Merge branch 'svn/trunk'
Conflicts: gtsam_unstable/slam/BetweenFactorEM.h gtsam_unstable/slam/tests/testBetweenFactorEM.cpprelease/4.3a0
Richard Roberts
commit
d9c9682f6e
|
|
@ -73,7 +73,7 @@ Values BatchOptimize(const NonlinearFactorGraph& graph, const Values& theta, int
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, equals )
|
TEST( ConcurrentIncrementalSmootherDL, equals )
|
||||||
{
|
{
|
||||||
// TODO: Test 'equals' more vigorously
|
// TODO: Test 'equals' more vigorously
|
||||||
|
|
||||||
|
|
@ -99,7 +99,7 @@ TEST( ConcurrentIncrementalSmoother, equals )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getFactors )
|
TEST( ConcurrentIncrementalSmootherDL, getFactors )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -150,7 +150,7 @@ TEST( ConcurrentIncrementalSmoother, getFactors )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getLinearizationPoint )
|
TEST( ConcurrentIncrementalSmootherDL, getLinearizationPoint )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -201,19 +201,19 @@ TEST( ConcurrentIncrementalSmoother, getLinearizationPoint )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getOrdering )
|
TEST( ConcurrentIncrementalSmootherDL, getOrdering )
|
||||||
{
|
{
|
||||||
// TODO: Think about how to check ordering...
|
// TODO: Think about how to check ordering...
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getDelta )
|
TEST( ConcurrentIncrementalSmootherDL, getDelta )
|
||||||
{
|
{
|
||||||
// TODO: Think about how to check ordering...
|
// TODO: Think about how to check ordering...
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, calculateEstimate )
|
TEST( ConcurrentIncrementalSmootherDL, calculateEstimate )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -287,7 +287,7 @@ TEST( ConcurrentIncrementalSmoother, calculateEstimate )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, update_empty )
|
TEST( ConcurrentIncrementalSmootherDL, update_empty )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -300,7 +300,7 @@ TEST( ConcurrentIncrementalSmoother, update_empty )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, update_multiple )
|
TEST( ConcurrentIncrementalSmootherDL, update_multiple )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -358,7 +358,7 @@ TEST( ConcurrentIncrementalSmoother, update_multiple )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_empty )
|
TEST( ConcurrentIncrementalSmootherDL, synchronize_empty )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -388,7 +388,7 @@ TEST( ConcurrentIncrementalSmoother, synchronize_empty )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_1 )
|
TEST( ConcurrentIncrementalSmootherDL, synchronize_1 )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -450,7 +450,7 @@ TEST( ConcurrentIncrementalSmoother, synchronize_1 )
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_2 )
|
TEST( ConcurrentIncrementalSmootherDL, synchronize_2 )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -531,7 +531,7 @@ TEST( ConcurrentIncrementalSmoother, synchronize_2 )
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_3 )
|
TEST( ConcurrentIncrementalSmootherDL, synchronize_3 )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
|
||||||
|
|
@ -73,7 +73,7 @@ Values BatchOptimize(const NonlinearFactorGraph& graph, const Values& theta, int
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, equals )
|
TEST( ConcurrentIncrementalSmootherGN, equals )
|
||||||
{
|
{
|
||||||
// TODO: Test 'equals' more vigorously
|
// TODO: Test 'equals' more vigorously
|
||||||
|
|
||||||
|
|
@ -99,7 +99,7 @@ TEST( ConcurrentIncrementalSmoother, equals )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getFactors )
|
TEST( ConcurrentIncrementalSmootherGN, getFactors )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -150,7 +150,7 @@ TEST( ConcurrentIncrementalSmoother, getFactors )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getLinearizationPoint )
|
TEST( ConcurrentIncrementalSmootherGN, getLinearizationPoint )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -201,19 +201,19 @@ TEST( ConcurrentIncrementalSmoother, getLinearizationPoint )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getOrdering )
|
TEST( ConcurrentIncrementalSmootherGN, getOrdering )
|
||||||
{
|
{
|
||||||
// TODO: Think about how to check ordering...
|
// TODO: Think about how to check ordering...
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, getDelta )
|
TEST( ConcurrentIncrementalSmootherGN, getDelta )
|
||||||
{
|
{
|
||||||
// TODO: Think about how to check ordering...
|
// TODO: Think about how to check ordering...
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, calculateEstimate )
|
TEST( ConcurrentIncrementalSmootherGN, calculateEstimate )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -287,7 +287,7 @@ TEST( ConcurrentIncrementalSmoother, calculateEstimate )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, update_empty )
|
TEST( ConcurrentIncrementalSmootherGN, update_empty )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -300,7 +300,7 @@ TEST( ConcurrentIncrementalSmoother, update_empty )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, update_multiple )
|
TEST( ConcurrentIncrementalSmootherGN, update_multiple )
|
||||||
{
|
{
|
||||||
// Create a Concurrent Batch Smoother
|
// Create a Concurrent Batch Smoother
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -358,7 +358,7 @@ TEST( ConcurrentIncrementalSmoother, update_multiple )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_empty )
|
TEST( ConcurrentIncrementalSmootherGN, synchronize_empty )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -388,7 +388,7 @@ TEST( ConcurrentIncrementalSmoother, synchronize_empty )
|
||||||
}
|
}
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_1 )
|
TEST( ConcurrentIncrementalSmootherGN, synchronize_1 )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -450,7 +450,7 @@ TEST( ConcurrentIncrementalSmoother, synchronize_1 )
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_2 )
|
TEST( ConcurrentIncrementalSmootherGN, synchronize_2 )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
@ -531,7 +531,7 @@ TEST( ConcurrentIncrementalSmoother, synchronize_2 )
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( ConcurrentIncrementalSmoother, synchronize_3 )
|
TEST( ConcurrentIncrementalSmootherGN, synchronize_3 )
|
||||||
{
|
{
|
||||||
// Create a set of optimizer parameters
|
// Create a set of optimizer parameters
|
||||||
ISAM2Params parameters;
|
ISAM2Params parameters;
|
||||||
|
|
|
||||||
|
|
@ -232,8 +232,8 @@ namespace gtsam {
|
||||||
Matrix invCov_inlier = model_inlier_->R().transpose() * model_inlier_->R();
|
Matrix invCov_inlier = model_inlier_->R().transpose() * model_inlier_->R();
|
||||||
Matrix invCov_outlier = model_outlier_->R().transpose() * model_outlier_->R();
|
Matrix invCov_outlier = model_outlier_->R().transpose() * model_outlier_->R();
|
||||||
|
|
||||||
double p_inlier = prior_inlier_ * invCov_inlier.norm() * exp( -0.5 * err_wh_inlier.dot(err_wh_inlier) );
|
double p_inlier = prior_inlier_ * std::sqrt(invCov_inlier.norm()) * exp( -0.5 * err_wh_inlier.dot(err_wh_inlier) );
|
||||||
double p_outlier = prior_outlier_ * invCov_outlier.norm() * exp( -0.5 * err_wh_outlier.dot(err_wh_outlier) );
|
double p_outlier = prior_outlier_ * std::sqrt(invCov_outlier.norm()) * exp( -0.5 * err_wh_outlier.dot(err_wh_outlier) );
|
||||||
|
|
||||||
double sumP = p_inlier + p_outlier;
|
double sumP = p_inlier + p_outlier;
|
||||||
p_inlier /= sumP;
|
p_inlier /= sumP;
|
||||||
|
|
|
||||||
|
|
@ -173,8 +173,8 @@ TEST( SmartProjectionFactor, noisy ){
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( SmartProjectionFactor, 3poses ){
|
TEST( SmartProjectionFactor, 3poses_smart_projection_factor ){
|
||||||
cout << " ************************ MultiProjectionFactor: 3 cams + 3 landmarks **********************" << endl;
|
cout << " ************************ SmartProjectionFactor: 3 cams + 3 landmarks **********************" << endl;
|
||||||
|
|
||||||
Symbol x1('X', 1);
|
Symbol x1('X', 1);
|
||||||
Symbol x2('X', 2);
|
Symbol x2('X', 2);
|
||||||
|
|
@ -239,17 +239,19 @@ TEST( SmartProjectionFactor, 3poses ){
|
||||||
graph.push_back(PriorFactor<Pose3>(x1, pose1, noisePrior));
|
graph.push_back(PriorFactor<Pose3>(x1, pose1, noisePrior));
|
||||||
graph.push_back(PriorFactor<Pose3>(x2, pose2, noisePrior));
|
graph.push_back(PriorFactor<Pose3>(x2, pose2, noisePrior));
|
||||||
|
|
||||||
Pose3 noise_pose = Pose3(Rot3::ypr(-M_PI/100, 0., -M_PI/100), gtsam::Point3(0.1,0.1,0.1));
|
// Pose3 noise_pose = Pose3(Rot3::ypr(-M_PI/10, 0., -M_PI/10), gtsam::Point3(0.5,0.1,0.3)); // noise from regular projection factor test below
|
||||||
|
Pose3 noise_pose = Pose3(Rot3::ypr(-M_PI/100, 0., -M_PI/100), gtsam::Point3(0.1,0.1,0.1)); // smaller noise
