| Line |
Branch |
Exec |
Source |
| 1 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
| 2 |
|
|
// BSD 3-Clause License |
| 3 |
|
|
// |
| 4 |
|
|
// Copyright (C) 2021-2025, LAAS-CNRS, New York University, |
| 5 |
|
|
// Max Planck Gesellschaft, |
| 6 |
|
|
// University of Edinburgh, INRIA, |
| 7 |
|
|
// University of Trento, Heriot-Watt University |
| 8 |
|
|
// Copyright note valid unless otherwise controld in individual files. |
| 9 |
|
|
// All rights reserved. |
| 10 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
| 11 |
|
|
|
| 12 |
|
|
#define BOOST_TEST_NO_MAIN |
| 13 |
|
|
#define BOOST_TEST_ALTERNATIVE_INIT_API |
| 14 |
|
|
|
| 15 |
|
|
#include "crocoddyl/core/numdiff/control.hpp" |
| 16 |
|
|
#include "factory/control.hpp" |
| 17 |
|
|
#include "unittest_common.hpp" |
| 18 |
|
|
|
| 19 |
|
|
using namespace boost::unit_test; |
| 20 |
|
|
using namespace crocoddyl::unittest; |
| 21 |
|
|
|
| 22 |
|
|
//----------------------------------------------------------------------------// |
| 23 |
|
|
|
| 24 |
|
✗ |
void test_calcDiff_num_diff(ControlTypes::Type control_type) { |
| 25 |
|
✗ |
ControlFactory factory; |
| 26 |
|
|
const std::shared_ptr<crocoddyl::ControlParametrizationModelAbstract>& |
| 27 |
|
✗ |
control = factory.create(control_type, 10); |
| 28 |
|
|
|
| 29 |
|
|
const std::shared_ptr<crocoddyl::ControlParametrizationDataAbstract>& data = |
| 30 |
|
✗ |
control->createData(); |
| 31 |
|
|
|
| 32 |
|
|
// Generating random values for the control parameters |
| 33 |
|
✗ |
const Eigen::VectorXd p = Eigen::VectorXd::Random(control->get_nu()); |
| 34 |
|
✗ |
double t = Eigen::VectorXd::Random(1)(0) * 0.5 + 1.; // random in [0, 1] |
| 35 |
|
|
|
| 36 |
|
|
// Get the num diff control |
| 37 |
|
✗ |
crocoddyl::ControlParametrizationModelNumDiff control_num_diff(control); |
| 38 |
|
|
std::shared_ptr<crocoddyl::ControlParametrizationDataAbstract> data_num_diff = |
| 39 |
|
✗ |
control_num_diff.createData(); |
| 40 |
|
|
|
| 41 |
|
|
// Computing the partial derivatives of the value function |
| 42 |
|
✗ |
control->calc(data, t, p); |
| 43 |
|
✗ |
control_num_diff.calc(data_num_diff, t, p); |
| 44 |
|
✗ |
control->calcDiff(data, t, p); |
| 45 |
|
✗ |
control_num_diff.calcDiff(data_num_diff, t, p); |
| 46 |
|
|
// Tolerance defined as in |
| 47 |
|
|
// http://www.it.uom.gr/teaching/linearalgebra/NumericalRecipiesInC/c5-7.pdf |
| 48 |
|
✗ |
double tol = std::pow(control_num_diff.get_disturbance(), 1. / 3.); |
| 49 |
|
✗ |
BOOST_CHECK((data->dw_du - data_num_diff->dw_du).isZero(tol)); |
| 50 |
|
|
|
| 51 |
|
|
// Checking that casted computation is the same |
| 52 |
|
|
const std::shared_ptr< |
| 53 |
|
|
crocoddyl::ControlParametrizationModelAbstractTpl<float>>& |
