1 // Ceres Solver - A fast non-linear least squares minimizer 2 // Copyright 2014 Google Inc. All rights reserved. 3 // http://code.google.com/p/ceres-solver/ 4 // 5 // Redistribution and use in source and binary forms, with or without 6 // modification, are permitted provided that the following conditions are met: 7 // 8 // * Redistributions of source code must retain the above copyright notice, 9 // this list of conditions and the following disclaimer. 10 // * Redistributions in binary form must reproduce the above copyright notice, 11 // this list of conditions and the following disclaimer in the documentation 12 // and/or other materials provided with the distribution. 13 // * Neither the name of Google Inc. nor the names of its contributors may be 14 // used to endorse or promote products derived from this software without 15 // specific prior written permission. 16 // 17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 27 // POSSIBILITY OF SUCH DAMAGE. 28 // 29 // Author: joydeepb (at) ri.cmu.edu (Joydeep Biswas) 30 // 31 // This example demonstrates how to use the DynamicAutoDiffCostFunction 32 // variant of CostFunction. The DynamicAutoDiffCostFunction is meant to 33 // be used in cases where the number of parameter blocks or the sizes are not 34 // known at compile time. 35 // 36 // This example simulates a robot traversing down a 1-dimension hallway with 37 // noise odometry readings and noisy range readings of the end of the hallway. 38 // By fusing the noisy odometry and sensor readings this example demonstrates 39 // how to compute the maximum likelihood estimate (MLE) of the robot's pose at 40 // each timestep. 41 // 42 // The robot starts at the origin, and it is travels to the end of a corridor of 43 // fixed length specified by the "--corridor_length" flag. It executes a series 44 // of motion commands to move forward a fixed length, specified by the 45 // "--pose_separation" flag, at which pose it receives relative odometry 46 // measurements as well as a range reading of the distance to the end of the 47 // hallway. The odometry readings are drawn with Gaussian noise and standard 48 // deviation specified by the "--odometry_stddev" flag, and the range readings 49 // similarly with standard deviation specified by the "--range-stddev" flag. 50 // 51 // There are two types of residuals in this problem: 52 // 1) The OdometryConstraint residual, that accounts for the odometry readings 53 // between successive pose estimatess of the robot. 54 // 2) The RangeConstraint residual, that accounts for the errors in the observed 55 // range readings from each pose. 56 // 57 // The OdometryConstraint residual is modeled as an AutoDiffCostFunction with 58 // a fixed parameter block size of 1, which is the relative odometry being 59 // solved for, between a pair of successive poses of the robot. Differences 60 // between observed and computed relative odometry values are penalized weighted 61 // by the known standard deviation of the odometry readings. 62 // 63 // The RangeConstraint residual is modeled as a DynamicAutoDiffCostFunction 64 // which sums up the relative odometry estimates to compute the estimated 65 // global pose of the robot, and then computes the expected range reading. 66 // Differences between the observed and expected range readings are then 67 // penalized weighted by the standard deviation of readings of the sensor. 68 // Since the number of poses of the robot is not known at compile time, this 69 // cost function is implemented as a DynamicAutoDiffCostFunction. 70 // 71 // The outputs of the example are the initial values of the odometry and range 72 // readings, and the range and odometry errors for every pose of the robot. 73 // After computing the MLE, the computed poses and corrected odometry values 74 // are printed out, along with the corresponding range and odometry errors. Note 75 // that as an MLE of a noisy system the errors will not be reduced to zero, but 76 // the odometry estimates will be updated to maximize the joint likelihood of 77 // all odometry and range readings of the robot. 78 // 79 // Mathematical Formulation 80 // ====================================================== 81 // 82 // Let p_0, .., p_N be (N+1) robot poses, where the robot moves down the 83 // corridor starting from p_0 and ending at p_N. We assume that p_0 is the 84 // origin of the coordinate system. 