| Line |
Branch |
Exec |
Source |
| 1 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
| 2 |
|
|
// BSD 3-Clause License |
| 3 |
|
|
// |
| 4 |
|
|
// Copyright (C) 2019-2025, LAAS-CNRS, University of Edinburgh, |
| 5 |
|
|
// Heriot-Watt University |
| 6 |
|
|
// Copyright note valid unless otherwise stated in individual files. |
| 7 |
|
|
// All rights reserved. |
| 8 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
| 9 |
|
|
|
| 10 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 11 |
|
|
#include <omp.h> |
| 12 |
|
|
#endif |
| 13 |
|
|
#ifndef CROCODDYL_WITH_NTHREADS |
| 14 |
|
|
#define CROCODDYL_WITH_NTHREADS 1 |
| 15 |
|
|
#endif |
| 16 |
|
|
|
| 17 |
|
|
#ifdef CROCODDYL_WITH_CODEGEN |
| 18 |
|
|
#include "crocoddyl/core/codegen/action-base.hpp" |
| 19 |
|
|
#endif |
| 20 |
|
|
|
| 21 |
|
|
#include "crocoddyl/core/solvers/fddp.hpp" |
| 22 |
|
|
#include "crocoddyl/core/utils/file-io.hpp" |
| 23 |
|
|
#include "crocoddyl/core/utils/timer.hpp" |
| 24 |
|
|
#include "factory/arm-kinova.hpp" |
| 25 |
|
|
#include "factory/arm.hpp" |
| 26 |
|
|
#include "factory/legged-robots.hpp" |
| 27 |
|
|
|
| 28 |
|
|
#define STDDEV(vec) \ |
| 29 |
|
|
std::sqrt(((vec - vec.mean())).square().sum() / \ |
| 30 |
|
|
(static_cast<double>(vec.size()) - 1)) |
| 31 |
|
|
#define AVG(vec) (vec.mean()) |
| 32 |
|
|
|
| 33 |
|
✗ |
void print_benchmark(RobotEENames robot) { |
| 34 |
|
✗ |
unsigned int N = 100; // number of nodes |
| 35 |
|
✗ |
unsigned int T = 1e3; // number of trials |
| 36 |
|
|
|
| 37 |
|
|
// Building the running and terminal models |
| 38 |
|
✗ |
std::shared_ptr<crocoddyl::ActionModelAbstract> runningModel, terminalModel; |
| 39 |
|
✗ |
if (robot.robot_name == "Talos_arm") { |
| 40 |
|
✗ |
crocoddyl::benchmark::build_arm_action_models(runningModel, terminalModel); |
| 41 |
|
✗ |
} else if (robot.robot_name == "Kinova_arm") { |
| 42 |
|
✗ |
crocoddyl::benchmark::build_arm_kinova_action_models(runningModel, |
| 43 |
|
|
terminalModel); |
| 44 |
|
|
} else { |
| 45 |
|
✗ |
crocoddyl::benchmark::build_contact_action_models(robot, runningModel, |
| 46 |
|
|
terminalModel); |
| 47 |
|
|
} |
| 48 |
|
|
|
| 49 |
|
|
// Get the initial state |
| 50 |
|
|
std::shared_ptr<crocoddyl::StateMultibody> state = |
| 51 |
|
|
std::static_pointer_cast<crocoddyl::StateMultibody>( |
| 52 |
|
✗ |
runningModel->get_state()); |
| 53 |
|
✗ |
std::cout << "NQ: " << state->get_nq() << std::endl; |
| 54 |
|
✗ |
std::cout << "Number of nodes: " << N << std::endl << std::endl; |
| 55 |
|
|
|
| 56 |
|
✗ |
Eigen::VectorXd default_state(state->get_nq() + state->get_nv()); |
| 57 |
|
|
std::shared_ptr<crocoddyl::IntegratedActionModelEuler> rm = |
| 58 |
|
|
std::static_pointer_cast<crocoddyl::IntegratedActionModelEuler>( |
| 59 |
|
✗ |
runningModel); |
| 60 |
|
✗ |
if (robot.robot_name == "Talos_arm" || robot.robot_name == "Kinova_arm") { |
| 61 |
|
|
std::shared_ptr<crocoddyl::DifferentialActionModelFreeFwdDynamics> dm = |
| 62 |
|
|
std::static_pointer_cast< |
| 63 |
|
|
crocoddyl::DifferentialActionModelFreeFwdDynamics>( |
| 64 |
|
✗ |
rm->get_differential()); |
| 65 |
|
|
default_state |
| 66 |
|
✗ |
<< dm->get_pinocchio().referenceConfigurations[robot.reference_conf], |
| 67 |
|
✗ |
Eigen::VectorXd::Zero(state->get_nv()); |
| 68 |
|
✗ |
} else { |
| 69 |
|
|
std::shared_ptr<crocoddyl::DifferentialActionModelContactFwdDynamics> dm = |
| 70 |
|
|
std::static_pointer_cast< |
| 71 |
|
|
crocoddyl::DifferentialActionModelContactFwdDynamics>( |
| 72 |
|
✗ |
rm->get_differential()); |
| 73 |
|
|
default_state |
| 74 |
|
✗ |
<< dm->get_pinocchio().referenceConfigurations[robot.reference_conf], |
| 75 |
|
✗ |
Eigen::VectorXd::Zero(state->get_nv()); |
| 76 |
|
|
} |
| 77 |
|
✗ |