|
||||||
Values values;
|
Values values;
|
||||||
values.insert(x1, pose1);
|
values.insert(x1, pose1);
|
||||||
values.insert(x2, pose2*noise_pose);
|
values.insert(x2, pose2);
|
||||||
values.insert(x3, pose3);
|
// initialize third pose with some noise, we expect it to move back to original pose3
|
||||||
|
values.insert(x3, pose3*noise_pose);
|
||||||
|
values.at<Pose3>(x3).print("Smart: Pose3 before optimization: ");
|
||||||
|
|
||||||
LevenbergMarquardtParams params;
|
LevenbergMarquardtParams params;
|
||||||
params.verbosityLM = LevenbergMarquardtParams::TRYLAMBDA;
|
params.verbosityLM = LevenbergMarquardtParams::TRYLAMBDA;
|
||||||
params.verbosity = NonlinearOptimizerParams::ERROR;
|
params.verbosity = NonlinearOptimizerParams::ERROR;
|
||||||
|
|
||||||
|
|
||||||
Values result;
|
Values result;
|
||||||
gttic_(SmartProjectionFactor);
|
gttic_(SmartProjectionFactor);
|
||||||
LevenbergMarquardtOptimizer optimizer(graph, values, params);
|
LevenbergMarquardtOptimizer optimizer(graph, values, params);
|
||||||
|
|
@ -257,7 +259,9 @@ TEST( SmartProjectionFactor, 3poses ){
|
||||||
gttoc_(SmartProjectionFactor);
|
gttoc_(SmartProjectionFactor);
|
||||||
tictoc_finishedIteration_();
|
tictoc_finishedIteration_();
|
||||||
|
|
||||||
result.print("results of 3 camera, 3 landmark optimization \n");
|
// result.print("results of 3 camera, 3 landmark optimization \n");
|
||||||
|
result.at<Pose3>(x3).print("Smart: Pose3 after optimization: ");
|
||||||
|
EXPECT(assert_equal(pose3,result.at<Pose3>(x3)));
|
||||||
tictoc_print_();
|
tictoc_print_();
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
@ -265,7 +269,7 @@ TEST( SmartProjectionFactor, 3poses ){
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( SmartProjectionFactor, 3poses_projection_factor ){
|
TEST( SmartProjectionFactor, 3poses_projection_factor ){
|
||||||
cout << " ************************ Normal ProjectionFactor: 3 cams + 3 landmarks **********************" << endl;
|
// cout << " ************************ Normal ProjectionFactor: 3 cams + 3 landmarks **********************" << endl;
|
||||||
|
|
||||||
Symbol x1('X', 1);
|
Symbol x1('X', 1);
|
||||||
Symbol x2('X', 2);
|
Symbol x2('X', 2);
|
||||||
|
|
@ -287,7 +291,6 @@ TEST( SmartProjectionFactor, 3poses_projection_factor ){
|
||||||
|
|
||||||
// create third camera 1 meter above the first camera
|
// create third camera 1 meter above the first camera
|
||||||
Pose3 pose3 = pose1 * Pose3(Rot3(), Point3(0,-1,0));
|
Pose3 pose3 = pose1 * Pose3(Rot3(), Point3(0,-1,0));
|
||||||
pose3.print("Pose3: ");
|
|
||||||
SimpleCamera cam3(pose3, *K);
|
SimpleCamera cam3(pose3, *K);
|
||||||
|
|
||||||
// three landmarks ~5 meters infront of camera
|
// three landmarks ~5 meters infront of camera
|
||||||
|
|
@ -324,6 +327,7 @@ TEST( SmartProjectionFactor, 3poses_projection_factor ){
|
||||||
values.insert(L(1), landmark1);
|
values.insert(L(1), landmark1);
|
||||||
values.insert(L(2), landmark2);
|
values.insert(L(2), landmark2);
|
||||||
values.insert(L(3), landmark3);
|
values.insert(L(3), landmark3);
|
||||||
|
// values.at<Pose3>(x3).print("Pose3 before optimization: ");
|
||||||
|
|
||||||
LevenbergMarquardtParams params;
|
LevenbergMarquardtParams params;
|
||||||
// params.verbosityLM = LevenbergMarquardtParams::TRYLAMBDA;
|
// params.verbosityLM = LevenbergMarquardtParams::TRYLAMBDA;
|
||||||
|
|
@ -331,14 +335,15 @@ TEST( SmartProjectionFactor, 3poses_projection_factor ){
|
||||||
LevenbergMarquardtOptimizer optimizer(graph, values, params);
|
LevenbergMarquardtOptimizer optimizer(graph, values, params);
|
||||||
Values result = optimizer.optimize();
|
Values result = optimizer.optimize();
|
||||||
|
|
||||||
result.print("Regular Projection Factor: results of 3 camera, 3 landmark optimization \n");
|
// result.at<Pose3>(x3).print("Pose3 after optimization: ");
|
||||||
|
EXPECT(assert_equal(pose3,result.at<Pose3>(x3)));
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
TEST( SmartProjectionFactor, Hessian ){
|
TEST( SmartProjectionFactor, Hessian ){
|
||||||
cout << " ************************ Normal ProjectionFactor: Hessian **********************" << endl;
|
cout << " ************************ SmartProjectionFactor: Hessian **********************" << endl;
|
||||||
|
|
||||||
Symbol x1('X', 1);
|
Symbol x1('X', 1);
|
||||||
Symbol x2('X', 2);
|
Symbol x2('X', 2);
|
||||||
|
|
|
||||||
|
|
@ -1,226 +0,0 @@
|
||||||
%close all
|
|
||||||
%clc
|
|
||||||
|
|
||||||
import gtsam.*;
|
|
||||||
|
|
||||||
%% Read metadata and compute relative sensor pose transforms
|
|
||||||
IMU_metadata = importdata('KittiEquivBiasedImu_metadata.txt');
|
|
||||||
IMU_metadata = cell2struct(num2cell(IMU_metadata.data), IMU_metadata.colheaders, 2);
|
|
||||||
IMUinBody = Pose3.Expmap([
|
|
||||||
IMU_metadata.BodyPtx; IMU_metadata.BodyPty; IMU_metadata.BodyPtz;
|
|
||||||
IMU_metadata.BodyPrx; IMU_metadata.BodyPry; IMU_metadata.BodyPrz; ]);
|
|
||||||
if ~IMUinBody.equals(Pose3, 1e-5)
|
|
||||||
error 'Currently only support IMUinBody is identity, i.e. IMU and body frame are the same';
|
|
||||||
end
|
|
||||||
|
|
||||||
VO_metadata = importdata('KittiRelativePose_metadata.txt');
|
|
||||||
VO_metadata = cell2struct(num2cell(VO_metadata.data), VO_metadata.colheaders, 2);
|
|
||||||
VOinBody = Pose3.Expmap([
|
|
||||||
VO_metadata.BodyPtx; VO_metadata.BodyPty; VO_metadata.BodyPtz;
|
|
||||||
VO_metadata.BodyPrx; VO_metadata.BodyPry; VO_metadata.BodyPrz; ]);
|
|
||||||
|
|
||||||
GPS_metadata = importdata('KittiGps_metadata.txt');
|
|
||||||
GPS_metadata = cell2struct(num2cell(GPS_metadata.data), GPS_metadata.colheaders, 2);
|
|
||||||
GPSinBody = Pose3.Expmap([
|
|
||||||
GPS_metadata.BodyPtx; GPS_metadata.BodyPty; GPS_metadata.BodyPtz;
|
|
||||||
GPS_metadata.BodyPrx; GPS_metadata.BodyPry; GPS_metadata.BodyPrz; ]);
|
|
||||||
|
|
||||||
VOinIMU = IMUinBody.inverse().compose(VOinBody);
|
|
||||||
GPSinIMU = IMUinBody.inverse().compose(GPSinBody);
|
|
||||||
|
|
||||||
%% Read data and change coordinate frame of GPS and VO measurements to IMU frame
|
|
||||||
IMU_data = importdata('KittiEquivBiasedImu.txt');
|
|
||||||
IMU_data = cell2struct(num2cell(IMU_data.data), IMU_data.colheaders, 2);
|
|
||||||
imum = cellfun(@(x) x', num2cell([ [IMU_data.accelX]' [IMU_data.accelY]' [IMU_data.accelZ]' [IMU_data.omegaX]' [IMU_data.omegaY]' [IMU_data.omegaZ]' ], 2), 'UniformOutput', false);
|
|
||||||
[IMU_data.acc_omega] = deal(imum{:});
|
|
||||||
IMU_data = rmfield(IMU_data, { 'accelX' 'accelY' 'accelZ' 'omegaX' 'omegaY' 'omegaZ' });
|
|
||||||
clear imum
|
|
||||||
|
|
||||||
VO_data = importdata('KittiRelativePose.txt');
|
|
||||||
VO_data = cell2struct(num2cell(VO_data.data), VO_data.colheaders, 2);
|
|
||||||
% Merge relative pose fields and convert to Pose3
|
|
||||||
logposes = [ [VO_data.dtx]' [VO_data.dty]' [VO_data.dtz]' [VO_data.drx]' [VO_data.dry]' [VO_data.drz]' ];
|
|
||||||
logposes = num2cell(logposes, 2);
|
|
||||||
relposes = arrayfun(@(x) {gtsam.Pose3.Expmap(x{:}')}, logposes);
|
|
||||||
relposes = arrayfun(@(x) {VOinIMU.compose(x{:}).compose(VOinIMU.inverse())}, relposes);
|
|
||||||
[VO_data.RelativePose] = deal(relposes{:});
|
|
||||||
VO_data = rmfield(VO_data, { 'dtx' 'dty' 'dtz' 'drx' 'dry' 'drz' });
|
|
||||||
clear logposes relposes
|
|
||||||
|
|
||||||
GPS_data = importdata('KittiGps.txt');
|
|
||||||
GPS_data = cell2struct(num2cell(GPS_data.data), GPS_data.colheaders, 2);
|
|
||||||
|
|
||||||
%% Set initial conditions for the estimated trajectory
|
|
||||||
disp('TODO: we have GPS so this initialization is not right')
|
|
||||||
currentPoseGlobal = Pose3; % initial pose is the reference frame (navigation frame)
|
|
||||||
currentVelocityGlobal = [0;0;0]; % the vehicle is stationary at the beginning
|
|
||||||
bias_acc = [0;0;0]; % we initialize accelerometer biases to zero
|
|
||||||
bias_omega = [0;0;0]; % we initialize gyro biases to zero
|
|
||||||
|
|
||||||
%% Solver object
|
|
||||||
isamParams = ISAM2Params;
|
|
||||||
isamParams.setRelinearizeSkip(1);
|
|
||||||
isam = gtsam.ISAM2(isamParams);
|
|
||||||
|
|
||||||
%% create nonlinear factor graph
|
|
||||||
factors = NonlinearFactorGraph;
|
|
||||||
values = Values;
|
|
||||||
|
|
||||||
%% Create prior on initial pose, velocity, and biases
|
|
||||||
sigma_init_x = 1.0;
|
|
||||||
sigma_init_v = 1.0;
|
|
||||||
sigma_init_b = 1.0;
|
|
||||||
|
|
||||||
values.insert(symbol('x',0), currentPoseGlobal);
|
|
||||||
values.insert(symbol('v',0), LieVector(currentVelocityGlobal) );
|
|
||||||
values.insert(symbol('b',0), imuBias.ConstantBias(bias_acc,bias_omega) );
|
|
||||||
|
|
||||||
disp('TODO: we have GPS so this initialization is not right')
|
|
||||||
% Prior on initial pose
|
|
||||||
factors.add(PriorFactorPose3(symbol('x',0), ...