85 // Odometry u_i is the observed relative odometry between pose p_(i-1) and p_i, 86 // and range reading y_i is the range reading of the end of the corridor from 87 // pose p_i. Both odometry as well as range readings are noisy, but we wish to 88 // compute the maximum likelihood estimate (MLE) of corrected odometry values 89 // u*_0 to u*_(N-1), such that the Belief is optimized: 90 // 91 // Belief(u*_(0:N-1) | u_(0:N-1), y_(0:N-1)) 1. 92 // = P(u*_(0:N-1) | u_(0:N-1), y_(0:N-1)) 2. 93 // \propto P(y_(0:N-1) | u*_(0:N-1), u_(0:N-1)) P(u*_(0:N-1) | u_(0:N-1)) 3. 94 // = \prod_i{ P(y_i | u*_(0:i)) P(u*_i | u_i) } 4. 95 // 96 // Here, the subscript "(0:i)" is used as shorthand to indicate entries from all 97 // timesteps 0 to i for that variable, both inclusive. 98 // 99 // Bayes' rule is used to derive eq. 3 from 2, and the independence of 100 // odometry observations and range readings is expolited to derive 4 from 3. 101 // 102 // Thus, the Belief, up to scale, is factored as a product of a number of 103 // terms, two for each pose, where for each pose term there is one term for the 104 // range reading, P(y_i | u*_(0:i) and one term for the odometry reading, 105 // P(u*_i | u_i) . Note that the term for the range reading is dependent on all 106 // odometry values u*_(0:i), while the odometry term, P(u*_i | u_i) depends only 107 // on a single value, u_i. Both the range reading as well as odoemtry 108 // probability terms are modeled as the Normal distribution, and have the form: 109 // 110 // p(x) \propto \exp{-((x - x_mean) / x_stddev)^2} 111 // 112 // where x refers to either the MLE odometry u* or range reading y, and x_mean 113 // is the corresponding mean value, u for the odometry terms, and y_expected, 114 // the expected range reading based on all the previous odometry terms. 115 // The MLE is thus found by finding those values x* which minimize: 116 // 117 // x* = \arg\min{((x - x_mean) / x_stddev)^2} 118 // 119 // which is in the nonlinear least-square form, suited to being solved by Ceres. 120 // The non-linear component arise from the computation of x_mean. The residuals 121 // ((x - x_mean) / x_stddev) for the residuals that Ceres will optimize. As 122 // mentioned earlier, the odometry term for each pose depends only on one 123 // variable, and will be computed by an AutoDiffCostFunction, while the term 124 // for the range reading will depend on all previous odometry observations, and 125 // will be computed by a DynamicAutoDiffCostFunction since the number of 126 // odoemtry observations will only be known at run time. 127 128 #include <cstdio> 129 #include <math.h> 130 #include <vector> 131 132 #include "ceres/ceres.h" 133 #include "ceres/dynamic_autodiff_cost_function.h" 134 #include "gflags/gflags.h" 135 #include "glog/logging.h" 136 #include "random.h" 137 138 using ceres::AutoDiffCostFunction; 139 using ceres::DynamicAutoDiffCostFunction; 140 using ceres::CauchyLoss; 141 using ceres::CostFunction; 142 using ceres::LossFunction; 143 using ceres::Problem; 144 using ceres::Solve; 145 using ceres::Solver; 146 using ceres::examples::RandNormal; 147 using std::min; 148 using std::vector; 149 150 DEFINE_double(corridor_length, 30.0, "Length of the corridor that the robot is " 151 "travelling down."); 152 153 DEFINE_double(pose_separation, 0.5, "The distance that the robot traverses " 154 "between successive odometry updates."); 155 156 DEFINE_double(odometry_stddev, 0.1, "The standard deviation of " 157 "odometry error of the robot."); 158 159 DEFINE_double(range_stddev, 0.01, "The standard deviation of range readings of " 160 "the robot."); 161 162 // The stride length of the dynamic_autodiff_cost_function evaluator. 163 static const int kStride = 10; 164 165 struct OdometryConstraint { 166 typedef AutoDiffCostFunction<OdometryConstraint, 1, 1> OdometryCostFunction; 167 168 OdometryConstraint(double odometry_mean, double odometry_stddev) : 169 odometry_mean(odometry_mean), odometry_stddev(odometry_stddev) {} 170 171 template <typename T> 172 bool operator()(const T* const odometry, T* residual) const { 173 *residual = (*odometry - T(odometry_mean)) / T(odometry_stddev); 174 return true; 175 } 176 177 static OdometryCostFunction* Create(const double odometry_value) { 178 return new OdometryCostFunction( 179 new OdometryConstraint(odometry_value, FLAGS_odometry_stddev)); 180 } 181 182 const double odometry_mean; 183 const double odometry_stddev; 184 }; 185 186 struct RangeConstraint { 187 typedef DynamicAutoDiffCostFunction<RangeConstraint, kStride> 188 RangeCostFunction; 189 190 RangeConstraint( 191 int pose_index, 192 double range_reading, 193 double range_stddev, 194 double corridor_length) : 195 pose_index(pose_index), range_reading(range_reading), 196 range_stddev(range_stddev), corridor_length(corridor_length) {} 197 198 template <typename T> 199 bool operator()(T const* const* relative_poses, T* residuals) const { 200 T global_pose(0); 201 for (int i = 0; i <= pose_index; ++i) { 202 global_pose += relative_poses[i][0]; 203 } 204 residuals[0] = (global_pose + T(range_reading) - T(corridor_length)) / 205 T(range_stddev); 206 return true; 207 } 208 209 // Factory method to create a CostFunction from a RangeConstraint to 210 // conveniently add to a ceres problem. 