Eigen::VectorXd x0(default_state); |
| 78 |
|
|
std::vector<std::shared_ptr<crocoddyl::ActionModelAbstract>> runningModels( |
| 79 |
|
✗ |
N, runningModel); |
| 80 |
|
|
std::shared_ptr<crocoddyl::ShootingProblem> problem = |
| 81 |
|
|
std::make_shared<crocoddyl::ShootingProblem>(x0, runningModels, |
| 82 |
|
✗ |
terminalModel); |
| 83 |
|
|
|
| 84 |
|
|
// Computing the warm-start |
| 85 |
|
✗ |
std::vector<Eigen::VectorXd> xs(N + 1, x0); |
| 86 |
|
|
std::vector<Eigen::VectorXd> us( |
| 87 |
|
✗ |
N, Eigen::VectorXd::Zero(runningModel->get_nu())); |
| 88 |
|
✗ |
for (unsigned int i = 0; i < N; ++i) { |
| 89 |
|
|
const std::shared_ptr<crocoddyl::ActionModelAbstract>& model = |
| 90 |
|
✗ |
problem->get_runningModels()[i]; |
| 91 |
|
|
const std::shared_ptr<crocoddyl::ActionDataAbstract>& data = |
| 92 |
|
✗ |
problem->get_runningDatas()[i]; |
| 93 |
|
✗ |
model->quasiStatic(data, us[i], x0); |
| 94 |
|
|
} |
| 95 |
|
✗ |
crocoddyl::SolverFDDP ddp(problem); |
| 96 |
|
✗ |
ddp.setCandidate(xs, us, false); |
| 97 |
|
|
std::shared_ptr<crocoddyl::ActionDataAbstract> runningData = |
| 98 |
|
✗ |
runningModel->createData(); |
| 99 |
|
|
|
| 100 |
|
|
// Casting objects to floats |
| 101 |
|
✗ |
Eigen::VectorXf x0_f = x0.cast<float>(); |
| 102 |
|
|
std::vector<std::shared_ptr<crocoddyl::ActionModelAbstractTpl<float>>> |
| 103 |
|
✗ |
casted_runningModels = crocoddyl::vector_cast<float>(runningModels); |
| 104 |
|
|
std::shared_ptr<crocoddyl::ShootingProblemTpl<float>> casted_problem = |
| 105 |
|
|
std::make_shared<crocoddyl::ShootingProblemTpl<float>>( |
| 106 |
|
✗ |
problem->cast<float>()); |
| 107 |
|
✗ |
std::vector<Eigen::VectorXf> xs_f; |
| 108 |
|
✗ |
std::vector<Eigen::VectorXf> us_f; |
| 109 |
|
✗ |
for (unsigned int i = 0; i < N; ++i) { |
| 110 |
|
✗ |
xs_f.push_back(xs[i].cast<float>()); |
| 111 |
|
✗ |
us_f.push_back(us[i].cast<float>()); |
| 112 |
|
|
} |
| 113 |
|
✗ |
xs_f.push_back(xs.back().cast<float>()); |
| 114 |
|
|
|
| 115 |
|
|
#ifdef CROCODDYL_WITH_CODEGEN |
| 116 |
|
|
// Code generation of the running an terminal models |
| 117 |
|
|
typedef CppAD::AD<CppAD::cg::CG<double>> ADScalar; |
| 118 |
|
|
std::shared_ptr<crocoddyl::ActionModelAbstractTpl<ADScalar>> ad_runningModel, |
| 119 |
|
|
ad_terminalModel; |
| 120 |
|
|
if (robot.robot_name == "Talos_arm") { |
| 121 |
|
|
crocoddyl::benchmark::build_arm_action_models(ad_runningModel, |
| 122 |
|
|
ad_terminalModel); |
| 123 |
|
|
} else if (robot.robot_name == "Kinova_arm") { |
| 124 |
|
|
crocoddyl::benchmark::build_arm_kinova_action_models(ad_runningModel, |
| 125 |
|
|
ad_terminalModel); |
| 126 |
|
|
} else { |
| 127 |
|
|
crocoddyl::benchmark::build_contact_action_models(robot, ad_runningModel, |
| 128 |
|
|
ad_terminalModel); |
| 129 |
|
|
} |
| 130 |
|
|
|
| 131 |
|
|
std::shared_ptr<crocoddyl::ActionModelAbstract> cg_runningModel = |
| 132 |
|
|
std::make_shared<crocoddyl::ActionModelCodeGen>( |
| 133 |
|
|
ad_runningModel, robot.robot_name + "_running_cg"); |
| 134 |
|
|
std::shared_ptr<crocoddyl::ActionModelAbstract> cg_terminalModel = |
| 135 |
|
|
std::make_shared<crocoddyl::ActionModelCodeGen>( |
| 136 |
|
|
ad_terminalModel, robot.robot_name + "_terminal_cg"); |
| 137 |
|
|
|
| 138 |
|
|
// Defining the shooting problem for both cases: with and without code |
| 139 |
|
|
// generation |
| 140 |
|
|
std::vector<std::shared_ptr<crocoddyl::ActionModelAbstract>> cg_runningModels( |
| 141 |
|
|
N, cg_runningModel); |
| 142 |
|
|
std::shared_ptr<crocoddyl::ShootingProblem> cg_problem = |
| 143 |
|
|
std::make_shared<crocoddyl::ShootingProblem>(x0, cg_runningModels, |
| 144 |
|
|
cg_terminalModel); |
| 145 |
|
|
|
| 146 |
|
|
// Check that code-generated action model is the same as original. |