|
|
||||||
currentPoseGlobal, noiseModel.Isotropic.Sigma(6, sigma_init_x)));
|
|
||||||
% Prior on initial velocity
|
|
||||||
factors.add(PriorFactorLieVector(symbol('v',0), ...
|
|
||||||
LieVector(currentVelocityGlobal), noiseModel.Isotropic.Sigma(3, sigma_init_v)));
|
|
||||||
% Prior on initial bias
|
|
||||||
factors.add(PriorFactorConstantBias(symbol('b',0), ...
|
|
||||||
imuBias.ConstantBias(bias_acc,bias_omega), noiseModel.Isotropic.Sigma(6, sigma_init_b)));
|
|
||||||
|
|
||||||
%% Main loop:
|
|
||||||
% (1) we read the measurements
|
|
||||||
% (2) we create the corresponding factors in the graph
|
|
||||||
% (3) we solve the graph to obtain and optimal estimate of robot trajectory
|
|
||||||
|
|
||||||
% lastTime = 0; TODO: delete?
|
|
||||||
% lastIndex = 0; TODO: delete?
|
|
||||||
currentSummarizedMeasurement = [];
|
|
||||||
|
|
||||||
% Measurement types:
|
|
||||||
% 1: VO
|
|
||||||
% 2: GPS
|
|
||||||
% 3: IMU
|
|
||||||
times = sortrows( [ ...
|
|
||||||
[VO_data.Time]' 1*ones(length([VO_data.Time]), 1); ...
|
|
||||||
%[GPS_data.Time]' 2*ones(length([GPS_data.Time]), 1); ...
|
|
||||||
[IMU_data.Time]' 3*ones(length([IMU_data.Time]), 1); ...
|
|
||||||
], 1); % this are the time-stamps at which we want to initialize a new node in the graph
|
|
||||||
|
|
||||||
t_previous = 0;
|
|
||||||
poseIndex = 0;
|
|
||||||
for measurementIndex = 1:size(times,1)
|
|
||||||
% At each non=IMU measurement we initialize a new node in the graph
|
|
||||||
currentPoseKey = symbol('x',poseIndex);
|
|
||||||
currentVelKey = symbol('v',poseIndex);
|
|
||||||
currentBiasKey = symbol('b',poseIndex);
|
|
||||||
|
|
||||||
t = times(measurementIndex, 1);
|
|
||||||
type = times(measurementIndex, 2);
|
|
||||||
|
|
||||||
if type == 3
|
|
||||||
% Integrate IMU
|
|
||||||
|
|
||||||
if isempty(currentSummarizedMeasurement)
|
|
||||||
% Create initial empty summarized measurement
|
|
||||||
% we assume that each row of the IMU.txt file has the following structure:
|
|
||||||
% timestamp delta_t acc_x acc_y acc_z omega_x omega_y omega_z
|
|
||||||
currentBias = isam.calculateEstimate(currentBiasKey - 1);
|
|
||||||
currentSummarizedMeasurement = gtsam.ImuFactorPreintegratedMeasurements( ...
|
|
||||||
currentBias, IMU_metadata.AccelerometerSigma.^2 * eye(3), ...
|
|
||||||
IMU_metadata.GyroscopeSigma.^2 * eye(3), IMU_metadata.IntegrationSigma.^2 * eye(3));
|
|
||||||
end
|
|
||||||
|
|
||||||
% Accumulate preintegrated measurement
|
|
||||||
deltaT = IMU_data(index).dt;
|
|
||||||
accMeas = IMU_data(index).acc_omega(1:3);
|
|
||||||
omegaMeas = IMU_data(index).acc_omega(4:6);
|
|
||||||
currentSummarizedMeasurement.integrateMeasurement(accMeas, omegaMeas, deltaT);
|
|
||||||
|
|
||||||
else
|
|
||||||
% Create IMU factor
|
|
||||||
factors.add(ImuFactor( ...
|
|
||||||
currentPoseKey-1, currentVelKey-1, ...
|
|
||||||
currentPoseKey, currentVelKey, ...
|
|
||||||
currentBiasKey-1, currentSummarizedMeasurement, g, cor_v, ...
|
|
||||||
currentSummarizedMeasurement.PreintMeasCov));
|
|
||||||
|
|
||||||
% Reset summarized measurement
|
|
||||||
currentSummarizedMeasurement = [];
|
|
||||||
|
|
||||||
if type == 1
|
|
||||||
% Create VO factor
|
|
||||||
elseif type == 2
|
|
||||||
% Create GPS factor
|
|
||||||
end
|
|
||||||
|
|
||||||
poseIndex = poseIndex + 1;
|
|
||||||
end
|
|
||||||
|
|
||||||
|
|
||||||
% =======================================================================
|
|
||||||
|
|
||||||
|
|
||||||
%% add factor corresponding to GPS measurements (if available at the current time)
|
|
||||||
% % =======================================================================
|
|
||||||
% if isempty( find(GPS_data(:,1) == t ) ) == 0 % it is a GPS measurement
|
|
||||||
% if length( find(GPS_data(:,1)) ) > 1
|
|
||||||
% error('more GPS measurements at the same time stamp: it should be an error')
|
|
||||||
% end
|
|
||||||
%
|
|
||||||
% index = find(GPS_data(:,1) == t ); % the row of the IMU_data matrix that we have to integrate
|
|
||||||
% GPSmeas = GPS_data(index,2:4);
|
|
||||||
%
|
|
||||||
% noiseModelGPS = ???; % noiseModelGPS.Isotropic.Sigma(6, sigma_init_x))
|
|
||||||
%
|
|
||||||
% % add factor
|
|
||||||
% disp('TODO: is the GPS noise right?')
|
|
||||||
% factors.add(PriorFactor???(currentPoseKey, GPSmeas, noiseModelGPS) );
|
|
||||||
% end
|
|
||||||
% =======================================================================
|
|
||||||
|
|
||||||
|
|
||||||
%% add factor corresponding to VO measurements (if available at the current time)
|
|
||||||
% =======================================================================
|
|
||||||
if isempty( find([VO_data.Time] == t, 1) )== 0 % it is a GPS measurement
|
|
||||||
if length( find([VO_data.Time] == t) ) > 1
|
|
||||||
error('more VO measurements at the same time stamp: it should be an error')
|
|
||||||
end
|
|
||||||
|
|
||||||
index = find([VO_data.Time] == t, 1); % the row of the IMU_data matrix that we have to integrate
|
|
||||||
|
|
||||||
VOpose = VO_data(index).RelativePose;
|
|
||||||
noiseModelVO = noiseModel.Diagonal.Sigmas([ IMU_metadata.RotationSigma * [1;1;1]; IMU_metadata.TranslationSigma * [1;1;1] ]);
|
|
||||||
|
|
||||||
% add factor
|
|
||||||
disp('TODO: is the VO noise right?')