211 static RangeCostFunction* Create(const int pose_index, 212 const double range_reading, 213 vector<double>* odometry_values, 214 vector<double*>* parameter_blocks) { 215 RangeConstraint* constraint = new RangeConstraint( 216 pose_index, range_reading, FLAGS_range_stddev, FLAGS_corridor_length); 217 RangeCostFunction* cost_function = new RangeCostFunction(constraint); 218 // Add all the parameter blocks that affect this constraint. 219 parameter_blocks->clear(); 220 for (int i = 0; i <= pose_index; ++i) { 221 parameter_blocks->push_back(&((*odometry_values)[i])); 222 cost_function->AddParameterBlock(1); 223 } 224 cost_function->SetNumResiduals(1); 225 return (cost_function); 226 } 227 228 const int pose_index; 229 const double range_reading; 230 const double range_stddev; 231 const double corridor_length; 232 }; 233 234 void SimulateRobot(vector<double>* odometry_values, 235 vector<double>* range_readings) { 236 const int num_steps = static_cast<int>( 237 ceil(FLAGS_corridor_length / FLAGS_pose_separation)); 238 239 // The robot starts out at the origin. 240 double robot_location = 0.0; 241 for (int i = 0; i < num_steps; ++i) { 242 const double actual_odometry_value = min( 243 FLAGS_pose_separation, FLAGS_corridor_length - robot_location); 244 robot_location += actual_odometry_value; 245 const double actual_range = FLAGS_corridor_length - robot_location; 246 const double observed_odometry = 247 RandNormal() * FLAGS_odometry_stddev + actual_odometry_value; 248 const double observed_range = 249 RandNormal() * FLAGS_range_stddev + actual_range; 250 odometry_values->push_back(observed_odometry); 251 range_readings->push_back(observed_range); 252 } 253 } 254 255 void PrintState(const vector<double>& odometry_readings, 256 const vector<double>& range_readings) { 257 CHECK_EQ(odometry_readings.size(), range_readings.size()); 258 double robot_location = 0.0; 259 printf("pose: location odom range r.error o.error\n"); 260 for (int i = 0; i < odometry_readings.size(); ++i) { 261 robot_location += odometry_readings[i]; 262 const double range_error = 263 robot_location + range_readings[i] - FLAGS_corridor_length; 264 const double odometry_error = 265 FLAGS_pose_separation - odometry_readings[i]; 266 printf("%4d: %8.3f %8.3f %8.3f %8.3f %8.3f\n", 267 static_cast<int>(i), robot_location, odometry_readings[i], 268 range_readings[i], range_error, odometry_error); 269 } 270 } 271 272 int main(int argc, char** argv) { 273 google::InitGoogleLogging(argv[0]); 274 google::ParseCommandLineFlags(&argc, &argv, true); 275 // Make sure that the arguments parsed are all positive. 276 CHECK_GT(FLAGS_corridor_length, 0.0); 277 CHECK_GT(FLAGS_pose_separation, 0.0); 278 CHECK_GT(FLAGS_odometry_stddev, 0.0); 279 CHECK_GT(FLAGS_range_stddev, 0.0); 280 281 vector<double> odometry_values; 282 vector<double> range_readings; 283 SimulateRobot(&odometry_values, &range_readings); 284 285 printf("Initial values:\n"); 286 PrintState(odometry_values, range_readings); 287 ceres::Problem problem; 288 289 for (int i = 0; i < odometry_values.size(); ++i) { 290 // Create and add a DynamicAutoDiffCostFunction for the RangeConstraint from 291 // pose i. 292 vector<double*> parameter_blocks; 293 RangeConstraint::RangeCostFunction* range_cost_function = 294 RangeConstraint::Create( 295 i, range_readings[i], &odometry_values, ¶meter_blocks); 296 problem.AddResidualBlock(range_cost_function, NULL, parameter_blocks); 297 298 // Create and add an AutoDiffCostFunction for the OdometryConstraint for 299 // pose i. 300 problem.AddResidualBlock(OdometryConstraint::Create(odometry_values[i]), 301 NULL, 302 &(odometry_values[i])); 303 } 304 305 ceres::Solver::Options solver_options; 306 solver_options.minimizer_progress_to_stdout = true; 307 308 Solver::Summary summary; 309 printf("Solving...\n"); 310 Solve(solver_options, &problem, &summary); 311 printf("Done.\n"); 312 std::cout << summary.FullReport() << "\n"; 313 printf("Final values:\n"); 314 PrintState(odometry_values, range_readings); 315 return 0; 316 } 317