Created a nonlinear "AntiFactor" to perform downdating
Stephen Williams
parent
918019c605
commit
3e6054122c
|
|
@ -0,0 +1,138 @@
|
|||
/* ----------------------------------------------------------------------------
|
||||
|
||||
* 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 AntiFactor.h
|
||||
* @author Stephen Williams
|
||||
**/
|
||||
#pragma once
|
||||
|
||||
#include <ostream>
|
||||
|
||||
#include <gtsam/nonlinear/NonlinearFactor.h>
|
||||
#include <gtsam/linear/GaussianFactor.h>
|
||||
|
||||
namespace gtsam {
|
||||
|
||||
/**
|
||||
* A class for a measurement predicted by "between(config[key1],config[key2])"
|
||||
* KEY1::Value is the Lie Group type
|
||||
* T is the measurement type, by default the same
|
||||
*/
|
||||
template<class VALUES>
|
||||
class AntiFactor: public NonlinearFactor<VALUES> {
|
||||
|
||||
private:
|
||||
|
||||
typedef AntiFactor<VALUES> This;
|
||||
typedef NonlinearFactor<VALUES> Base;
|
||||
typedef typename NonlinearFactor<VALUES>::shared_ptr sharedFactor;
|
||||
|
||||
sharedFactor factor_;
|
||||
|
||||
public:
|
||||
|
||||
// shorthand for a smart pointer to a factor
|
||||
typedef typename boost::shared_ptr<AntiFactor> shared_ptr;
|
||||
|
||||
/** default constructor - only use for serialization */
|
||||
AntiFactor() {}
|
||||
|
||||
/** constructor - Creates the equivalent AntiFactor from an existing factor */
|
||||
AntiFactor(typename NonlinearFactor<VALUES>::shared_ptr factor) : Base(factor->begin(), factor->end()), factor_(factor) {}
|
||||
|
||||
virtual ~AntiFactor() {}
|
||||
|
||||
/** implement functions needed for Testable */
|
||||
|
||||
/** print */
|
||||
virtual void print(const std::string& s) const {
|
||||
std::cout << s << "AntiFactor version of:" << std::endl;
|
||||
factor_->print(s);
|
||||
}
|
||||
|
||||
/** equals */
|
||||
virtual bool equals(const NonlinearFactor<VALUES>& expected, double tol=1e-9) const {
|
||||
const This *e = dynamic_cast<const This*> (&expected);
|
||||
return e != NULL && Base::equals(*e, tol) && this->factor_->equals(*e->factor_, tol);
|
||||
}
|
||||
|
||||
/** implement functions needed to derive from Factor */
|
||||
|
||||
/**
|
||||
* Calculate the error of the factor
|
||||
* This is typically equal to log-likelihood, e.g. 0.5(h(x)-z)^2/sigma^2 in case of Gaussian.
|
||||
* You can override this for systems with unusual noise models.
|
||||
*/
|
||||
double error(const VALUES& c) const { return -factor_->error(c); }
|
||||
|
||||
/** get the dimension of the factor (number of rows on linearization) */
|
||||
size_t dim() const { return factor_->dim(); }
|
||||
|
||||
/**
|
||||
* Checks whether a factor should be used based on a set of values.
|
||||
* This is primarily used to implment inequality constraints that
|
||||
* require a variable active set. For all others, the default implementation
|
||||
* returning true solves this problem.
|
||||
*
|
||||
* In an inequality/bounding constraint, this active() returns true
|
||||
* when the constraint is *NOT* fulfilled.