|
|
||||||
factors.add(BetweenFactorPose3(lastVOPoseKey, currentPoseKey, VOpose, noiseModelVO));
|
|
||||||
|
|
||||||
lastVOPoseKey = currentPoseKey;
|
|
||||||
end
|
|
||||||
% =======================================================================
|
|
||||||
|
|
||||||
disp('TODO: add values')
|
|
||||||
% values.insert(, initialPose);
|
|
||||||
% values.insert(symbol('v',lastIndex+1), initialVel);
|
|
||||||
|
|
||||||
%% Update solver
|
|
||||||
% =======================================================================
|
|
||||||
isam.update(factors, values);
|
|
||||||
factors = NonlinearFactorGraph;
|
|
||||||
values = Values;
|
|
||||||
|
|
||||||
isam.calculateEstimate(currentPoseKey);
|
|
||||||
% M = isam.marginalCovariance(key_pose);
|
|
||||||
% =======================================================================
|
|
||||||
|
|
||||||
previousPoseKey = currentPoseKey;
|
|
||||||
previousVelKey = currentVelKey;
|
|
||||||
t_previous = t;
|
|
||||||
end
|
|
||||||
|
|
||||||
disp('TODO: display results')
|
|
||||||
% figure(1)
|
|
||||||
% hold on;
|
|
||||||
% plot(positions(1,:), positions(2,:), '-b');
|
|
||||||
% plot3DTrajectory(isam.calculateEstimate, 'g-');
|
|
||||||
% axis equal;
|
|
||||||
% legend('true trajectory', 'traj integrated in body', 'traj integrated in nav')
|
|
||||||
|
|
@ -0,0 +1,191 @@
|
||||||
|
close all
|
||||||
|
clc
|
||||||
|
|
||||||
|
import gtsam.*;
|
||||||
|
disp('Example of application of ISAM2 for visual-inertial navigation on the KITTI VISION BENCHMARK SUITE (http://www.computervisiononline.com/dataset/kitti-vision-benchmark-suite)')
|
||||||
|
|
||||||
|
%% Read metadata and compute relative sensor pose transforms
|
||||||
|
% IMU metadata
|
||||||
|
disp('-- Reading sensor metadata')
|
||||||
|
IMU_metadata = importdata('KittiEquivBiasedImu_metadata.txt');
|
||||||
|
IMU_metadata = cell2struct(num2cell(IMU_metadata.data), IMU_metadata.colheaders, 2);
|
||||||
|
IMUinBody = Pose3.Expmap([IMU_metadata.BodyPtx; IMU_metadata.BodyPty; IMU_metadata.BodyPtz;
|
||||||
|
IMU_metadata.BodyPrx; IMU_metadata.BodyPry; IMU_metadata.BodyPrz; ]);
|
||||||
|
if ~IMUinBody.equals(Pose3, 1e-5)
|
||||||
|
error 'Currently only support IMUinBody is identity, i.e. IMU and body frame are the same';
|
||||||
|
end
|
||||||
|
|
||||||
|
% VO metadata
|
||||||
|
VO_metadata = importdata('KittiRelativePose_metadata.txt');
|
||||||
|
VO_metadata = cell2struct(num2cell(VO_metadata.data), VO_metadata.colheaders, 2);
|
||||||
|
VOinBody = Pose3.Expmap([VO_metadata.BodyPtx; VO_metadata.BodyPty; VO_metadata.BodyPtz;
|
||||||
|
VO_metadata.BodyPrx; VO_metadata.BodyPry; VO_metadata.BodyPrz; ]);
|
||||||
|
VOinIMU = IMUinBody.inverse().compose(VOinBody);
|
||||||
|
|
||||||
|
% GPS metadata
|
||||||
|
GPS_metadata = importdata('KittiGps_metadata.txt');
|
||||||
|
GPS_metadata = cell2struct(num2cell(GPS_metadata.data), GPS_metadata.colheaders, 2);
|
||||||
|
GPSinBody = Pose3.Expmap([GPS_metadata.BodyPtx; GPS_metadata.BodyPty; GPS_metadata.BodyPtz;
|
||||||
|
GPS_metadata.BodyPrx; GPS_metadata.BodyPry; GPS_metadata.BodyPrz; ]);
|
||||||
|
GPSinIMU = IMUinBody.inverse().compose(GPSinBody);
|
||||||
|
|
||||||
|
%% Read data and change coordinate frame of GPS and VO measurements to IMU frame
|
||||||
|
disp('-- Reading sensor data from file')
|
||||||
|
% IMU data
|
||||||
|
IMU_data = importdata('KittiEquivBiasedImu.txt');
|
||||||
|
IMU_data = cell2struct(num2cell(IMU_data.data), IMU_data.colheaders, 2);
|
||||||
|
imum = cellfun(@(x) x', num2cell([ [IMU_data.accelX]' [IMU_data.accelY]' [IMU_data.accelZ]' [IMU_data.omegaX]' [IMU_data.omegaY]' [IMU_data.omegaZ]' ], 2), 'UniformOutput', false);
|
||||||
|
[IMU_data.acc_omega] = deal(imum{:});
|
||||||
|
%IMU_data = rmfield(IMU_data, { 'accelX' 'accelY' 'accelZ' 'omegaX' 'omegaY' 'omegaZ' });
|
||||||
|
clear imum
|
||||||
|
|
||||||
|
% VO data
|
||||||
|
VO_data = importdata('KittiRelativePose.txt');
|
||||||
|
VO_data = cell2struct(num2cell(VO_data.data), VO_data.colheaders, 2);
|
||||||
|
% Merge relative pose fields and convert to Pose3
|
||||||
|
logposes = [ [VO_data.dtx]' [VO_data.dty]' [VO_data.dtz]' [VO_data.drx]' [VO_data.dry]' [VO_data.drz]' ];
|
||||||
|
logposes = num2cell(logposes, 2);
|
||||||
|
relposes = arrayfun(@(x) {gtsam.Pose3.Expmap(x{:}')}, logposes);
|
||||||
|
relposes = arrayfun(@(x) {VOinIMU.compose(x{:}).compose(VOinIMU.inverse())}, relposes);
|
||||||
|
[VO_data.RelativePose] = deal(relposes{:});
|
||||||
|
VO_data = rmfield(VO_data, { 'dtx' 'dty' 'dtz' 'drx' 'dry' 'drz' });
|
||||||
|
noiseModelVO = noiseModel.Diagonal.Sigmas([ VO_metadata.RotationSigma * [1;1;1]; VO_metadata.TranslationSigma * [1;1;1] ]);
|
||||||
|
clear logposes relposes
|
||||||
|
|
||||||
|
% % % GPS data
|
||||||
|
% % GPS_data = importdata('KittiGps.txt');
|
||||||
|
% % GPS_data = cell2struct(num2cell(GPS_data.data), GPS_data.colheaders, 2);
|
||||||
|
% % % Convert GPS from lat/long to meters
|
||||||
|
% % [ x, y, ~ ] = deg2utm( [GPS_data.Latitude], [GPS_data.Longitude] );
|
||||||
|
% % for i = 1:numel(x)
|
||||||
|
% % GPS_data(i).Position = gtsam.Point3(x(i), y(i), GPS_data(i).Altitude);
|
||||||
|
% % end
|
||||||
|
% % % % Calculate GPS sigma in meters
|
||||||
|
% % % [ xSig, ySig, ~ ] = deg2utm( [GPS_data.Latitude] + [GPS_data.PositionSigma], ...
|
||||||
|
% % % [GPS_data.Longitude] + [GPS_data.PositionSigma]);
|
||||||
|
% % % xSig = xSig - x;
|
||||||
|
% % % ySig = ySig - y;
|
||||||
|
% % %% Start at time of first GPS measurement
|
||||||
|
% % % firstGPSPose = 1;
|
||||||
|
|
||||||
|
%% Get initial conditions for the estimated trajectory
|
||||||
|
% % % currentPoseGlobal = Pose3(Rot3, GPS_data(firstGPSPose).Position); % initial pose is the reference frame (navigation frame)
|
||||||
|
currentPoseGlobal = Pose3;
|
||||||
|
currentVelocityGlobal = LieVector([0;0;0]); % the vehicle is stationary at the beginning
|
||||||
|
currentBias = imuBias.ConstantBias(zeros(3,1), zeros(3,1));
|
||||||
|
sigma_init_x = noiseModel.Isotropic.Sigma(6, 0.01);
|
||||||
|
sigma_init_v = noiseModel.Isotropic.Sigma(3, 1000.0);
|
||||||
|
sigma_init_b = noiseModel.Isotropic.Sigma(6, 0.01);
|
||||||
|
g = [0;0;-9.8];
|
||||||
|
w_coriolis = [0;0;0];
|
||||||
|
|
||||||
|
%% Solver object
|
||||||
|
isamParams = ISAM2Params;
|
||||||
|
isamParams.setFactorization('QR');
|
||||||
|
isamParams.setRelinearizeSkip(1);
|
||||||
|
isam = gtsam.ISAM2(isamParams);
|
||||||
|
newFactors = NonlinearFactorGraph;
|
||||||
|
newValues = Values;
|
||||||
|
|
||||||
|
%% Main loop:
|
||||||
|
% (1) we read the measurements
|
||||||
|
% (2) we create the corresponding factors in the graph
|
||||||
|
% (3) we solve the graph to obtain and optimal estimate of robot trajectory
|
||||||
|
timestamps = sortrows( [ ...
|
||||||
|
[VO_data.Time]' 1*ones(length([VO_data.Time]), 1); ...
|
||||||
|
% % %[GPS_data.Time]' 2*ones(length([GPS_data.Time]), 1); ...