| 147 |
|
|
/**************************************************************************/ |
| 148 |
|
|
crocoddyl::SolverFDDP cg_ddp(cg_problem); |
| 149 |
|
|
cg_ddp.setCandidate(xs, us, false); |
| 150 |
|
|
std::shared_ptr<crocoddyl::ActionDataAbstract> cg_runningData = |
| 151 |
|
|
cg_runningModel->createData(); |
| 152 |
|
|
Eigen::VectorXd x_rand = cg_runningModel->get_state()->rand(); |
| 153 |
|
|
Eigen::VectorXd u_rand = Eigen::VectorXd::Random(cg_runningModel->get_nu()); |
| 154 |
|
|
runningModel->calc(runningData, x_rand, u_rand); |
| 155 |
|
|
runningModel->calcDiff(runningData, x_rand, u_rand); |
| 156 |
|
|
cg_runningModel->calc(cg_runningData, x_rand, u_rand); |
| 157 |
|
|
cg_runningModel->calcDiff(cg_runningData, x_rand, u_rand); |
| 158 |
|
|
assert_pretty(cg_runningData->xnext.isApprox(runningData->xnext), |
| 159 |
|
|
"Problem in xnext"); |
| 160 |
|
|
assert_pretty(std::abs(cg_runningData->cost - runningData->cost) < 1e-10, |
| 161 |
|
|
"Problem in cost"); |
| 162 |
|
|
assert_pretty(cg_runningData->Lx.isApprox(runningData->Lx), "Problem in Lx"); |
| 163 |
|
|
assert_pretty(cg_runningData->Lu.isApprox(runningData->Lu), "Problem in Lu"); |
| 164 |
|
|
assert_pretty(cg_runningData->Lxx.isApprox(runningData->Lxx), |
| 165 |
|
|
"Problem in Lxx"); |
| 166 |
|
|
assert_pretty(cg_runningData->Lxu.isApprox(runningData->Lxu), |
| 167 |
|
|
"Problem in Lxu"); |
| 168 |
|
|
assert_pretty(cg_runningData->Luu.isApprox(runningData->Luu), |
| 169 |
|
|
"Problem in Luu"); |
| 170 |
|
|
assert_pretty(cg_runningData->Fx.isApprox(runningData->Fx), "Problem in Fx"); |
| 171 |
|
|
assert_pretty(cg_runningData->Fu.isApprox(runningData->Fu), "Problem in Fu"); |
| 172 |
|
|
#endif // CROCODDYL_WITH_CODEGEN |
| 173 |
|
|
|
| 174 |
|
|
/******************************* create csv file |
| 175 |
|
|
* *******************************/ |
| 176 |
|
✗ |
const std::string csv_filename = "/tmp/" + robot.robot_name + "_" + |
| 177 |
|
✗ |
std::to_string(state->get_nq()) + |
| 178 |
|
✗ |
"DoF.bench"; |
| 179 |
|
✗ |
CsvStream csv(csv_filename); |
| 180 |
|
✗ |
csv << "fn_name" |
| 181 |
|
✗ |
<< "nthreads" |
| 182 |
|
✗ |
<< "with_cg" |
| 183 |
|
✗ |
<< "mean" |
| 184 |
|
✗ |
<< "stddev" |
| 185 |
|
✗ |
<< "max" |
| 186 |
|
✗ |
<< "min" |
| 187 |
|
✗ |
<< "mean_per_nodes" |
| 188 |
|
✗ |
<< "stddev_per_nodes" << csv.endl; |
| 189 |
|
|
|
| 190 |
|
|
/*******************************************************************************/ |
| 191 |
|
|
/*********************************** TIMINGS |
| 192 |
|
|
* ***********************************/ |
| 193 |
|
✗ |
Eigen::ArrayXd duration(T); |
| 194 |
|
✗ |
Eigen::ArrayXd avg(CROCODDYL_WITH_NTHREADS); |
| 195 |
|
✗ |
Eigen::ArrayXd stddev(CROCODDYL_WITH_NTHREADS); |
| 196 |
|
|
|
| 197 |
|
|
/*******************************************************************************/ |
| 198 |
|
|
/*************************** ACTION MODEL TIMINGS |
| 199 |
|
|
* ******************************/ |
| 200 |
|
✗ |
std::cout << "Without Code Generation:" << std::endl; |
| 201 |
|
✗ |
problem->calc(xs, us); |
| 202 |
|
|
// calcDiff timings |
| 203 |
|
✗ |
for (int ithread = 0; ithread < CROCODDYL_WITH_NTHREADS; ++ithread) { |
| 204 |
|
✗ |
duration.setZero(); |
| 205 |
|
✗ |
for (unsigned int i = 0; i < T; ++i) { |
| 206 |
|
✗ |
crocoddyl::Timer timer; |
| 207 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 208 |
|
|
#pragma omp parallel for num_threads(ithread + 1) |
| 209 |
|
|
#endif |
| 210 |
|
✗ |
for (unsigned int j = 0; j < N; ++j) { |
| 211 |
|
✗ |
runningModels[j]->calcDiff(problem->get_runningDatas()[j], xs[j], |
| 212 |
|
✗ |
us[j]); |
| 213 |
|
|
} |
| 214 |
|
✗ |
duration[i] = timer.get_us_duration(); |
| 215 |
|
|
} |
| 216 |
