| Line |
Branch |
Exec |
Source |
| 1 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
| 2 |
|
|
// BSD 3-Clause License |
| 3 |
|
|
// |
| 4 |
|
|
// Copyright (C) 2019-2025, University of Edinburgh, University of Trento, |
| 5 |
|
|
// LAAS-CNRS, IRI: CSIC-UPC, Heriot-Watt University |
| 6 |
|
|
// Copyright note valid unless otherwise stated in individual files. |
| 7 |
|
|
// All rights reserved. |
| 8 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
| 9 |
|
|
|
| 10 |
|
|
namespace crocoddyl { |
| 11 |
|
|
|
| 12 |
|
|
template <typename Scalar> |
| 13 |
|
✗ |
IntegratedActionModelRKTpl<Scalar>::IntegratedActionModelRKTpl( |
| 14 |
|
|
std::shared_ptr<DifferentialActionModelAbstract> model, |
| 15 |
|
|
std::shared_ptr<ControlParametrizationModelAbstract> control, |
| 16 |
|
|
const RKType rktype, const Scalar time_step, const bool with_cost_residual) |
| 17 |
|
✗ |
: Base(model, control, time_step, with_cost_residual), rk_type_(rktype) { |
| 18 |
|
✗ |
set_rk_type(rktype); |
| 19 |
|
|
} |
| 20 |
|
|
|
| 21 |
|
|
template <typename Scalar> |
| 22 |
|
✗ |
IntegratedActionModelRKTpl<Scalar>::IntegratedActionModelRKTpl( |
| 23 |
|
|
std::shared_ptr<DifferentialActionModelAbstract> model, const RKType rktype, |
| 24 |
|
|
const Scalar time_step, const bool with_cost_residual) |
| 25 |
|
✗ |
: Base(model, time_step, with_cost_residual), rk_type_(rktype) { |
| 26 |
|
✗ |
set_rk_type(rktype); |
| 27 |
|
|
} |
| 28 |
|
|
|
| 29 |
|
|
template <typename Scalar> |
| 30 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::calc( |
| 31 |
|
|
const std::shared_ptr<ActionDataAbstract>& data, |
| 32 |
|
|
const Eigen::Ref<const VectorXs>& x, const Eigen::Ref<const VectorXs>& u) { |
| 33 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 34 |
|
✗ |
throw_pretty( |
| 35 |
|
|
"Invalid argument: " << "x has wrong dimension (it should be " + |
| 36 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 37 |
|
|
} |
| 38 |
|
✗ |
if (static_cast<std::size_t>(u.size()) != nu_) { |
| 39 |
|
✗ |
throw_pretty( |
| 40 |
|
|
"Invalid argument: " << "u has wrong dimension (it should be " + |
| 41 |
|
|
std::to_string(nu_) + ")"); |
| 42 |
|
|
} |
| 43 |
|
✗ |
const std::size_t nv = state_->get_nv(); |
| 44 |
|
✗ |
Data* d = static_cast<Data*>(data.get()); |
| 45 |
|
|
|
| 46 |
|
|
const std::shared_ptr<DifferentialActionDataAbstract>& k0_data = |
| 47 |
|
✗ |
d->differential[0]; |
| 48 |
|
|
const std::shared_ptr<ControlParametrizationDataAbstract>& u0_data = |
| 49 |
|
✗ |
d->control[0]; |
| 50 |
|
✗ |
control_->calc(u0_data, rk_c_[0], u); |
| 51 |
|
✗ |
d->ws[0] = u0_data->w; |
| 52 |
|
✗ |
differential_->calc(k0_data, x, d->ws[0]); |
| 53 |
|
✗ |
d->y[0] = x; |
| 54 |
|
✗ |
d->ki[0].head(nv) = d->y[0].tail(nv); |
| 55 |
|
✗ |
d->ki[0].tail(nv) = k0_data->xout; |
| 56 |
|