|
||||
* @return true if the constraint is active
|
||||
*/
|
||||
bool active(const VALUES& c) const { return factor_->active(c); }
|
||||
|
||||
/** linearize to a GaussianFactor */
|
||||
boost::shared_ptr<GaussianFactor>
|
||||
linearize(const VALUES& c, const Ordering& ordering) const {
|
||||
|
||||
// Generate the linearized factor from the contained nonlinear factor
|
||||
GaussianFactor::shared_ptr gaussianFactor = factor_->linearize(c, ordering);
|
||||
|
||||
// Cast the GaussianFactor to a Hessian
|
||||
HessianFactor::shared_ptr hessianFactor = boost::dynamic_pointer_cast<HessianFactor>(gaussianFactor);
|
||||
|
||||
// If the cast fails, convert it to a Hessian
|
||||
if(!hessianFactor){
|
||||
hessianFactor = HessianFactor::shared_ptr(new HessianFactor(*gaussianFactor));
|
||||
}
|
||||
|
||||
// Copy Hessian Blocks from Hessian factor and invert
|
||||
std::vector<Index> js;
|
||||
std::vector<Matrix> Gs;
|
||||
std::vector<Vector> gs;
|
||||
double f;
|
||||
js.insert(js.end(), hessianFactor->begin(), hessianFactor->end());
|
||||
for(size_t i = 0; i < js.size(); ++i){
|
||||
for(size_t j = i; j < js.size(); ++j){
|
||||
Gs.push_back( -hessianFactor->info(hessianFactor->begin()+i, hessianFactor->begin()+j) );
|
||||
}
|
||||
gs.push_back( -hessianFactor->linearTerm(hessianFactor->begin()+i) );
|
||||
}
|
||||
f = -hessianFactor->constantTerm();
|
||||
|
||||
// Create the Anti-Hessian Factor from the negated blocks
|
||||
return HessianFactor::shared_ptr(new HessianFactor(js, Gs, gs, f));
|
||||
}
|
||||
|
||||
|
||||
private:
|
||||
|
||||
/** Serialization function */
|
||||
friend class boost::serialization::access;
|
||||
template<class ARCHIVE>
|
||||
void serialize(ARCHIVE & ar, const unsigned int version) {
|
||||
ar & boost::serialization::make_nvp("AntiFactor",
|
||||
boost::serialization::base_object<Base>(*this));
|
||||
ar & BOOST_SERIALIZATION_NVP(factor_);
|
||||
}
|
||||
}; // \class AntiFactor
|
||||
|
||||
} /// namespace gtsam
|
||||
|
|
@ -26,6 +26,8 @@ check_PROGRAMS += tests/testSimulated3D
|
|||
# Generic SLAM headers
|
||||
headers += BetweenFactor.h PriorFactor.h PartialPriorFactor.h
|
||||
headers += BearingFactor.h RangeFactor.h BearingRangeFactor.h
|
||||
headers += AntiFactor.h
|
||||
check_PROGRAMS += tests/testAntiFactor
|
||||
|
||||
# Generic constraint headers
|
||||
headers += BoundingConstraint.h
|
||||
|
|
|
|||
|
|
@ -0,0 +1,139 @@
|
|||
/* ----------------------------------------------------------------------------
|
||||
|
||||
* 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 testAntiFactor.cpp
|
||||
* @brief Unit test for the AntiFactor
|
||||
* @author Stephen Williams
|
||||
*/
|
||||
|
||||
#include <CppUnitLite/TestHarness.h>
|
||||
|
||||
#include <gtsam/slam/pose3SLAM.h>
|
||||
#include <gtsam/slam/AntiFactor.h>
|
||||
#include <gtsam/nonlinear/NonlinearFactorGraph.h>
|
||||
#include <gtsam/nonlinear/NonlinearOptimizer.h>
|
||||
#include <gtsam/linear/GaussianSequentialSolver.h>
|
||||
#include <gtsam/geometry/Pose3.h>
|
||||
|
||||
using namespace std;
|
||||
using namespace gtsam;
|
||||
|
||||
/* ************************************************************************* */
|
||||
TEST( AntiFactor, NegativeHessian)
|
||||
{
|
||||
// The AntiFactor should produce a Hessian Factor with negative matrices
|
||||
|
||||
// Create linearization points
|
||||
Pose3 pose1(Rot3(), Point3(0, 0, 0));
|
||||
Pose3 pose2(Rot3(), Point3(2, 1, 3));
|
||||
Pose3 z(Rot3(), Point3(1, 1, 1));
|
||||
SharedNoiseModel sigma(noiseModel::Unit::Create(Pose3::Dim()));
|
||||
|
||||
// Create a configuration corresponding to the ground truth
|
||||
boost::shared_ptr<pose3SLAM::Values> values(new pose3SLAM::Values());
|
||||
values->insert(1, pose1);
|
||||
values->insert(2, pose2);
|
||||
|
||||
// Define an elimination ordering
|
||||
Ordering::shared_ptr ordering(new Ordering());
|
||||
ordering->insert(pose3SLAM::Key(1), 0);
|
||||
ordering->insert(pose3SLAM::Key(2), 1);
|
||||
|
||||
|
||||
// Create a "standard" factor
|
||||
BetweenFactor<pose3SLAM::Values,pose3SLAM::Key>::shared_ptr originalFactor(new BetweenFactor<pose3SLAM::Values,pose3SLAM::Key>(1, 2, z, sigma));
|
||||
|
||||
// Linearize it into a Jacobian Factor