|
✗ |
avg[ithread] = AVG(duration); |
| 217 |
|
✗ |
stddev[ithread] = STDDEV(duration); |
| 218 |
|
✗ |
std::cout << ithread + 1 << " threaded calcDiff [us]:\t" << avg[ithread] |
| 219 |
|
✗ |
<< " +- " << stddev[ithread] << " (max: " << duration.maxCoeff() |
| 220 |
|
✗ |
<< ", min: " << duration.minCoeff() |
| 221 |
|
✗ |
<< ", per nodes: " << avg[ithread] * (ithread + 1) / N << " +- " |
| 222 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << ")" << std::endl; |
| 223 |
|
✗ |
csv << "calcDiff" << (ithread + 1) << false << avg[ithread] |
| 224 |
|
✗ |
<< stddev[ithread] << duration.maxCoeff() << duration.minCoeff() |
| 225 |
|
✗ |
<< avg[ithread] * (ithread + 1) / N |
| 226 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << csv.endl; |
| 227 |
|
|
} |
| 228 |
|
|
|
| 229 |
|
|
// calc timings |
| 230 |
|
✗ |
for (int ithread = 0; ithread < CROCODDYL_WITH_NTHREADS; ++ithread) { |
| 231 |
|
✗ |
duration.setZero(); |
| 232 |
|
✗ |
for (unsigned int i = 0; i < T; ++i) { |
| 233 |
|
✗ |
crocoddyl::Timer timer; |
| 234 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 235 |
|
|
#pragma omp parallel for num_threads(ithread + 1) |
| 236 |
|
|
#endif |
| 237 |
|
✗ |
for (unsigned int j = 0; j < N; ++j) { |
| 238 |
|
✗ |
runningModels[j]->calc(problem->get_runningDatas()[j], xs[j], us[j]); |
| 239 |
|
|
} |
| 240 |
|
✗ |
duration[i] = timer.get_us_duration(); |
| 241 |
|
|
} |
| 242 |
|
✗ |
avg[ithread] = AVG(duration); |
| 243 |
|
✗ |
stddev[ithread] = STDDEV(duration); |
| 244 |
|
✗ |
std::cout << ithread + 1 << " threaded calc [us]: \t" << avg[ithread] |
| 245 |
|
✗ |
<< " +- " << stddev[ithread] << " (max: " << duration.maxCoeff() |
| 246 |
|
✗ |
<< ", min: " << duration.minCoeff() |
| 247 |
|
✗ |
<< ", per nodes: " << avg[ithread] * (ithread + 1) / N << " +- " |
| 248 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << ")" << std::endl; |
| 249 |
|
✗ |
csv << "calc" << (ithread + 1) << false << avg[ithread] << stddev[ithread] |
| 250 |
|
✗ |
<< duration.maxCoeff() << duration.minCoeff() |
| 251 |
|
✗ |
<< avg[ithread] * (ithread + 1) / N |
| 252 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << csv.endl; |
| 253 |
|
|
} |
| 254 |
|
|
|
| 255 |
|
✗ |
std::cout << "Without Code Generation (floats):" << std::endl; |
| 256 |
|
✗ |
casted_problem->calc(xs_f, us_f); |
| 257 |
|
|
// calcDiff timings |
| 258 |
|
✗ |
for (int ithread = 0; ithread < CROCODDYL_WITH_NTHREADS; ++ithread) { |
| 259 |
|
✗ |
duration.setZero(); |
| 260 |
|
✗ |
for (unsigned int i = 0; i < T; ++i) { |
| 261 |
|
✗ |
crocoddyl::Timer timer; |
| 262 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 263 |
|
|
#pragma omp parallel for num_threads(ithread + 1) |
| 264 |
|
|
#endif |
| 265 |
|
✗ |
for (unsigned int j = 0; j < N; ++j) { |
| 266 |
|
✗ |
casted_runningModels[j]->calcDiff(casted_problem->get_runningDatas()[j], |
| 267 |
|
✗ |
xs_f[j], us_f[j]); |
| 268 |
|
|
} |
| 269 |
|
✗ |
duration[i] = timer.get_us_duration(); |
| 270 |
|
|
} |
| 271 |
|
✗ |
avg[ithread] = AVG(duration); |
| 272 |
|
✗ |
stddev[ithread] = STDDEV(duration); |
| 273 |
|
✗ |
std::cout << ithread + 1 << " threaded calcDiff [us]:\t" << avg[ithread] |
| 274 |
|
✗ |
<< " +- " << stddev[ithread] << " (max: " << duration.maxCoeff() |
| 275 |
|
✗ |
<< ", min: " << duration.minCoeff() |
| 276 |
|
✗ |
<< ", per nodes: " << avg[ithread] * (ithread + 1) / N << " +- " |
| 277 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << ")" << std::endl; |
| 278 |
|
✗ |
csv << "calcDiff" << (ithread + 1) << false << avg[ithread] |
| 279 |
|
✗ |
<< stddev[ithread] << duration.maxCoeff() << duration.minCoeff() |
| 280 |
|
✗ |
<< avg[ithread] * (ithread + 1) / N |
| 281 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << csv.endl; |
