |
| 198 | countK = countK + 1; |
| 199 | } |
| 200 | |
| 201 | // Check whether to update iSAM 2 |
| 202 | if ((k > minK) && (countK > incK)) { |
| 203 | if (!initialized) { // Do a full optimize for first minK ranges |
| 204 | std::cout << "Initializing at time " << k << std::endl; |
| 205 | gttic_(batchInitialization); |
| 206 | gtsam::LevenbergMarquardtOptimizer batchOptimizer(newFactors, initial); |
| 207 | initial = batchOptimizer.optimize(); |
| 208 | gttoc_(batchInitialization); |
| 209 | initialized = true; |
| 210 | } |
| 211 | gttic_(update); |
| 212 | isam.update(newFactors, newTheta: initial); |
| 213 | gttoc_(update); |
| 214 | gttic_(calculateEstimate); |
| 215 | gtsam::Values result = isam.calculateEstimate(); |
| 216 | gttoc_(calculateEstimate); |
| 217 | lastPose = result.at<Pose2>(j: i); |
| 218 | newFactors = gtsam::NonlinearFactorGraph(); |
| 219 | initial = gtsam::Values(); |
| 220 | countK = 0; |
| 221 | } |
| 222 | i += 1; |
| 223 | //--------------------------------- odometry loop -------------------------- |
| 224 | } // end for |
| 225 | gttoc_(iSAM); |
| 226 | |
| 227 | // Print timings |
| 228 | tictoc_print_(); |
| 229 | |
| 230 | // Print optimized landmarks: |
| 231 | gtsam::Values finalResult = isam.calculateEstimate(); |
| 232 | for (auto&& landmark_key : initializedLandmarks) { |
| 233 | Point2 p = finalResult.at<Point2>(j: landmark_key); |
| 234 | std::cout << landmark_key << ":" << p.transpose() << "\n" ; |
| 235 | } |
| 236 | |
| 237 | exit(status: 0); |
| 238 | } |
| 239 | |