130 lines
		
	
	
		
			3.9 KiB
		
	
	
	
		
			C++
		
	
	
			
		
		
	
	
			130 lines
		
	
	
		
			3.9 KiB
		
	
	
	
		
			C++
		
	
	
/* ----------------------------------------------------------------------------
 | 
						|
 | 
						|
 * 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    testGaussMarkov1stOrderFactor.cpp
 | 
						|
 * @brief   Unit tests for the GaussMarkov1stOrder factor
 | 
						|
 * @author  Vadim Indelman
 | 
						|
 * @date    Jan 17, 2012
 | 
						|
 */
 | 
						|
 | 
						|
#include <CppUnitLite/TestHarness.h>
 | 
						|
#include <gtsam/base/Vector.h>
 | 
						|
#include <gtsam/base/numericalDerivative.h>
 | 
						|
#include <gtsam/inference/Key.h>
 | 
						|
#include <gtsam/nonlinear/Values.h>
 | 
						|
#include <gtsam_unstable/slam/GaussMarkov1stOrderFactor.h>
 | 
						|
 | 
						|
using namespace std::placeholders;
 | 
						|
using namespace std;
 | 
						|
using namespace gtsam;
 | 
						|
 | 
						|
//! Factors
 | 
						|
typedef GaussMarkov1stOrderFactor<Vector3> GaussMarkovFactor;
 | 
						|
 | 
						|
/* ************************************************************************* */
 | 
						|
Vector predictionError(const Vector& v1, const Vector& v2,
 | 
						|
                       const GaussMarkovFactor factor) {
 | 
						|
  return factor.evaluateError(v1, v2);
 | 
						|
}
 | 
						|
 | 
						|
/* ************************************************************************* */
 | 
						|
TEST( GaussMarkovFactor, equals )
 | 
						|
{
 | 
						|
  // Create two identical factors and make sure they're equal
 | 
						|
  Key x1(1);
 | 
						|
  Key x2(2);
 | 
						|
  double delta_t = 0.10;
 | 
						|
  Vector tau = Vector3(100.0, 150.0, 10.0);
 | 
						|
  SharedGaussian model = noiseModel::Isotropic::Sigma(3, 1.0);
 | 
						|
 | 
						|
  GaussMarkovFactor factor1(x1, x2, delta_t, tau, model);
 | 
						|
  GaussMarkovFactor factor2(x1, x2, delta_t, tau, model);
 | 
						|
 | 
						|
  CHECK(assert_equal(factor1, factor2));
 | 
						|
}
 | 
						|
 | 
						|
/* ************************************************************************* */
 | 
						|
TEST( GaussMarkovFactor, error )
 | 
						|
{
 | 
						|
  Values linPoint;
 | 
						|
  Key x1(1);
 | 
						|
  Key x2(2);
 | 
						|
  double delta_t = 0.10;
 | 
						|
  Vector3 tau(100.0, 150.0, 10.0);
 | 
						|
  SharedGaussian model = noiseModel::Isotropic::Sigma(3, 1.0);
 | 
						|
 | 
						|
  Vector3 v1(10.0, 12.0, 13.0);
 | 
						|
  Vector3 v2(10.0, 15.0, 14.0);
 | 
						|
 | 
						|
  // Create two nodes
 | 
						|
  linPoint.insert(x1, v1);
 | 
						|
  linPoint.insert(x2, v2);
 | 
						|
 | 
						|
  GaussMarkovFactor factor(x1, x2, delta_t, tau, model);
 | 
						|
  Vector3 error1 = factor.evaluateError(v1, v2);
 | 
						|
 | 
						|
  // Manually calculate the error
 | 
						|
  Vector3 alpha(tau.size());
 | 
						|
  Vector3 alpha_v1(tau.size());
 | 
						|
  for(int i=0; i<tau.size(); i++){
 | 
						|
    alpha(i) = exp(- 1/tau(i)*delta_t );
 | 
						|
    alpha_v1(i) = alpha(i) * v1(i);
 | 
						|
  }
 | 
						|
  Vector3 error2 = v2 - alpha_v1;
 | 
						|
 | 
						|
  CHECK(assert_equal(error1, error2, 1e-8));
 | 
						|
}
 | 
						|
 | 
						|
/* ************************************************************************* */
 | 
						|
TEST (GaussMarkovFactor, jacobian ) {
 | 
						|
 | 
						|
  Values linPoint;
 | 
						|
  Key x1(1);
 | 
						|
  Key x2(2);
 | 
						|
  double delta_t = 0.10;
 | 
						|
  Vector3 tau(100.0, 150.0, 10.0);
 | 
						|
  SharedGaussian model = noiseModel::Isotropic::Sigma(3, 1.0);
 | 
						|
 | 
						|
  GaussMarkovFactor factor(x1, x2, delta_t, tau, model);
 | 
						|
 | 
						|
  // Update the linearization point
 | 
						|
  Vector3 v1_upd(0.5, -0.7, 0.3);
 | 
						|
  Vector3 v2_upd(-0.7, 0.4, 0.9);
 | 
						|
 | 
						|
  // Calculate the Jacobian matrix using the factor
 | 
						|
  Matrix computed_H1, computed_H2;
 | 
						|
  factor.evaluateError(v1_upd, v2_upd, computed_H1, computed_H2);
 | 
						|
 | 
						|
  // Calculate the Jacobian matrices H1 and H2 using the numerical derivative function
 | 
						|
  Matrix numerical_H1, numerical_H2;
 | 
						|
  numerical_H1 = numericalDerivative21<Vector3, Vector3, Vector3>(
 | 
						|
      std::bind(&predictionError, std::placeholders::_1, std::placeholders::_2,
 | 
						|
                factor),
 | 
						|
      v1_upd, v2_upd);
 | 
						|
  numerical_H2 = numericalDerivative22<Vector3, Vector3, Vector3>(
 | 
						|
      std::bind(&predictionError, std::placeholders::_1, std::placeholders::_2,
 | 
						|
                factor),
 | 
						|
      v1_upd, v2_upd);
 | 
						|
 | 
						|
  // Verify they are equal for this choice of state
 | 
						|
  CHECK(assert_equal(numerical_H1, computed_H1, 1e-9));
 | 
						|
  CHECK(assert_equal(numerical_H2, computed_H2, 1e-9));
 | 
						|
}
 | 
						|
 | 
						|
/* ************************************************************************* */
 | 
						|
int main()
 | 
						|
{
 | 
						|
  TestResult tr; return TestRegistry::runAllTests(tr);
 | 
						|
}
 | 
						|
/* ************************************************************************* */
 | 
						|
 |