|
||||||
|
], 1); % this are the time-stamps at which we want to initialize a new node in the graph
|
||||||
|
|
||||||
|
timestamps = timestamps(15:end,:); % there seem to be issues with the initial IMU measurements
|
||||||
|
IMUtimes = [IMU_data.Time];
|
||||||
|
VOPoseKeys = []; % here we store the keys of the poses involved in VO (between) factors
|
||||||
|
|
||||||
|
for measurementIndex = 1:length(timestamps)
|
||||||
|
|
||||||
|
% At each non=IMU measurement we initialize a new node in the graph
|
||||||
|
currentPoseKey = symbol('x',measurementIndex);
|
||||||
|
currentVelKey = symbol('v',measurementIndex);
|
||||||
|
currentBiasKey = symbol('b',measurementIndex);
|
||||||
|
t = timestamps(measurementIndex, 1);
|
||||||
|
type = timestamps(measurementIndex, 2);
|
||||||
|
|
||||||
|
%% bookkeeping
|
||||||
|
if type == 1 % we store the keys corresponding to VO measurements
|
||||||
|
VOPoseKeys = [VOPoseKeys; currentPoseKey];
|
||||||
|
end
|
||||||
|
|
||||||
|
if measurementIndex == 1
|
||||||
|
%% Create initial estimate and prior on initial pose, velocity, and biases
|
||||||
|
newValues.insert(currentPoseKey, currentPoseGlobal);
|
||||||
|
newValues.insert(currentVelKey, currentVelocityGlobal);
|
||||||
|
newValues.insert(currentBiasKey, currentBias);
|
||||||
|
newFactors.add(PriorFactorPose3(currentPoseKey, currentPoseGlobal, sigma_init_x));
|
||||||
|
newFactors.add(PriorFactorLieVector(currentVelKey, currentVelocityGlobal, sigma_init_v));
|
||||||
|
newFactors.add(PriorFactorConstantBias(currentBiasKey, currentBias, sigma_init_b));
|
||||||
|
else
|
||||||
|
t_previous = timestamps(measurementIndex-1, 1);
|
||||||
|
%% Summarize IMU data between the previous GPS measurement and now
|
||||||
|
IMUindices = find(IMUtimes >= t_previous & IMUtimes <= t);
|
||||||
|
|
||||||
|
if ~isempty(IMUindices) % if there are IMU measurements to integrate
|
||||||
|
currentSummarizedMeasurement = gtsam.ImuFactorPreintegratedMeasurements( ...
|
||||||
|
currentBias, IMU_metadata.AccelerometerSigma.^2 * eye(3), ...
|
||||||
|
IMU_metadata.GyroscopeSigma.^2 * eye(3), IMU_metadata.IntegrationSigma.^2 * eye(3));
|
||||||
|
|
||||||
|
for imuIndex = IMUindices
|
||||||
|
accMeas = [ IMU_data(imuIndex).accelX; IMU_data(imuIndex).accelY; IMU_data(imuIndex).accelZ ];
|
||||||
|
omegaMeas = [ IMU_data(imuIndex).omegaX; IMU_data(imuIndex).omegaY; IMU_data(imuIndex).omegaZ ];
|
||||||
|
deltaT = IMU_data(imuIndex).dt;
|
||||||
|
currentSummarizedMeasurement.integrateMeasurement(accMeas, omegaMeas, deltaT);
|
||||||
|
end
|
||||||
|
|
||||||
|
% Create IMU factor
|
||||||
|
newFactors.add(ImuFactor( ...
|
||||||
|
currentPoseKey-1, currentVelKey-1, ...
|
||||||
|
currentPoseKey, currentVelKey, ...
|
||||||
|
currentBiasKey, currentSummarizedMeasurement, g, w_coriolis));
|
||||||
|
|
||||||
|
else % if there are no IMU measurements
|
||||||
|
error('no IMU measurements in [t_previous, t]')
|
||||||
|
end
|
||||||
|
|
||||||
|
% LC: sigma_init_b is wrong: this should be some uncertainty on bias evolution given in the IMU metadata
|
||||||
|
newFactors.add(BetweenFactorConstantBias(currentBiasKey-1, currentBiasKey, imuBias.ConstantBias(zeros(3,1), zeros(3,1)), sigma_init_b));
|
||||||
|
|
||||||
|
%% Create GPS factor
|
||||||
|
if type == 2
|
||||||
|
newFactors.add(PriorFactorPose3(currentPoseKey, Pose3(currentPoseGlobal.rotation, GPS_data(measurementIndex).Position), ...
|
||||||
|
noiseModel.Diagonal.Precisions([ zeros(3,1); 1./(GPS_data(measurementIndex).PositionSigma).^2*ones(3,1) ])));
|
||||||
|
end
|
||||||
|
|
||||||
|
%% Create VO factor
|
||||||
|
if type == 1
|
||||||
|
VOpose = VO_data(measurementIndex).RelativePose;
|
||||||
|
newFactors.add(BetweenFactorPose3(VOPoseKeys(end-1), VOPoseKeys(end), VOpose, noiseModelVO));
|
||||||
|
end
|
||||||
|
|
||||||
|
% Add initial value
|
||||||
|
% newValues.insert(currentPoseKey, Pose3(currentPoseGlobal.rotation, GPS_data(measurementIndex).Position));
|
||||||
|
newValues.insert(currentPoseKey,currentPoseGlobal);
|
||||||
|
newValues.insert(currentVelKey, currentVelocityGlobal);
|
||||||
|
newValues.insert(currentBiasKey, currentBias);
|
||||||
|
|
||||||
|
% Update solver
|
||||||
|
% =======================================================================
|
||||||
|
isam.update(newFactors, newValues);
|
||||||
|
newFactors = NonlinearFactorGraph;
|
||||||
|
newValues = Values;
|
||||||
|
|
||||||
|
if rem(measurementIndex,100)==0 % plot every 100 time steps
|
||||||
|
cla;
|
||||||
|
plot3DTrajectory(isam.calculateEstimate, 'g-');
|
||||||
|
axis equal
|
||||||
|
drawnow;
|
||||||
|
end
|
||||||
|
% =======================================================================
|
||||||
|
currentPoseGlobal = isam.calculateEstimate(currentPoseKey);
|
||||||
|
currentVelocityGlobal = isam.calculateEstimate(currentVelKey);
|
||||||
|
currentBias = isam.calculateEstimate(currentBiasKey);
|
||||||
|
end
|
||||||
|
|
||||||
|
end % end main loop
|
||||||
|
|
@ -0,0 +1,149 @@
|
||||||
|
close all
|
||||||
|
clc
|
||||||
|
|
||||||
|
import gtsam.*;
|
||||||
|
disp('Example of application of ISAM2 for GPS-aided navigation on the KITTI VISION BENCHMARK SUITE (http://www.computervisiononline.com/dataset/kitti-vision-benchmark-suite)')
|
||||||
|
|
||||||
|
%% Read metadata and compute relative sensor pose transforms
|
||||||
|
% IMU metadata
|
||||||
|
disp('-- Reading sensor metadata')
|
||||||
|
IMU_metadata = importdata('KittiEquivBiasedImu_metadata.txt');
|
||||||
|
IMU_metadata = cell2struct(num2cell(IMU_metadata.data), IMU_metadata.colheaders, 2);
|
||||||
|
IMUinBody = Pose3.Expmap([IMU_metadata.BodyPtx; IMU_metadata.BodyPty; IMU_metadata.BodyPtz;
|
||||||
|
IMU_metadata.BodyPrx; IMU_metadata.BodyPry; IMU_metadata.BodyPrz; ]);
|
||||||
|
if ~IMUinBody.equals(Pose3, 1e-5)
|
||||||
|
error 'Currently only support IMUinBody is identity, i.e. IMU and body frame are the same';
|
||||||
|
end
|
||||||
|
|
||||||
|
% GPS metadata
|
||||||
|
GPS_metadata = importdata('KittiRelativePose_metadata.txt');
|
||||||
|
GPS_metadata = cell2struct(num2cell(GPS_metadata.data), GPS_metadata.colheaders, 2);
|
||||||
|
|
||||||
|
%% Read data
|
||||||
|
disp('-- Reading sensor data from file')
|
||||||
|
% IMU data
|
||||||
|
IMU_data = importdata('KittiEquivBiasedImu.txt');
|
||||||
|
IMU_data = cell2struct(num2cell(IMU_data.data), IMU_data.colheaders, 2);
|
||||||
|
imum = cellfun(@(x) x', num2cell([ [IMU_data.accelX]' [IMU_data.accelY]' [IMU_data.accelZ]' [IMU_data.omegaX]' [IMU_data.omegaY]' [IMU_data.omegaZ]' ], 2), 'UniformOutput', false);
|
||||||
|
[IMU_data.acc_omega] = deal(imum{:});
|
||||||
|
clear imum
|
||||||
|
|
||||||
|
% GPS data
|
||||||
|
GPS_data = importdata('Gps_converted.txt');
|
||||||
|
GPS_data = cell2struct(num2cell(GPS_data.data), GPS_data.colheaders, 2);
|
||||||
|
for i = 1:numel(GPS_data)
|
||||||
|
GPS_data(i).Position = gtsam.Point3(GPS_data(i).X, GPS_data(i).Y, GPS_data(i).Z);
|
||||||
|
end
|
||||||
|
noiseModelGPS = noiseModel.Diagonal.Precisions([ [0;0;0]; 1.0/0.07 * [1;1;1] ]);
|
||||||
|
firstGPSPose = 2;
|
||||||
|
GPSskip = 10; % Skip this many GPS measurements each time