✗ |
d->integral[0] = k0_data->cost; |
| 57 |
|
✗ |
for (std::size_t i = 1; i < ni_; ++i) { |
| 58 |
|
|
const std::shared_ptr<DifferentialActionDataAbstract>& ki_data = |
| 59 |
|
✗ |
d->differential[i]; |
| 60 |
|
|
const std::shared_ptr<ControlParametrizationDataAbstract>& ui_data = |
| 61 |
|
✗ |
d->control[i]; |
| 62 |
|
✗ |
d->dx_rk[i].noalias() = time_step_ * rk_c_[i] * d->ki[i - 1]; |
| 63 |
|
✗ |
state_->integrate(x, d->dx_rk[i], d->y[i]); |
| 64 |
|
✗ |
control_->calc(ui_data, rk_c_[i], u); |
| 65 |
|
✗ |
d->ws[i] = ui_data->w; |
| 66 |
|
✗ |
differential_->calc(ki_data, d->y[i], d->ws[i]); |
| 67 |
|
✗ |
d->ki[i].head(nv) = d->y[i].tail(nv); |
| 68 |
|
✗ |
d->ki[i].tail(nv) = ki_data->xout; |
| 69 |
|
✗ |
d->integral[i] = ki_data->cost; |
| 70 |
|
|
} |
| 71 |
|
|
|
| 72 |
|
✗ |
if (ni_ == 2) { |
| 73 |
|
✗ |
d->dx = d->ki[1] * time_step_; |
| 74 |
|
✗ |
d->cost = d->integral[1] * time_step_; |
| 75 |
|
✗ |
} else if (ni_ == 3) { |
| 76 |
|
✗ |
d->dx = (d->ki[0] + Scalar(3.) * d->ki[2]) * time_step_ / Scalar(4.); |
| 77 |
|
✗ |
d->cost = (d->integral[0] + Scalar(3.) * d->integral[2]) * time_step_ / |
| 78 |
|
|
Scalar(4.); |
| 79 |
|
|
} else { |
| 80 |
|
✗ |
d->dx = |
| 81 |
|
✗ |
(d->ki[0] + Scalar(2.) * d->ki[1] + Scalar(2.) * d->ki[2] + d->ki[3]) * |
| 82 |
|
✗ |
time_step_ / Scalar(6.); |
| 83 |
|
✗ |
d->cost = (d->integral[0] + Scalar(2.) * d->integral[1] + |
| 84 |
|
✗ |
Scalar(2.) * d->integral[2] + d->integral[3]) * |
| 85 |
|
✗ |
time_step_ / Scalar(6.); |
| 86 |
|
|
} |
| 87 |
|
✗ |
state_->integrate(x, d->dx, d->xnext); |
| 88 |
|
✗ |
d->g = k0_data->g; |
| 89 |
|
✗ |
d->h = k0_data->h; |
| 90 |
|
✗ |
if (with_cost_residual_) { |
| 91 |
|
✗ |
d->r = k0_data->r; |
| 92 |
|
|
} |
| 93 |
|
|
} |
| 94 |
|
|
|
| 95 |
|
|
template <typename Scalar> |
| 96 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::calc( |
| 97 |
|
|
const std::shared_ptr<ActionDataAbstract>& data, |
| 98 |
|
|
const Eigen::Ref<const VectorXs>& x) { |
| 99 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 100 |
|
✗ |
throw_pretty( |
| 101 |
|
|
"Invalid argument: " << "x has wrong dimension (it should be " + |
| 102 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 103 |
|
|
} |
| 104 |
|
✗ |
Data* d = static_cast<Data*>(data.get()); |
| 105 |
|
|
|
| 106 |
|
|
const std::shared_ptr<DifferentialActionDataAbstract>& k0_data = |
| 107 |
|
✗ |
d->differential[0]; |
| 108 |
|
✗ |
differential_->calc(k0_data, x); |
| 109 |
|
✗ |
d->dx.setZero(); |
| 110 |
|
✗ |
d->xnext = x; |
| 111 |
|
✗ |
d->cost = k0_data->cost; |
| 112 |
|
✗ |
d->g = k0_data->g; |
| 113 |
|
✗ |
d->h = k0_data->h; |
| 114 |
|
✗ |
if (with_cost_residual_) { |
| 115 |
|
✗ |
d->r = k0_data->r; |
| 116 |
|
|
} |
| 117 |
|
|
} |
| 118 |
|
|
|
| 119 |
|
|
template <typename Scalar> |