|
||||
GaussianFactor::shared_ptr originalJacobian = originalFactor->linearize(*values, *ordering);
|
||||
|
||||
// Convert it to a Hessian Factor
|
||||
HessianFactor::shared_ptr originalHessian = HessianFactor::shared_ptr(new HessianFactor(*originalJacobian));
|
||||
|
||||
// Create the AntiFactor version of the original nonlinear factor
|
||||
AntiFactor<pose3SLAM::Values>::shared_ptr antiFactor(new AntiFactor<pose3SLAM::Values>(originalFactor));
|
||||
|
||||
// Linearize the AntiFactor into a Hessian Factor
|
||||
GaussianFactor::shared_ptr antiGaussian = antiFactor->linearize(*values, *ordering);
|
||||
HessianFactor::shared_ptr antiHessian = boost::dynamic_pointer_cast<HessianFactor>(antiGaussian);
|
||||
|
||||
|
||||
// Compare Hessian blocks
|
||||
size_t variable_count = originalFactor->size();
|
||||
for(size_t i = 0; i < variable_count; ++i){
|
||||
for(size_t j = i; j < variable_count; ++j){
|
||||
Matrix expected_G = -originalHessian->info(originalHessian->begin()+i, originalHessian->begin()+j);
|
||||
Matrix actual_G = antiHessian->info(antiHessian->begin()+i, antiHessian->begin()+j);
|
||||
CHECK(assert_equal(expected_G, actual_G, 1e-5));
|
||||
}
|
||||
Vector expected_g = -originalHessian->linearTerm(originalHessian->begin()+i);
|
||||
Vector actual_g = antiHessian->linearTerm(antiHessian->begin()+i);
|
||||
CHECK(assert_equal(expected_g, actual_g, 1e-5));
|
||||
}
|
||||
double expected_f = -originalHessian->constantTerm();
|
||||
double actual_f = antiHessian->constantTerm();
|
||||
EXPECT_DOUBLES_EQUAL(expected_f, actual_f, 1e-5);
|
||||
}
|
||||
|
||||
/* ************************************************************************* */
|
||||
TEST( AntiFactor, EquivalentBayesNet)
|
||||
{
|
||||
// Test the AntiFactor by creating a simple graph and eliminating into a BayesNet
|
||||
// Then add an additional factor and the corresponding AntiFactor and eliminate
|
||||
// The resulting BayesNet should be identical to the first
|
||||
|
||||
Pose3 pose1(Rot3(), Point3(0, 0, 0));
|
||||
Pose3 pose2(Rot3(), Point3(2, 1, 3));
|
||||
Pose3 z(Rot3(), Point3(1, 1, 1));
|
||||
SharedNoiseModel sigma(noiseModel::Unit::Create(Pose3::Dim()));
|
||||
|
||||
boost::shared_ptr<pose3SLAM::Graph> graph(new pose3SLAM::Graph());
|
||||
graph->addPrior(1, pose1, sigma);
|
||||
graph->addConstraint(1, 2, pose1.between(pose2), sigma);
|
||||
|
||||
// Create a configuration corresponding to the ground truth
|
||||
boost::shared_ptr<pose3SLAM::Values> values(new pose3SLAM::Values());
|
||||
values->insert(1, pose1);
|
||||
values->insert(2, pose2);
|
||||
|
||||
// Define an elimination ordering
|
||||
Ordering::shared_ptr ordering = graph->orderingCOLAMD(*values);
|
||||
|
||||
// Eliminate into a BayesNet
|
||||
GaussianSequentialSolver solver1(*graph->linearize(*values, *ordering));
|
||||
GaussianBayesNet::shared_ptr expectedBayesNet = solver1.eliminate();
|
||||
|
||||
// Back-substitute to find the optimal deltas
|
||||
VectorValues expectedDeltas = optimize(*expectedBayesNet);
|
||||
|
||||
// Add an additional factor between Pose1 and Pose2
|
||||
pose3SLAM::Constraint::shared_ptr f1(new pose3SLAM::Constraint(1, 2, z, sigma));
|
||||
graph->push_back(f1);
|
||||
|
||||
// Add the corresponding AntiFactor between Pose1 and Pose2
|
||||
AntiFactor<pose3SLAM::Values>::shared_ptr f2(new AntiFactor<pose3SLAM::Values>(f1));
|
||||
graph->push_back(f2);
|
||||
|
||||
// Again, Eliminate into a BayesNet
|
||||
GaussianSequentialSolver solver2(*graph->linearize(*values, *ordering));
|
||||
GaussianBayesNet::shared_ptr actualBayesNet = solver2.eliminate();
|
||||
|
||||
// Back-substitute to find the optimal deltas
|
||||
VectorValues actualDeltas = optimize(*actualBayesNet);
|
||||
|
||||
// Verify the BayesNets are identical
|
||||
CHECK(assert_equal(*expectedBayesNet, *actualBayesNet, 1e-5));
|
||||
CHECK(assert_equal(expectedDeltas, actualDeltas, 1e-5));
|
||||
}
|
||||
|
||||
/* ************************************************************************* */
|
||||
int main() { TestResult tr; return TestRegistry::runAllTests(tr);}
|
||||
/* ************************************************************************* */
|
||||
Loading…
Reference in New Issue