|
||||||
|
|
||||||
|
%% Get initial conditions for the estimated trajectory
|
||||||
|
currentPoseGlobal = Pose3(Rot3, GPS_data(firstGPSPose).Position); % initial pose is the reference frame (navigation frame)
|
||||||
|
currentVelocityGlobal = LieVector([0;0;0]); % the vehicle is stationary at the beginning
|
||||||
|
currentBias = imuBias.ConstantBias(zeros(3,1), zeros(3,1));
|
||||||
|
sigma_init_x = noiseModel.Isotropic.Precisions([ 0.0; 0.0; 0.0; 1; 1; 1 ]);
|
||||||
|
sigma_init_v = noiseModel.Isotropic.Sigma(3, 1000.0);
|
||||||
|
sigma_init_b = noiseModel.Isotropic.Sigmas([ 0.100; 0.100; 0.100; 5.00e-05; 5.00e-05; 5.00e-05 ]);
|
||||||
|
sigma_between_b = [ IMU_metadata.AccelerometerBiasSigma * ones(3,1); IMU_metadata.GyroscopeBiasSigma * ones(3,1) ];
|
||||||
|
g = [0;0;-9.8];
|
||||||
|
w_coriolis = [0;0;0];
|
||||||
|
|
||||||
|
%% Solver object
|
||||||
|
isamParams = ISAM2Params;
|
||||||
|
isamParams.setFactorization('CHOLESKY');
|
||||||
|
isamParams.setRelinearizeSkip(10);
|
||||||
|
isam = gtsam.ISAM2(isamParams);
|
||||||
|
newFactors = NonlinearFactorGraph;
|
||||||
|
newValues = Values;
|
||||||
|
|
||||||
|
%% Main loop:
|
||||||
|
% (1) we read the measurements
|
||||||
|
% (2) we create the corresponding factors in the graph
|
||||||
|
% (3) we solve the graph to obtain and optimal estimate of robot trajectory
|
||||||
|
IMUtimes = [IMU_data.Time];
|
||||||
|
|
||||||
|
disp('-- Starting main loop: inference is performed at each time step, but we plot trajectory every 10 steps')
|
||||||
|
|
||||||
|
for measurementIndex = firstGPSPose:length(GPS_data)
|
||||||
|
|
||||||
|
% At each non=IMU measurement we initialize a new node in the graph
|
||||||
|
currentPoseKey = symbol('x',measurementIndex);
|
||||||
|
currentVelKey = symbol('v',measurementIndex);
|
||||||
|
currentBiasKey = symbol('b',measurementIndex);
|
||||||
|
t = GPS_data(measurementIndex, 1).Time;
|
||||||
|
|
||||||
|
if measurementIndex == firstGPSPose
|
||||||
|
%% Create initial estimate and prior on initial pose, velocity, and biases
|
||||||
|
newValues.insert(currentPoseKey, currentPoseGlobal);
|
||||||
|
newValues.insert(currentVelKey, currentVelocityGlobal);
|
||||||
|
newValues.insert(currentBiasKey, currentBias);
|
||||||
|
newFactors.add(PriorFactorPose3(currentPoseKey, currentPoseGlobal, sigma_init_x));
|
||||||
|
newFactors.add(PriorFactorLieVector(currentVelKey, currentVelocityGlobal, sigma_init_v));
|
||||||
|
newFactors.add(PriorFactorConstantBias(currentBiasKey, currentBias, sigma_init_b));
|
||||||
|
else
|
||||||
|
t_previous = GPS_data(measurementIndex-1, 1).Time;
|
||||||
|
%% Summarize IMU data between the previous GPS measurement and now
|
||||||
|
IMUindices = find(IMUtimes >= t_previous & IMUtimes <= t);
|
||||||
|
|
||||||
|
currentSummarizedMeasurement = gtsam.ImuFactorPreintegratedMeasurements( ...
|
||||||
|
currentBias, IMU_metadata.AccelerometerSigma.^2 * eye(3), ...
|
||||||
|
IMU_metadata.GyroscopeSigma.^2 * eye(3), IMU_metadata.IntegrationSigma.^2 * eye(3));
|
||||||
|
|
||||||
|
for imuIndex = IMUindices
|
||||||
|
accMeas = [ IMU_data(imuIndex).accelX; IMU_data(imuIndex).accelY; IMU_data(imuIndex).accelZ ];
|
||||||
|
omegaMeas = [ IMU_data(imuIndex).omegaX; IMU_data(imuIndex).omegaY; IMU_data(imuIndex).omegaZ ];
|
||||||
|
deltaT = IMU_data(imuIndex).dt;
|
||||||
|
currentSummarizedMeasurement.integrateMeasurement(accMeas, omegaMeas, deltaT);
|
||||||
|
end
|
||||||
|
|
||||||
|
% Create IMU factor
|
||||||
|
newFactors.add(ImuFactor( ...
|
||||||
|
currentPoseKey-1, currentVelKey-1, ...
|
||||||
|
currentPoseKey, currentVelKey, ...
|
||||||
|
currentBiasKey, currentSummarizedMeasurement, g, w_coriolis));
|
||||||
|
|
||||||
|
% Bias evolution as given in the IMU metadata
|
||||||
|
newFactors.add(BetweenFactorConstantBias(currentBiasKey-1, currentBiasKey, imuBias.ConstantBias(zeros(3,1), zeros(3,1)), ...
|
||||||
|
noiseModel.Diagonal.Sigmas(sqrt(numel(IMUindices)) * sigma_between_b)));
|
||||||
|
|
||||||
|
% Create GPS factor
|
||||||
|
GPSPose = Pose3(currentPoseGlobal.rotation, GPS_data(measurementIndex).Position);
|
||||||
|
if mod(measurementIndex, GPSskip) == 0
|
||||||
|
newFactors.add(PriorFactorPose3(currentPoseKey, GPSPose, noiseModelGPS));
|
||||||
|
end
|
||||||
|
|
||||||
|
% Add initial value
|
||||||
|
newValues.insert(currentPoseKey, GPSPose);
|
||||||
|
newValues.insert(currentVelKey, currentVelocityGlobal);
|
||||||
|
newValues.insert(currentBiasKey, currentBias);
|
||||||
|
|
||||||
|
% Update solver
|
||||||
|
% =======================================================================
|
||||||
|
% We accumulate 2*GPSskip GPS measurements before updating the solver at
|
||||||
|
% first so that the heading becomes observable.
|
||||||
|
if measurementIndex > firstGPSPose + 2*GPSskip
|
||||||
|
isam.update(newFactors, newValues);
|
||||||
|
newFactors = NonlinearFactorGraph;
|
||||||
|
newValues = Values;
|
||||||
|
|
||||||
|
if rem(measurementIndex,10)==0 % plot every 10 time steps
|
||||||
|
cla;
|
||||||
|
plot3DTrajectory(isam.calculateEstimate, 'g-');
|
||||||
|
title('Estimated trajectory using ISAM2 (IMU+GPS)')
|
||||||
|
xlabel('[m]')
|
||||||
|
ylabel('[m]')
|
||||||
|
zlabel('[m]')
|
||||||
|
axis equal
|
||||||
|
drawnow;
|
||||||
|
end
|
||||||
|
% =======================================================================
|
||||||
|
currentPoseGlobal = isam.calculateEstimate(currentPoseKey);
|
||||||
|
currentVelocityGlobal = isam.calculateEstimate(currentVelKey);
|
||||||
|
currentBias = isam.calculateEstimate(currentBiasKey);
|
||||||
|
end
|
||||||
|
end
|
||||||
|
|
||||||
|
end % end main loop
|
||||||
|
|
||||||
|
disp('-- Reached end of sensor data')
|
||||||
|
|
@ -1,126 +0,0 @@
|
||||||
%close all
|
|
||||||
%clc
|
|
||||||
|
|
||||||
import gtsam.*;
|
|
||||||
|
|
||||||
%% Read data
|
|
||||||
IMU_metadata = importdata(gtsam.findExampleDataFile('KittiEquivBiasedImu_metadata.txt'));
|
|
||||||
IMU_data = importdata(gtsam.findExampleDataFile('KittiEquivBiasedImu.txt'));
|
|
||||||
% Make text file column headers into struct fields
|
|
||||||
IMU_metadata = cell2struct(num2cell(IMU_metadata.data), IMU_metadata.colheaders, 2);
|
|
||||||
IMU_data = cell2struct(num2cell(IMU_data.data), IMU_data.colheaders, 2);
|
|
||||||
|
|
||||||
GPS_metadata = importdata(gtsam.findExampleDataFile('KittiGps_metadata.txt'));
|
|
||||||
GPS_data = importdata(gtsam.findExampleDataFile('KittiGps.txt'));
|
|
||||||
% Make text file column headers into struct fields
|
|
||||||
GPS_metadata = cell2struct(num2cell(GPS_metadata.data), GPS_metadata.colheaders, 2);
|
|
||||||
GPS_data = cell2struct(num2cell(GPS_data.data), GPS_data.colheaders, 2);
|
|
||||||
|
|
||||||
%% Convert GPS from lat/long to meters
|
|
||||||
[ x, y, ~ ] = deg2utm( [GPS_data.Latitude], [GPS_data.Longitude] );
|
|
||||||
for i = 1:numel(x)
|
|
||||||
GPS_data(i).Position = gtsam.Point3(x(i), y(i), GPS_data(i).Altitude);
|
|
||||||
end
|
|
||||||
|
|
||||||
% % Calculate GPS sigma in meters
|
|
||||||
% [ xSig, ySig, ~ ] = deg2utm( [GPS_data.Latitude] + [GPS_data.PositionSigma], ...