| 120 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::calcDiff( |
| 121 |
|
|
const std::shared_ptr<ActionDataAbstract>& data, |
| 122 |
|
|
const Eigen::Ref<const VectorXs>& x, const Eigen::Ref<const VectorXs>& u) { |
| 123 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 124 |
|
✗ |
throw_pretty( |
| 125 |
|
|
"Invalid argument: " << "x has wrong dimension (it should be " + |
| 126 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 127 |
|
|
} |
| 128 |
|
✗ |
if (static_cast<std::size_t>(u.size()) != nu_) { |
| 129 |
|
✗ |
throw_pretty( |
| 130 |
|
|
"Invalid argument: " << "u has wrong dimension (it should be " + |
| 131 |
|
|
std::to_string(nu_) + ")"); |
| 132 |
|
|
} |
| 133 |
|
✗ |
const std::size_t nv = state_->get_nv(); |
| 134 |
|
✗ |
const std::size_t nu = control_->get_nu(); |
| 135 |
|
✗ |
Data* d = static_cast<Data*>(data.get()); |
| 136 |
|
✗ |
assert_pretty( |
| 137 |
|
|
MatrixXs(d->dyi_dx[0]) |
| 138 |
|
|
.isApprox(MatrixXs::Identity(state_->get_ndx(), state_->get_ndx())), |
| 139 |
|
|
"you have changed dyi_dx[0] values that supposed to be constant."); |
| 140 |
|
✗ |
assert_pretty( |
| 141 |
|
|
MatrixXs(d->dki_dx[0]) |
| 142 |
|
|
.topRightCorner(nv, nv) |
| 143 |
|
|
.isApprox(MatrixXs::Identity(nv, nv)), |
| 144 |
|
|
"you have changed dki_dx[0] values that supposed to be constant."); |
| 145 |
|
|
|
| 146 |
|
✗ |
for (std::size_t i = 0; i < ni_; ++i) { |
| 147 |
|
✗ |
differential_->calcDiff(d->differential[i], d->y[i], d->ws[i]); |
| 148 |
|
|
} |
| 149 |
|
|
|
| 150 |
|
|
const std::shared_ptr<DifferentialActionDataAbstract>& k0_data = |
| 151 |
|
✗ |
d->differential[0]; |
| 152 |
|
|
const std::shared_ptr<ControlParametrizationDataAbstract>& u0_data = |
| 153 |
|
✗ |
d->control[0]; |
| 154 |
|
✗ |
d->dki_dx[0].bottomRows(nv) = k0_data->Fx; |
| 155 |
|
✗ |
control_->multiplyByJacobian( |
| 156 |
|
✗ |
u0_data, k0_data->Fu, |
| 157 |
|
✗ |
d->dki_du[0].bottomRows(nv)); // dki_du = dki_dw * dw_du |
| 158 |
|
|
|
| 159 |
|
✗ |
d->dli_dx[0] = k0_data->Lx; |
| 160 |
|
✗ |
control_->multiplyJacobianTransposeBy( |
| 161 |
|
✗ |
u0_data, k0_data->Lu, |
| 162 |
|
✗ |
d->dli_du[0]); // dli_du = dli_dw * dw_du |
| 163 |
|
|
|
| 164 |
|
✗ |
d->ddli_ddx[0] = k0_data->Lxx; |
| 165 |
|
✗ |
d->ddli_ddw[0] = k0_data->Luu; |
| 166 |
|
✗ |
control_->multiplyByJacobian( |
| 167 |
|
✗ |
u0_data, d->ddli_ddw[0], |
| 168 |
|
✗ |
d->ddli_dwdu[0]); // ddli_dwdu = ddli_ddw * dw_du |
| 169 |
|
✗ |
control_->multiplyJacobianTransposeBy( |
| 170 |
|
✗ |
u0_data, d->ddli_dwdu[0], |
| 171 |
|
✗ |
d->ddli_ddu[0]); // ddli_ddu = dw_du.T * ddli_dwdu |
| 172 |
|
✗ |
d->ddli_dxdw[0] = k0_data->Lxu; |
| 173 |
|
✗ |
control_->multiplyByJacobian( |
| 174 |
|
✗ |
u0_data, d->ddli_dxdw[0], |
| 175 |
|
✗ |
d->ddli_dxdu[0]); // ddli_dxdu = ddli_dxdw * dw_du |
| 176 |
|
|
|
| 177 |
|
✗ |