| 282 |
|
|
} |
| 283 |
|
|
|
| 284 |
|
|
// calc timings |
| 285 |
|
✗ |
for (int ithread = 0; ithread < CROCODDYL_WITH_NTHREADS; ++ithread) { |
| 286 |
|
✗ |
duration.setZero(); |
| 287 |
|
✗ |
for (unsigned int i = 0; i < T; ++i) { |
| 288 |
|
✗ |
crocoddyl::Timer timer; |
| 289 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 290 |
|
|
#pragma omp parallel for num_threads(ithread + 1) |
| 291 |
|
|
#endif |
| 292 |
|
✗ |
for (unsigned int j = 0; j < N; ++j) { |
| 293 |
|
✗ |
casted_runningModels[j]->calc(casted_problem->get_runningDatas()[j], |
| 294 |
|
✗ |
xs_f[j], us_f[j]); |
| 295 |
|
|
} |
| 296 |
|
✗ |
duration[i] = timer.get_us_duration(); |
| 297 |
|
|
} |
| 298 |
|
✗ |
avg[ithread] = AVG(duration); |
| 299 |
|
✗ |
stddev[ithread] = STDDEV(duration); |
| 300 |
|
✗ |
std::cout << ithread + 1 << " threaded calc [us]: \t" << avg[ithread] |
| 301 |
|
✗ |
<< " +- " << stddev[ithread] << " (max: " << duration.maxCoeff() |
| 302 |
|
✗ |
<< ", min: " << duration.minCoeff() |
| 303 |
|
✗ |
<< ", per nodes: " << avg[ithread] * (ithread + 1) / N << " +- " |
| 304 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << ")" << std::endl; |
| 305 |
|
✗ |
csv << "calc" << (ithread + 1) << false << avg[ithread] << stddev[ithread] |
| 306 |
|
✗ |
<< duration.maxCoeff() << duration.minCoeff() |
| 307 |
|
✗ |
<< avg[ithread] * (ithread + 1) / N |
| 308 |
|
✗ |
<< stddev[ithread] * (ithread + 1) / N << csv.endl; |
| 309 |
|
|
} |
| 310 |
|
|
|
| 311 |
|
|
/*******************************************************************************/ |
| 312 |
|
|
/******************* DDP BACKWARD AND FORWARD PASSES TIMINGS |
| 313 |
|
|
* *******************/ |
| 314 |
|
|
// Backward pass timings |
| 315 |
|
✗ |
ddp.calcDiff(); |
| 316 |
|
✗ |
duration.setZero(); |
| 317 |
|
✗ |
for (unsigned int i = 0; i < T; ++i) { |
| 318 |
|
✗ |
crocoddyl::Timer timer; |
| 319 |
|
✗ |
ddp.backwardPass(); |
| 320 |
|
✗ |
duration[i] = timer.get_us_duration(); |
| 321 |
|
|
} |
| 322 |
|
✗ |
double avg_bp = AVG(duration); |
| 323 |
|
✗ |
double stddev_bp = STDDEV(duration); |
| 324 |
|
✗ |
std::cout << "backwardPass [us]:\t\t" << avg_bp << " +- " << stddev_bp |
| 325 |
|
✗ |
<< " (max: " << duration.maxCoeff() |
| 326 |
|
✗ |
<< ", min: " << duration.minCoeff() << ", per nodes: " << avg_bp / N |
| 327 |
|
✗ |
<< " +- " << stddev_bp / N << ")" << std::endl; |
| 328 |
|
|
|
| 329 |
|
✗ |
csv << "backwardPass" << 1 << false << avg_bp << stddev_bp |
| 330 |
|
✗ |
<< duration.maxCoeff() << duration.minCoeff() << avg_bp / N |
| 331 |
|
✗ |
<< stddev_bp / N << csv.endl; |
| 332 |
|
|
|
| 333 |
|
|
// Forward pass timings |
| 334 |
|
✗ |
duration.setZero(); |
| 335 |
|
✗ |
for (unsigned int i = 0; i < T; ++i) { |
| 336 |
|
✗ |
crocoddyl::Timer timer; |
| 337 |
|
✗ |
ddp.forwardPass(0.005); |
| 338 |
|
✗ |
duration[i] = timer.get_us_duration(); |
| 339 |
|
|
} |
| 340 |
|
✗ |
double avg_fp = AVG(duration); |
| 341 |
|
✗ |
double stddev_fp = STDDEV(duration); |
| 342 |
|
✗ |
std::cout << "forwardPass [us]: \t\t" << avg_fp << " +- " << stddev_fp |
| 343 |
|
✗ |
<< " (max: " << duration.maxCoeff() |
| 344 |
|
✗ |
<< ", min: " << duration.minCoeff() << ", per nodes: " << avg_fp / N |
| 345 |
|
✗ |
<< " +- " << stddev_fp / N << ")" << std::endl; |
| 346 |
|
|
|
| 347 |
|
✗ |
csv << "forwardPass" << 1 << false << avg_fp << stddev_fp |
| 348 |
|
✗ |
<< duration.maxCoeff() << duration.minCoeff() << avg_fp / N |
| 349 |
|
✗ |
<< stddev_fp / N << csv.endl; |
| 350 |
|
|
|
| 351 |
|
|
#ifdef CROCODDYL_WITH_CODEGEN |
| 352 |
|
|
|
| 353 |
|
|
/*******************************************************************************/ |
| 354 |
|
|
/*************************** CODE GENERATION TIMINGS |
| 355 |