| 54 |
|
✗ |
casted_control = control->cast<float>(); |
| 55 |
|
|
const std::shared_ptr< |
| 56 |
|
|
crocoddyl::ControlParametrizationDataAbstractTpl<float>>& casted_data = |
| 57 |
|
✗ |
casted_control->createData(); |
| 58 |
|
✗ |
const Eigen::VectorXf p_f = p.cast<float>(); |
| 59 |
|
✗ |
float t_f = float(t); |
| 60 |
|
✗ |
casted_control->calc(casted_data, t_f, p_f); |
| 61 |
|
✗ |
casted_control->calcDiff(casted_data, t_f, p_f); |
| 62 |
|
✗ |
float tol_f = std::sqrt(2.0f * std::numeric_limits<float>::epsilon()); |
| 63 |
|
✗ |
BOOST_CHECK((data->dw_du.cast<float>() - casted_data->dw_du).isZero(tol_f)); |
| 64 |
|
|
} |
| 65 |
|
|
|
| 66 |
|
✗ |
void test_multiplyByJacobian_num_diff(ControlTypes::Type control_type) { |
| 67 |
|
✗ |
ControlFactory factory; |
| 68 |
|
|
const std::shared_ptr<crocoddyl::ControlParametrizationModelAbstract>& |
| 69 |
|
✗ |
control = factory.create(control_type, 10); |
| 70 |
|
|
|
| 71 |
|
|
std::shared_ptr<crocoddyl::ControlParametrizationDataAbstract> data = |
| 72 |
|
✗ |
control->createData(); |
| 73 |
|
|
|
| 74 |
|
|
// Generating random values for the control parameters, the time, and the |
| 75 |
|
|
// matrix to multiply |
| 76 |
|
✗ |
const Eigen::VectorXd p = Eigen::VectorXd::Random(control->get_nu()); |
| 77 |
|
✗ |
double t = Eigen::VectorXd::Random(1)(0) * 0.5 + 1.; // random in [0, 1] |
| 78 |
|
✗ |
const Eigen::MatrixXd A = Eigen::MatrixXd::Random(5, control->get_nw()); |
| 79 |
|
|
|
| 80 |
|
|
// Get the num diff control and datas |
| 81 |
|
✗ |
crocoddyl::ControlParametrizationModelNumDiff control_num_diff(control); |
| 82 |
|
|
std::shared_ptr<crocoddyl::ControlParametrizationDataAbstract> data_num_diff = |
| 83 |
|
✗ |
control_num_diff.createData(); |
| 84 |
|
|
|
| 85 |
|
|
// Checking the operator |
| 86 |
|
✗ |
Eigen::MatrixXd A_J(Eigen::MatrixXd::Zero(A.rows(), control->get_nu())); |
| 87 |
|
|
Eigen::MatrixXd A_J_num_diff( |
| 88 |
|
✗ |
Eigen::MatrixXd::Zero(A.rows(), control->get_nu())); |
| 89 |
|
✗ |
control->calc(data, t, p); |
| 90 |
|
✗ |
control->calcDiff(data, t, p); |
| 91 |
|
✗ |
control_num_diff.calc(data_num_diff, t, p); |
| 92 |
|
✗ |
control_num_diff.calcDiff(data_num_diff, t, p); |
| 93 |
|
✗ |
control->multiplyByJacobian(data, A, A_J); |
| 94 |
|
✗ |
control_num_diff.multiplyByJacobian(data_num_diff, A, A_J_num_diff); |
| 95 |
|
|
// Tolerance defined as in |
| 96 |
|
|
// http://www.it.uom.gr/teaching/linearalgebra/NumericalRecipiesInC/c5-7.pdf |
| 97 |
|
✗ |
double tol = std::pow(control_num_diff.get_disturbance(), 1. / 3.); |
| 98 |
|
✗ |
BOOST_CHECK((A_J - A_J_num_diff).isZero(tol)); |
| 99 |
|
|
|
| 100 |
|
|
// Checking that casted computation is the same |
| 101 |
|
|
const std::shared_ptr< |
| 102 |
|
|