for (std::size_t i = 1; i < ni_; ++i) { |
| 178 |
|
|
const std::shared_ptr<DifferentialActionDataAbstract>& ki_data = |
| 179 |
|
✗ |
d->differential[i]; |
| 180 |
|
|
const std::shared_ptr<ControlParametrizationDataAbstract>& ui_data = |
| 181 |
|
✗ |
d->control[i]; |
| 182 |
|
✗ |
d->dyi_dx[i].noalias() = d->dki_dx[i - 1] * rk_c_[i] * time_step_; |
| 183 |
|
✗ |
d->dyi_du[i].noalias() = d->dki_du[i - 1] * rk_c_[i] * time_step_; |
| 184 |
|
✗ |
state_->JintegrateTransport(x, d->dx_rk[i], d->dyi_dx[i], second); |
| 185 |
|
✗ |
state_->Jintegrate(x, d->dx_rk[i], d->dyi_dx[i], d->dyi_dx[i], first, |
| 186 |
|
|
addto); |
| 187 |
|
✗ |
state_->JintegrateTransport(x, d->dx_rk[i], d->dyi_du[i], |
| 188 |
|
|
second); // dyi_du = Jintegrate * dyi_du |
| 189 |
|
|
|
| 190 |
|
|
// Sparse matrix-matrix multiplication for computing: |
| 191 |
|
✗ |
Eigen::Block<MatrixXs> dkvi_dq = d->dki_dx[i].bottomLeftCorner(nv, nv); |
| 192 |
|
✗ |
Eigen::Block<MatrixXs> dkvi_dv = d->dki_dx[i].bottomRightCorner(nv, nv); |
| 193 |
|
✗ |
Eigen::Block<MatrixXs> dkqi_du = d->dki_du[i].topLeftCorner(nv, nu); |
| 194 |
|
✗ |
Eigen::Block<MatrixXs> dkvi_du = d->dki_du[i].bottomLeftCorner(nv, nu); |
| 195 |
|
✗ |
const Eigen::Block<MatrixXs> dki_dqi = ki_data->Fx.bottomLeftCorner(nv, nv); |
| 196 |
|
|
const Eigen::Block<MatrixXs> dki_dvi = |
| 197 |
|
✗ |
ki_data->Fx.bottomRightCorner(nv, nv); |
| 198 |
|
✗ |
const Eigen::Block<MatrixXs> dqi_dq = d->dyi_dx[i].topLeftCorner(nv, nv); |
| 199 |
|
✗ |
const Eigen::Block<MatrixXs> dqi_dv = d->dyi_dx[i].topRightCorner(nv, nv); |
| 200 |
|
✗ |
const Eigen::Block<MatrixXs> dvi_dq = d->dyi_dx[i].bottomLeftCorner(nv, nv); |
| 201 |
|
|
const Eigen::Block<MatrixXs> dvi_dv = |
| 202 |
|
✗ |
d->dyi_dx[i].bottomRightCorner(nv, nv); |
| 203 |
|
✗ |
const Eigen::Block<MatrixXs> dqi_du = d->dyi_du[i].topLeftCorner(nv, nu); |
| 204 |
|
✗ |
const Eigen::Block<MatrixXs> dvi_du = d->dyi_du[i].bottomLeftCorner(nv, nu); |
| 205 |
|
|
// i. d->dki_dx[i].noalias() = d->dki_dy[i] * d->dyi_dx[i], where dki_dy |
| 206 |
|
|
// is ki_data.Fx |
| 207 |
|
✗ |
d->dki_dx[i].topRows(nv) = d->dyi_dx[i].bottomRows(nv); |
| 208 |
|
✗ |
dkvi_dq.noalias() = dki_dqi * dqi_dq; |
| 209 |
|
✗ |
if (i == 1) { |
| 210 |
|
✗ |
dkvi_dv = time_step_ / Scalar(2.) * dki_dqi; |
| 211 |
|
|
} else { |
| 212 |
|
✗ |
dkvi_dv.noalias() = dki_dqi * dqi_dv; |
| 213 |
|
|
} |
| 214 |
|
✗ |
dkvi_dq.noalias() += dki_dvi * dvi_dq; |
| 215 |
|
✗ |
dkvi_dv.noalias() += dki_dvi * dvi_dv; |
| 216 |
|
|
// ii. d->dki_du[i].noalias() = d->dki_dy[i] * d->dyi_du[i], where dki_dy |
| 217 |
|
|
// is ki_data.Fx |
| 218 |
|
✗ |
dkqi_du = dvi_du; |
| 219 |
|
✗ |
dkvi_du.noalias() = dki_dqi * dqi_du; |
| 220 |
|
✗ |
dkvi_du.noalias() += dki_dvi * dvi_du; |
| 221 |
|
|
|
| 222 |
|
✗ |
control_->multiplyByJacobian(ui_data, ki_data->Fu, |