|
|
* ***************************/ |
| 356 |
|
|
/*******************************************************************************/ |
| 357 |
|
|
|
| 358 |
|
|
/*******************************************************************************/ |
| 359 |
|
|
/****************************** ACTION MODEL TIMINGS |
| 360 |
|
|
* ***************************/ |
| 361 |
|
|
std::cout << std::endl << "With Code Generation:" << std::endl; |
| 362 |
|
|
// calcDiff timings |
| 363 |
|
|
cg_problem->calc(xs, us); |
| 364 |
|
|
for (int ithread = 0; ithread < CROCODDYL_WITH_NTHREADS; ++ithread) { |
| 365 |
|
|
duration.setZero(); |
| 366 |
|
|
for (unsigned int i = 0; i < T; ++i) { |
| 367 |
|
|
crocoddyl::Timer timer; |
| 368 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 369 |
|
|
#pragma omp parallel for num_threads(ithread + 1) |
| 370 |
|
|
#endif |
| 371 |
|
|
for (unsigned int j = 0; j < N; ++j) { |
| 372 |
|
|
cg_runningModels[j]->calcDiff(cg_problem->get_runningDatas()[j], xs[j], |
| 373 |
|
|
us[j]); |
| 374 |
|
|
} |
| 375 |
|
|
duration[i] = timer.get_us_duration(); |
| 376 |
|
|
} |
| 377 |
|
|
avg[ithread] = AVG(duration); |
| 378 |
|
|
stddev[ithread] = STDDEV(duration); |
| 379 |
|
|
std::cout << ithread + 1 << " threaded calcDiff [us]:\t" << avg[ithread] |
| 380 |
|
|
<< " +- " << stddev[ithread] << " (max: " << duration.maxCoeff() |
| 381 |
|
|
<< ", min: " << duration.minCoeff() |
| 382 |
|
|
<< ", per nodes: " << avg[ithread] * (ithread + 1) / N << " +- " |
| 383 |
|
|
<< stddev[ithread] * (ithread + 1) / N << ")" << std::endl; |
| 384 |
|
|
csv << "calcDiff" << (ithread + 1) << true << avg[ithread] |
| 385 |
|
|
<< stddev[ithread] << duration.maxCoeff() << duration.minCoeff() |
| 386 |
|
|
<< avg[ithread] * (ithread + 1) / N |
| 387 |
|
|
<< stddev[ithread] * (ithread + 1) / N << csv.endl; |
| 388 |
|
|
} |
| 389 |
|
|
|
| 390 |
|
|
// calc timings |
| 391 |
|
|
for (int ithread = 0; ithread < CROCODDYL_WITH_NTHREADS; ++ithread) { |
| 392 |
|
|
duration.setZero(); |
| 393 |
|
|
for (unsigned int i = 0; i < T; ++i) { |
| 394 |
|
|
crocoddyl::Timer timer; |
| 395 |
|
|
#ifdef CROCODDYL_WITH_MULTITHREADING |
| 396 |
|
|
#pragma omp parallel for num_threads(ithread + 1) |
| 397 |
|
|
#endif |
| 398 |
|
|
for (unsigned int j = 0; j < N; ++j) { |
| 399 |
|
|
cg_runningModels[j]->calc(cg_problem->get_runningDatas()[j], xs[j], |
| 400 |
|
|
us[j]); |
| 401 |
|
|
} |
| 402 |
|
|
duration[i] = timer.get_us_duration(); |
| 403 |
|
|
} |
| 404 |
|
|
avg[ithread] = AVG(duration); |
| 405 |
|
|
stddev[ithread] = STDDEV(duration); |
| 406 |
|
|
std::cout << ithread + 1 << " threaded calc [us]: \t" << avg[ithread] |
| 407 |
|
|
<< " +- " << stddev[ithread] << " (max: " << duration.maxCoeff() |
| 408 |
|
|
<< ", min: " << duration.minCoeff() |
| 409 |
|
|
<< ", per nodes: " << avg[ithread] * (ithread + 1) / N << " +- " |
| 410 |
|
|
<< stddev[ithread] * (ithread + 1) / N << ")" << std::endl; |
| 411 |
|
|
csv << "calc" << (ithread + 1) << true << avg[ithread] << stddev[ithread] |
| 412 |
|
|
<< duration.maxCoeff() << duration.minCoeff() |
| 413 |
|
|
<< avg[ithread] * (ithread + 1) / N |
| 414 |
|
|
<< stddev[ithread] * (ithread + 1) / N << csv.endl; |
| 415 |
|
|
} |
| 416 |
|
|
|
| 417 |
|
|
/*******************************************************************************/ |
| 418 |
|
|
/******************* DDP BACKWARD AND FORWARD PASSES TIMINGS |
| 419 |
|
|
* *******************/ |
| 420 |
|
|
// Backward pass timings |
| 421 |
|
|
cg_ddp.calcDiff(); |
| 422 |
|
|
for (unsigned int i = 0; i < T; ++i) { |
| 423 |
|
|
crocoddyl::Timer timer; |
| 424 |
|
|
cg_ddp.backwardPass(); |
| 425 |
|
|
duration[i] = timer.get_us_duration(); |
| 426 |
|
|
} |
| 427 |
|
|