|
|
||||||
% [GPS_data.Longitude] + [GPS_data.PositionSigma]);
|
|
||||||
% xSig = xSig - x;
|
|
||||||
% ySig = ySig - y;
|
|
||||||
|
|
||||||
%% Start at time of first GPS measurement
|
|
||||||
firstGPSPose = 2;
|
|
||||||
|
|
||||||
%% Get initial conditions for the estimated trajectory
|
|
||||||
currentPoseGlobal = Pose3(Rot3, GPS_data(firstGPSPose).Position); % initial pose is the reference frame (navigation frame)
|
|
||||||
currentVelocityGlobal = LieVector([0;0;0]); % the vehicle is stationary at the beginning
|
|
||||||
currentBias = imuBias.ConstantBias(zeros(3,1), zeros(3,1));
|
|
||||||
|
|
||||||
%% Solver object
|
|
||||||
isamParams = ISAM2Params;
|
|
||||||
isamParams.setFactorization('QR');
|
|
||||||
isamParams.setRelinearizeSkip(1);
|
|
||||||
isam = gtsam.ISAM2(isamParams);
|
|
||||||
newFactors = NonlinearFactorGraph;
|
|
||||||
newValues = Values;
|
|
||||||
|
|
||||||
%% Create initial estimate and prior on initial pose, velocity, and biases
|
|
||||||
newValues.insert(symbol('x',firstGPSPose), currentPoseGlobal);
|
|
||||||
newValues.insert(symbol('v',firstGPSPose), currentVelocityGlobal);
|
|
||||||
newValues.insert(symbol('b',1), currentBias);
|
|
||||||
|
|
||||||
sigma_init_x = noiseModel.Diagonal.Precisions([0;0;0; 1;1;1]);
|
|
||||||
sigma_init_v = noiseModel.Isotropic.Sigma(3, 1000.0);
|
|
||||||
sigma_init_b = noiseModel.Isotropic.Sigma(6, 0.01);
|
|
||||||
|
|
||||||
newFactors.add(PriorFactorPose3(symbol('x',firstGPSPose), currentPoseGlobal, sigma_init_x));
|
|
||||||
newFactors.add(PriorFactorLieVector(symbol('v',firstGPSPose), currentVelocityGlobal, sigma_init_v));
|
|
||||||
newFactors.add(PriorFactorConstantBias(symbol('b',1), currentBias, sigma_init_b));
|
|
||||||
|
|
||||||
%% Main loop:
|
|
||||||
% (1) we read the measurements
|
|
||||||
% (2) we create the corresponding factors in the graph
|
|
||||||
% (3) we solve the graph to obtain and optimal estimate of robot trajectory
|
|
||||||
|
|
||||||
for poseIndex = firstGPSPose:length(GPS_data)
|
|
||||||
% At each non=IMU measurement we initialize a new node in the graph
|
|
||||||
currentPoseKey = symbol('x',poseIndex);
|
|
||||||
currentVelKey = symbol('v',poseIndex);
|
|
||||||
currentBiasKey = symbol('b',1);
|
|
||||||
|
|
||||||
if poseIndex > firstGPSPose
|
|
||||||
% Summarize IMU data between the previous GPS measurement and now
|
|
||||||
IMUindices = find([IMU_data.Time] > GPS_data(poseIndex-1).Time ...
|
|
||||||
& [IMU_data.Time] <= GPS_data(poseIndex).Time);
|
|
||||||
|
|
||||||
currentSummarizedMeasurement = gtsam.ImuFactorPreintegratedMeasurements( ...
|
|
||||||
currentBias, IMU_metadata.AccelerometerSigma.^2 * eye(3), ...
|
|
||||||
IMU_metadata.GyroscopeSigma.^2 * eye(3), IMU_metadata.IntegrationSigma.^2 * eye(3));
|
|
||||||
|
|
||||||
for imuIndex = IMUindices
|
|
||||||
accMeas = [ IMU_data(imuIndex).accelX; IMU_data(imuIndex).accelY; IMU_data(imuIndex).accelZ ];
|
|
||||||
omegaMeas = [ IMU_data(imuIndex).omegaX; IMU_data(imuIndex).omegaY; IMU_data(imuIndex).omegaZ ];
|
|
||||||
deltaT = IMU_data(imuIndex).dt;
|
|
||||||
currentSummarizedMeasurement.integrateMeasurement(accMeas, omegaMeas, deltaT);
|
|
||||||
end
|
|
||||||
|
|
||||||
% Create IMU factor
|
|
||||||
newFactors.add(ImuFactor( ...
|
|
||||||
currentPoseKey-1, currentVelKey-1, ...
|
|
||||||
currentPoseKey, currentVelKey, ...
|
|
||||||
currentBiasKey, currentSummarizedMeasurement, [0;0;-9.8], [0;0;0]));
|
|
||||||
|
|
||||||
% Create GPS factor
|
|
||||||
newFactors.add(PriorFactorPose3(currentPoseKey, Pose3(currentPoseGlobal.rotation, GPS_data(poseIndex).Position), ...
|
|
||||||
noiseModel.Diagonal.Precisions([ zeros(3,1); 1./(GPS_data(poseIndex).PositionSigma).^2*ones(3,1) ])));
|
|
||||||
|
|
||||||
% Add initial value
|
|
||||||
newValues.insert(currentPoseKey, Pose3(currentPoseGlobal.rotation, GPS_data(poseIndex).Position));
|
|
||||||
newValues.insert(currentVelKey, currentVelocityGlobal);
|
|
||||||
%newValues.insert(currentBiasKey, currentBias);
|
|
||||||
|
|
||||||
% Update solver
|
|
||||||
% =======================================================================
|
|
||||||
isam.update(newFactors, newValues);
|
|
||||||
newFactors = NonlinearFactorGraph;
|
|
||||||
newValues = Values;
|
|
||||||
|
|
||||||
cla;
|
|
||||||
plot3DTrajectory(isam.calculateEstimate, 'g-');
|
|
||||||
drawnow;
|
|
||||||
% =======================================================================
|
|
||||||
|
|
||||||
currentPoseGlobal = isam.calculateEstimate(currentPoseKey);
|
|
||||||
currentVelocityGlobal = isam.calculateEstimate(currentVelKey);
|
|
||||||
currentBias = isam.calculateEstimate(currentBiasKey);
|
|
||||||
|
|
||||||
end
|
|
||||||
end
|
|
||||||
|
|
||||||
disp('TODO: display results')
|
|
||||||
% figure(1)
|
|
||||||
% hold on;
|
|
||||||
% plot(positions(1,:), positions(2,:), '-b');
|
|
||||||
% plot3DTrajectory(isam.calculateEstimate, 'g-');
|
|
||||||
% axis equal;
|
|
||||||
% legend('true trajectory', 'traj integrated in body', 'traj integrated in nav')
|
|
||||||
|
|
@ -0,0 +1,152 @@
|
||||||
|
close all
|
||||||
|
clc
|
||||||
|
|
||||||
|
import gtsam.*;
|
||||||
|
disp('Example of application of ISAM2 for visual-inertial navigation on the KITTI VISION BENCHMARK SUITE (http://www.computervisiononline.com/dataset/kitti-vision-benchmark-suite)')
|
||||||
|
|
||||||
|
%% Read metadata and compute relative sensor pose transforms
|
||||||
|
% IMU metadata
|
||||||
|
disp('-- Reading sensor metadata')
|
||||||
|
IMU_metadata = importdata('KittiEquivBiasedImu_metadata.txt');
|
||||||
|
IMU_metadata = cell2struct(num2cell(IMU_metadata.data), IMU_metadata.colheaders, 2);
|
||||||
|
IMUinBody = Pose3.Expmap([IMU_metadata.BodyPtx; IMU_metadata.BodyPty; IMU_metadata.BodyPtz;
|
||||||
|
IMU_metadata.BodyPrx; IMU_metadata.BodyPry; IMU_metadata.BodyPrz; ]);
|
||||||
|
if ~IMUinBody.equals(Pose3, 1e-5)
|
||||||
|
error 'Currently only support IMUinBody is identity, i.e. IMU and body frame are the same';
|
||||||
|
end
|
||||||
|
|
||||||
|
% VO metadata
|
||||||
|
VO_metadata = importdata('KittiRelativePose_metadata.txt');
|
||||||
|
VO_metadata = cell2struct(num2cell(VO_metadata.data), VO_metadata.colheaders, 2);
|
||||||
|
VOinBody = Pose3.Expmap([VO_metadata.BodyPtx; VO_metadata.BodyPty; VO_metadata.BodyPtz;
|
||||||
|
VO_metadata.BodyPrx; VO_metadata.BodyPry; VO_metadata.BodyPrz; ]);
|
||||||
|
VOinIMU = IMUinBody.inverse().compose(VOinBody);
|
||||||
|
|
||||||
|
%% Read data and change coordinate frame of GPS and VO measurements to IMU frame
|
||||||
|
disp('-- Reading sensor data from file')
|
||||||
|
% IMU data
|
||||||
|
IMU_data = importdata('KittiEquivBiasedImu.txt');
|
||||||
|
IMU_data = cell2struct(num2cell(IMU_data.data), IMU_data.colheaders, 2);