crocoddyl::ControlParametrizationModelAbstractTpl<float>>& |
| 103 |
|
✗ |
casted_control = control->cast<float>(); |
| 104 |
|
|
const std::shared_ptr< |
| 105 |
|
|
crocoddyl::ControlParametrizationDataAbstractTpl<float>>& casted_data = |
| 106 |
|
✗ |
casted_control->createData(); |
| 107 |
|
|
Eigen::MatrixXf A_J_f( |
| 108 |
|
✗ |
Eigen::MatrixXf::Zero(A.rows(), casted_control->get_nu())); |
| 109 |
|
✗ |
const Eigen::VectorXf p_f = p.cast<float>(); |
| 110 |
|
✗ |
float t_f = float(t); |
| 111 |
|
✗ |
const Eigen::MatrixXf A_f = A.cast<float>(); |
| 112 |
|
✗ |
casted_control->calc(casted_data, t_f, p_f); |
| 113 |
|
✗ |
casted_control->calcDiff(casted_data, t_f, p_f); |
| 114 |
|
✗ |
casted_control->multiplyByJacobian(casted_data, A_f, A_J_f); |
| 115 |
|
✗ |
float tol_f = std::sqrt(2.0f * std::numeric_limits<float>::epsilon()); |
| 116 |
|
✗ |
BOOST_CHECK((data->dw_du.cast<float>() - casted_data->dw_du).isZero(tol_f)); |
| 117 |
|
✗ |
BOOST_CHECK((A_J.cast<float>() - A_J_f).isZero(tol_f)); |
| 118 |
|
|
} |
| 119 |
|
|
|
| 120 |
|
✗ |
void test_multiplyJacobianTransposeBy_num_diff( |
| 121 |
|
|
ControlTypes::Type control_type) { |
| 122 |
|
✗ |
ControlFactory factory; |
| 123 |
|
|
const std::shared_ptr<crocoddyl::ControlParametrizationModelAbstract>& |
| 124 |
|
✗ |
control = factory.create(control_type, 10); |
| 125 |
|
|
|
| 126 |
|
|
std::shared_ptr<crocoddyl::ControlParametrizationDataAbstract> data = |
| 127 |
|
✗ |
control->createData(); |
| 128 |
|
|
|
| 129 |
|
|
// Generating random values for the control parameters, the time, and the |
| 130 |
|
|
// matrix to multiply |
| 131 |
|
✗ |
const Eigen::VectorXd p = Eigen::VectorXd::Random(control->get_nu()); |
| 132 |
|
✗ |
double t = Eigen::VectorXd::Random(1)(0) * 0.5 + 1.; // random in [0, 1] |
| 133 |
|
✗ |
const Eigen::MatrixXd A = Eigen::MatrixXd::Random(control->get_nw(), 5); |
| 134 |
|
|
|
| 135 |
|
|
// Get the num diff control and datas |
| 136 |
|
✗ |
crocoddyl::ControlParametrizationModelNumDiff control_num_diff(control); |
| 137 |
|
|
std::shared_ptr<crocoddyl::ControlParametrizationDataAbstract> data_num_diff = |
| 138 |
|
✗ |
control_num_diff.createData(); |
| 139 |
|
|
|
| 140 |
|
|
// Checking the operator |
| 141 |
|
✗ |
Eigen::MatrixXd JT_A(Eigen::MatrixXd::Zero(control->get_nu(), A.cols())); |
| 142 |
|
|
Eigen::MatrixXd JT_A_num_diff( |
| 143 |
|
✗ |
Eigen::MatrixXd::Zero(control->get_nu(), A.cols())); |
| 144 |
|
✗ |
control->calc(data, t, p); |
| 145 |
|
✗ |
control->calcDiff(data, t, p); |
| 146 |
|
✗ |
control_num_diff.calc(data_num_diff, t, p); |
| 147 |
|
✗ |
control_num_diff.calcDiff(data_num_diff, t, p); |
| 148 |
|
✗ |
control->multiplyJacobianTransposeBy(data, A, JT_A); |
| 149 |
|
✗ |