avg_bp = AVG(duration); |
| 428 |
|
|
stddev_bp = STDDEV(duration); |
| 429 |
|
|
std::cout << "backwardPass [us]:\t\t" << avg_bp << " +- " << stddev_bp |
| 430 |
|
|
<< " (max: " << duration.maxCoeff() |
| 431 |
|
|
<< ", min: " << duration.minCoeff() << ", per nodes: " << avg_bp / N |
| 432 |
|
|
<< " +- " << stddev_bp / N << ")" << std::endl; |
| 433 |
|
|
csv << "backwardPass" << 1 << true << avg_bp << stddev_bp |
| 434 |
|
|
<< duration.maxCoeff() << duration.minCoeff() << avg_bp / N |
| 435 |
|
|
<< stddev_bp / N << csv.endl; |
| 436 |
|
|
|
| 437 |
|
|
// Forward pass timings |
| 438 |
|
|
for (unsigned int i = 0; i < T; ++i) { |
| 439 |
|
|
crocoddyl::Timer timer; |
| 440 |
|
|
cg_ddp.forwardPass(0.005); |
| 441 |
|
|
duration[i] = timer.get_us_duration(); |
| 442 |
|
|
} |
| 443 |
|
|
avg_fp = AVG(duration); |
| 444 |
|
|
stddev_fp = STDDEV(duration); |
| 445 |
|
|
std::cout << "forwardPass [us]: \t\t" << avg_fp << " +- " << stddev_fp |
| 446 |
|
|
<< " (max: " << duration.maxCoeff() |
| 447 |
|
|
<< ", min: " << duration.minCoeff() << ", per nodes: " << avg_fp / N |
| 448 |
|
|
<< " +- " << stddev_fp / N << ")" << std::endl; |
| 449 |
|
|
|
| 450 |
|
|
csv << "forwardPass" << 1 << true << avg_fp << stddev_fp |
| 451 |
|
|
<< duration.maxCoeff() << duration.minCoeff() << avg_fp / N |
| 452 |
|
|
<< stddev_fp / N << csv.endl; |
| 453 |
|
|
|
| 454 |
|
|
#endif // CROCODDYL_WITH_CODEGEN |
| 455 |
|
|
} |
| 456 |
|
|
|
| 457 |
|
✗ |
int main() { |
| 458 |
|
|
// Arm Manipulation Benchmarks |
| 459 |
|
✗ |
std::cout << "********************Talos 4DoF Arm******************" |
| 460 |
|
✗ |
<< std::endl; |
| 461 |
|
✗ |
std::vector<std::string> contact_names; |
| 462 |
|
✗ |
std::vector<crocoddyl::ContactType> contact_types; |
| 463 |
|
|
RobotEENames talosArm4Dof( |
| 464 |
|
|
"Talos_arm", contact_names, contact_types, |
| 465 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/talos_data/robots/talos_left_arm.urdf", |
| 466 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/talos_data/srdf/talos.srdf", |
| 467 |
|
✗ |
"gripper_left_joint", "half_sitting"); |
| 468 |
|
|
|
| 469 |
|
✗ |
print_benchmark(talosArm4Dof); |
| 470 |
|
|
// Arm Manipulation Benchmarks |
| 471 |
|
✗ |
std::cout << "******************** Kinova Arm ******************" |
| 472 |
|
✗ |
<< std::endl; |
| 473 |
|
|
RobotEENames kinovaArm( |
| 474 |
|
|
"Kinova_arm", contact_names, contact_types, |
| 475 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/kinova_description/robots/kinova.urdf", |
| 476 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/kinova_description/srdf/kinova.srdf", |
| 477 |
|
✗ |
"gripper_left_joint", "arm_up"); |
| 478 |
|
|
|
| 479 |
|
✗ |
print_benchmark(kinovaArm); |
| 480 |
|
|
|
| 481 |
|
|
// Quadruped Solo Benchmarks |
| 482 |
|
✗ |
std::cout << "********************Quadruped Solo******************" |
| 483 |
|
✗ |
<< std::endl; |
| 484 |
|
✗ |
contact_names.clear(); |
| 485 |
|
✗ |
contact_types.clear(); |
| 486 |
|
✗ |
contact_names.push_back("FR_KFE"); |
| 487 |
|
✗ |
contact_names.push_back("HL_KFE"); |
| 488 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 489 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 490 |
|
|
RobotEENames quadrupedSolo( |
| 491 |
|
|
"Solo", contact_names, contact_types, |
| 492 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/solo_description/robots/solo.urdf", |
| 493 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/solo_description/srdf/solo.srdf", "HL_KFE", |
| 494 |
|
✗ |
"standing"); |
| 495 |
|
|
|
| 496 |
|
✗ |
print_benchmark(quadrupedSolo); |
| 497 |
|
|
|
| 498 |
|
|
// Quadruped Anymal Benchmarks |
| 499 |
|
✗ |
std::cout << "********************Quadruped Anymal******************" |
| 500 |