|
||||||
|
imum = cellfun(@(x) x', num2cell([ [IMU_data.accelX]' [IMU_data.accelY]' [IMU_data.accelZ]' [IMU_data.omegaX]' [IMU_data.omegaY]' [IMU_data.omegaZ]' ], 2), 'UniformOutput', false);
|
||||||
|
[IMU_data.acc_omega] = deal(imum{:});
|
||||||
|
clear imum
|
||||||
|
|
||||||
|
% VO data
|
||||||
|
VO_data = importdata('KittiRelativePose.txt');
|
||||||
|
VO_data = cell2struct(num2cell(VO_data.data), VO_data.colheaders, 2);
|
||||||
|
% Merge relative pose fields and convert to Pose3
|
||||||
|
logposes = [ [VO_data.dtx]' [VO_data.dty]' [VO_data.dtz]' [VO_data.drx]' [VO_data.dry]' [VO_data.drz]' ];
|
||||||
|
logposes = num2cell(logposes, 2);
|
||||||
|
relposes = arrayfun(@(x) {gtsam.Pose3.Expmap(x{:}')}, logposes);
|
||||||
|
relposes = arrayfun(@(x) {VOinIMU.compose(x{:}).compose(VOinIMU.inverse())}, relposes);
|
||||||
|
[VO_data.RelativePose] = deal(relposes{:});
|
||||||
|
VO_data = rmfield(VO_data, { 'dtx' 'dty' 'dtz' 'drx' 'dry' 'drz' });
|
||||||
|
noiseModelVO = noiseModel.Diagonal.Sigmas([ VO_metadata.RotationSigma * [1;1;1]; VO_metadata.TranslationSigma * [1;1;1] ]);
|
||||||
|
clear logposes relposes
|
||||||
|
|
||||||
|
%% Get initial conditions for the estimated trajectory
|
||||||
|
currentPoseGlobal = Pose3;
|
||||||
|
currentVelocityGlobal = LieVector([0;0;0]); % the vehicle is stationary at the beginning
|
||||||
|
currentBias = imuBias.ConstantBias(zeros(3,1), zeros(3,1));
|
||||||
|
sigma_init_x = noiseModel.Isotropic.Sigmas([ 1.0; 1.0; 0.01; 0.01; 0.01; 0.01 ]);
|
||||||
|
sigma_init_v = noiseModel.Isotropic.Sigma(3, 1000.0);
|
||||||
|
sigma_init_b = noiseModel.Isotropic.Sigmas([ 0.100; 0.100; 0.100; 5.00e-05; 5.00e-05; 5.00e-05 ]);
|
||||||
|
sigma_between_b = [ IMU_metadata.AccelerometerBiasSigma * ones(3,1); IMU_metadata.GyroscopeBiasSigma * ones(3,1) ];
|
||||||
|
g = [0;0;-9.8];
|
||||||
|
w_coriolis = [0;0;0];
|
||||||
|
|
||||||
|
%% Solver object
|
||||||
|
isamParams = ISAM2Params;
|
||||||
|
isamParams.setFactorization('CHOLESKY');
|
||||||
|
isamParams.setRelinearizeSkip(10);
|
||||||
|
isam = gtsam.ISAM2(isamParams);
|
||||||
|
newFactors = NonlinearFactorGraph;
|
||||||
|
newValues = Values;
|
||||||
|
|
||||||
|
%% Main loop:
|
||||||
|
% (1) we read the measurements
|
||||||
|
% (2) we create the corresponding factors in the graph
|
||||||
|
% (3) we solve the graph to obtain and optimal estimate of robot trajectory
|
||||||
|
timestamps = [VO_data.Time]';
|
||||||
|
|
||||||
|
timestamps = timestamps(15:end,:); % there seem to be issues with the initial IMU measurements
|
||||||
|
IMUtimes = [IMU_data.Time];
|
||||||
|
|
||||||
|
disp('-- Starting main loop: inference is performed at each time step, but we plot trajectory every 100 steps')
|
||||||
|
|
||||||
|
for measurementIndex = 1:length(timestamps)
|
||||||
|
|
||||||
|
% At each non=IMU measurement we initialize a new node in the graph
|
||||||
|
currentPoseKey = symbol('x',measurementIndex);
|
||||||
|
currentVelKey = symbol('v',measurementIndex);
|
||||||
|
currentBiasKey = symbol('b',measurementIndex);
|
||||||
|
t = timestamps(measurementIndex, 1);
|
||||||
|
|
||||||
|
if measurementIndex == 1
|
||||||
|
%% Create initial estimate and prior on initial pose, velocity, and biases
|
||||||
|
newValues.insert(currentPoseKey, currentPoseGlobal);
|
||||||
|
newValues.insert(currentVelKey, currentVelocityGlobal);
|
||||||
|
newValues.insert(currentBiasKey, currentBias);
|
||||||
|
newFactors.add(PriorFactorPose3(currentPoseKey, currentPoseGlobal, sigma_init_x));
|
||||||
|
newFactors.add(PriorFactorLieVector(currentVelKey, currentVelocityGlobal, sigma_init_v));
|
||||||
|
newFactors.add(PriorFactorConstantBias(currentBiasKey, currentBias, sigma_init_b));
|
||||||
|
else
|
||||||
|
t_previous = timestamps(measurementIndex-1, 1);
|
||||||
|
%% Summarize IMU data between the previous GPS measurement and now
|
||||||
|
IMUindices = find(IMUtimes >= t_previous & IMUtimes <= t);
|
||||||
|
|
||||||
|
currentSummarizedMeasurement = gtsam.ImuFactorPreintegratedMeasurements( ...
|
||||||
|
currentBias, IMU_metadata.AccelerometerSigma.^2 * eye(3), ...
|
||||||
|
IMU_metadata.GyroscopeSigma.^2 * eye(3), IMU_metadata.IntegrationSigma.^2 * eye(3));
|
||||||
|
|
||||||
|
for imuIndex = IMUindices
|
||||||
|
accMeas = [ IMU_data(imuIndex).accelX; IMU_data(imuIndex).accelY; IMU_data(imuIndex).accelZ ];
|
||||||
|
omegaMeas = [ IMU_data(imuIndex).omegaX; IMU_data(imuIndex).omegaY; IMU_data(imuIndex).omegaZ ];
|
||||||
|
deltaT = IMU_data(imuIndex).dt;
|
||||||
|
currentSummarizedMeasurement.integrateMeasurement(accMeas, omegaMeas, deltaT);
|
||||||
|
end
|
||||||
|
|
||||||
|
% Create IMU factor
|
||||||
|
newFactors.add(ImuFactor( ...
|
||||||
|
currentPoseKey-1, currentVelKey-1, ...
|
||||||
|
currentPoseKey, currentVelKey, ...
|
||||||
|
currentBiasKey, currentSummarizedMeasurement, g, w_coriolis));
|
||||||
|
|
||||||
|
% LC: sigma_init_b is wrong: this should be some uncertainty on bias evolution given in the IMU metadata
|
||||||
|
newFactors.add(BetweenFactorConstantBias(currentBiasKey-1, currentBiasKey, imuBias.ConstantBias(zeros(3,1), zeros(3,1)), ...
|
||||||
|
noiseModel.Diagonal.Sigmas(sqrt(numel(IMUindices)) * sigma_between_b)));
|
||||||
|
|
||||||
|
%% Create VO factor
|
||||||
|
VOpose = VO_data(measurementIndex).RelativePose;
|
||||||
|
newFactors.add(BetweenFactorPose3(currentPoseKey - 1, currentPoseKey, VOpose, noiseModelVO));
|
||||||
|
|
||||||
|
% Add initial value
|
||||||
|
newValues.insert(currentPoseKey, currentPoseGlobal.compose(VOpose));
|
||||||
|
newValues.insert(currentVelKey, currentVelocityGlobal);
|
||||||
|
newValues.insert(currentBiasKey, currentBias);
|
||||||
|
|
||||||
|
% Update solver
|
||||||
|
% =======================================================================
|
||||||
|
isam.update(newFactors, newValues);
|
||||||
|
newFactors = NonlinearFactorGraph;
|
||||||
|
newValues = Values;
|
||||||
|
|
||||||
|
if rem(measurementIndex,100)==0 % plot every 100 time steps
|
||||||
|
cla;
|
||||||
|
plot3DTrajectory(isam.calculateEstimate, 'g-');
|
||||||
|
title('Estimated trajectory using ISAM2 (IMU+VO)')
|
||||||
|
xlabel('[m]')
|
||||||
|
ylabel('[m]')
|
||||||
|
zlabel('[m]')
|
||||||
|
axis equal
|
||||||
|
drawnow;
|
||||||
|
end
|
||||||
|
% =======================================================================
|
||||||
|
currentPoseGlobal = isam.calculateEstimate(currentPoseKey);
|
||||||
|
currentVelocityGlobal = isam.calculateEstimate(currentVelKey);
|
||||||
|
currentBias = isam.calculateEstimate(currentBiasKey);
|
||||||
|
end
|
||||||
|
|
||||||
|
end % end main loop
|
||||||
|
|
||||||
|
disp('-- Reached end of sensor data')
|
||||||
Loading…
Reference in New Issue