control_num_diff.multiplyJacobianTransposeBy(data_num_diff, A, JT_A_num_diff); |
| 150 |
|
|
// Tolerance defined as in |
| 151 |
|
|
// http://www.it.uom.gr/teaching/linearalgebra/NumericalRecipiesInC/c5-7.pdf |
| 152 |
|
✗ |
double tol = std::pow(control_num_diff.get_disturbance(), 1. / 3.); |
| 153 |
|
✗ |
BOOST_CHECK((JT_A - JT_A_num_diff).isZero(tol)); |
| 154 |
|
|
|
| 155 |
|
|
// Checking that casted computation is the same |
| 156 |
|
|
const std::shared_ptr< |
| 157 |
|
|
crocoddyl::ControlParametrizationModelAbstractTpl<float>>& |
| 158 |
|
✗ |
casted_control = control->cast<float>(); |
| 159 |
|
|
const std::shared_ptr< |
| 160 |
|
|
crocoddyl::ControlParametrizationDataAbstractTpl<float>>& casted_data = |
| 161 |
|
✗ |
casted_control->createData(); |
| 162 |
|
✗ |
const Eigen::VectorXf p_f = p.cast<float>(); |
| 163 |
|
✗ |
float t_f = float(t); |
| 164 |
|
✗ |
const Eigen::MatrixXf A_f = A.cast<float>(); |
| 165 |
|
|
Eigen::MatrixXf JT_A_f( |
| 166 |
|
✗ |
Eigen::MatrixXf::Zero(casted_control->get_nu(), A_f.cols())); |
| 167 |
|
✗ |
casted_control->calc(casted_data, t_f, p_f); |
| 168 |
|
✗ |
casted_control->calcDiff(casted_data, t_f, p_f); |
| 169 |
|
✗ |
casted_control->multiplyJacobianTransposeBy(casted_data, A_f, JT_A_f); |
| 170 |
|
✗ |
float tol_f = std::sqrt(2.0f * std::numeric_limits<float>::epsilon()); |
| 171 |
|
✗ |
BOOST_CHECK((JT_A.cast<float>() - JT_A_f).isZero(tol_f)); |
| 172 |
|
|
} |
| 173 |
|
|
|
| 174 |
|
|
//----------------------------------------------------------------------------// |
| 175 |
|
|
|
| 176 |
|
✗ |
void register_control_unit_tests(ControlTypes::Type control_type) { |
| 177 |
|
✗ |
boost::test_tools::output_test_stream test_name; |
| 178 |
|
✗ |
test_name << "test_" << control_type; |
| 179 |
|
✗ |
std::cout << "Running " << test_name.str() << std::endl; |
| 180 |
|
✗ |
test_suite* ts = BOOST_TEST_SUITE(test_name.str()); |
| 181 |
|
✗ |
ts->add(BOOST_TEST_CASE(boost::bind(&test_calcDiff_num_diff, control_type))); |
| 182 |
|
✗ |
ts->add(BOOST_TEST_CASE( |
| 183 |
|
|
boost::bind(&test_multiplyByJacobian_num_diff, control_type))); |
| 184 |
|
✗ |
ts->add(BOOST_TEST_CASE( |
| 185 |
|
|
boost::bind(&test_multiplyJacobianTransposeBy_num_diff, control_type))); |
| 186 |
|
✗ |
framework::master_test_suite().add(ts); |
| 187 |
|
|
} |
| 188 |
|
|
|
| 189 |
|
✗ |
bool init_function() { |
| 190 |
|
✗ |
for (size_t i = 0; i < ControlTypes::all.size(); ++i) { |
| 191 |
|
✗ |
register_control_unit_tests(ControlTypes::all[i]); |
| 192 |
|
|
} |
| 193 |
|
✗ |
return true; |
| 194 |
|
|
} |
| 195 |
|
|
|
| 196 |
|
✗ |
int main(int argc, char** argv) { |
| 197 |
|
✗ |
return ::boost::unit_test::unit_test_main(&init_function, argc, argv); |
| 198 |
|
|
} |
| 199 |
|
|
|