|
✗ |
<< std::endl; |
| 501 |
|
✗ |
contact_names.clear(); |
| 502 |
|
✗ |
contact_types.clear(); |
| 503 |
|
✗ |
contact_names.push_back("RF_KFE"); |
| 504 |
|
✗ |
contact_names.push_back("LF_KFE"); |
| 505 |
|
✗ |
contact_names.push_back("LH_KFE"); |
| 506 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 507 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 508 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 509 |
|
|
RobotEENames quadrupedAnymal( |
| 510 |
|
|
"Anymal", contact_names, contact_types, |
| 511 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR |
| 512 |
|
|
"/anymal_b_simple_description/robots/anymal.urdf", |
| 513 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR |
| 514 |
|
|
"/anymal_b_simple_description/srdf/anymal.srdf", |
| 515 |
|
✗ |
"RH_KFE", "standing"); |
| 516 |
|
|
|
| 517 |
|
✗ |
print_benchmark(quadrupedAnymal); |
| 518 |
|
|
|
| 519 |
|
|
// Quadruped HyQ Benchmarks |
| 520 |
|
✗ |
std::cout << "******************** Quadruped HyQ ******************" |
| 521 |
|
✗ |
<< std::endl; |
| 522 |
|
✗ |
contact_names.clear(); |
| 523 |
|
✗ |
contact_types.clear(); |
| 524 |
|
✗ |
contact_names.push_back("rf_kfe_joint"); |
| 525 |
|
✗ |
contact_names.push_back("lf_kfe_joint"); |
| 526 |
|
✗ |
contact_names.push_back("lh_kfe_joint"); |
| 527 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 528 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 529 |
|
✗ |
contact_types.push_back(crocoddyl::Contact3D); |
| 530 |
|
|
RobotEENames quadrupedHyQ("HyQ", contact_names, contact_types, |
| 531 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR |
| 532 |
|
|
"/hyq_description/robots/hyq_no_sensors.urdf", |
| 533 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR |
| 534 |
|
|
"/hyq_description/srdf/hyq.srdf", |
| 535 |
|
✗ |
"rh_kfe_joint", "standing"); |
| 536 |
|
|
|
| 537 |
|
✗ |
print_benchmark(quadrupedHyQ); |
| 538 |
|
|
|
| 539 |
|
|
// Biped icub Benchmarks |
| 540 |
|
✗ |
std::cout << "********************Biped iCub ***********************" |
| 541 |
|
✗ |
<< std::endl; |
| 542 |
|
✗ |
contact_names.clear(); |
| 543 |
|
✗ |
contact_types.clear(); |
| 544 |
|
✗ |
contact_names.push_back("r_ankle_roll"); |
| 545 |
|
✗ |
contact_names.push_back("l_ankle_roll"); |
| 546 |
|
✗ |
contact_types.push_back(crocoddyl::Contact6D); |
| 547 |
|
✗ |
contact_types.push_back(crocoddyl::Contact6D); |
| 548 |
|
|
|
| 549 |
|
|
RobotEENames bipedIcub( |
| 550 |
|
|
"iCub", contact_names, contact_types, |
| 551 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/icub_description/robots/icub_reduced.urdf", |
| 552 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/icub_description/srdf/icub.srdf", |
| 553 |
|
✗ |
"r_wrist_yaw", "half_sitting"); |
| 554 |
|
✗ |
print_benchmark(bipedIcub); |
| 555 |
|
|
|
| 556 |
|
|
// Biped icub Benchmarks |
| 557 |
|
✗ |
std::cout << "********************Biped Talos***********************" |
| 558 |
|
✗ |
<< std::endl; |
| 559 |
|
✗ |
contact_names.clear(); |
| 560 |
|
✗ |
contact_types.clear(); |
| 561 |
|
✗ |
contact_names.push_back("leg_right_6_joint"); |
| 562 |
|
✗ |
contact_names.push_back("leg_left_6_joint"); |
| 563 |
|
✗ |
contact_types.push_back(crocoddyl::Contact6D); |
| 564 |
|
✗ |
contact_types.push_back(crocoddyl::Contact6D); |
| 565 |
|
|
|
| 566 |
|
|
RobotEENames bipedTalos( |
| 567 |
|
|
"Talos", contact_names, contact_types, |
| 568 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/talos_data/robots/talos_reduced.urdf", |
| 569 |
|
|
EXAMPLE_ROBOT_DATA_MODEL_DIR "/talos_data/srdf/talos.srdf", |
| 570 |
|
✗ |
"arm_right_7_joint", "half_sitting"); |
| 571 |
|
✗ |
print_benchmark(bipedTalos); |
| 572 |
|
|
|
| 573 |
|
✗ |
return 0; |
| 574 |
|
|
} |
| 575 |
|
|
|