| 223 |
|
✗ |
d->dki_du[i].bottomRows(nv), |
| 224 |
|
|
addto); // dfi_du = dki_dw * dw_du |
| 225 |
|
|
|
| 226 |
|
✗ |
d->dli_dx[i].noalias() = ki_data->Lx.transpose() * d->dyi_dx[i]; |
| 227 |
|
✗ |
control_->multiplyJacobianTransposeBy(ui_data, ki_data->Lu, |
| 228 |
|
✗ |
d->dli_du[i]); // dli_du = Lu * dw_du |
| 229 |
|
✗ |
d->dli_du[i].noalias() += ki_data->Lx.transpose() * d->dyi_du[i]; |
| 230 |
|
|
|
| 231 |
|
✗ |
d->Lxx_partialx[i].noalias() = ki_data->Lxx * d->dyi_dx[i]; |
| 232 |
|
✗ |
d->ddli_ddx[i].noalias() = d->dyi_dx[i].transpose() * d->Lxx_partialx[i]; |
| 233 |
|
|
|
| 234 |
|
✗ |
control_->multiplyByJacobian(ui_data, ki_data->Lxu, |
| 235 |
|
✗ |
d->Lxu_i[i]); // Lxu = Lxw * dw_du |
| 236 |
|
✗ |
d->Luu_partialx[i].noalias() = d->Lxu_i[i].transpose() * d->dyi_du[i]; |
| 237 |
|
✗ |
d->Lxx_partialu[i].noalias() = ki_data->Lxx * d->dyi_du[i]; |
| 238 |
|
✗ |
control_->multiplyByJacobian( |
| 239 |
|
✗ |
ui_data, ki_data->Luu, |
| 240 |
|
✗ |
d->ddli_dwdu[i]); // ddli_dwdu = ddli_ddw * dw_du |
| 241 |
|
✗ |
control_->multiplyJacobianTransposeBy( |
| 242 |
|
✗ |
ui_data, d->ddli_dwdu[i], |
| 243 |
|
✗ |
d->ddli_ddu[i]); // ddli_ddu = dw_du.T * ddli_dwdu |
| 244 |
|
✗ |
d->ddli_ddu[i].noalias() += d->Luu_partialx[i].transpose() + |
| 245 |
|
✗ |
d->Luu_partialx[i] + |
| 246 |
|
✗ |
d->dyi_du[i].transpose() * d->Lxx_partialu[i]; |
| 247 |
|
|
|
| 248 |
|
✗ |
d->ddli_dxdw[i].noalias() = d->dyi_dx[i].transpose() * ki_data->Lxu; |
| 249 |
|
✗ |
control_->multiplyByJacobian( |
| 250 |
|
✗ |
ui_data, d->ddli_dxdw[i], |
| 251 |
|
✗ |
d->ddli_dxdu[i]); // ddli_dxdu = ddli_dxdw * dw_du |
| 252 |
|
✗ |
d->ddli_dxdu[i].noalias() += d->dyi_dx[i].transpose() * d->Lxx_partialu[i]; |
| 253 |
|
|
} |
| 254 |
|
|
|
| 255 |
|
✗ |
if (ni_ == 2) { |
| 256 |
|
✗ |
d->Fx.noalias() = time_step_ * d->dki_dx[1]; |
| 257 |
|
✗ |
d->Fu.noalias() = time_step_ * d->dki_du[1]; |
| 258 |
|
✗ |
d->Lx.noalias() = time_step_ * d->dli_dx[1]; |
| 259 |
|
✗ |
d->Lu.noalias() = time_step_ * d->dli_du[1]; |
| 260 |
|
✗ |
d->Lxx.noalias() = time_step_ * d->ddli_ddx[1]; |
| 261 |
|
✗ |
d->Luu.noalias() = time_step_ * d->ddli_ddu[1]; |
| 262 |
|
✗ |
d->Lxu.noalias() = time_step_ * d->ddli_dxdu[1]; |
| 263 |
|
✗ |
} else if (ni_ == 3) { |
| 264 |
|
✗ |
d->Fx.noalias() = |
| 265 |
|
✗ |
time_step_ / Scalar(4.) * (d->dki_dx[0] + Scalar(3.) * d->dki_dx[2]); |
| 266 |
|
✗ |
d->Fu.noalias() = |
| 267 |
|
✗ |
time_step_ / Scalar(4.) * (d->dki_du[0] + Scalar(3.) * d->dki_du[2]); |
| 268 |
|
✗ |
d->Lx.noalias() = |
| 269 |
|
✗ |
time_step_ / Scalar(4.) * (d->dli_dx[0] + Scalar(3.) * d->dli_dx[2]); |
| 270 |
|
✗ |
d->Lu.noalias() = |
| 271 |
|
✗ |
time_step_ / Scalar(4.) * (d->dli_du[0] + Scalar(3.) * d->dli_du[2]); |
| 272 |
|
✗ |
d->Lxx.noalias() = time_step_ / Scalar(4.) * |
| 273 |
|
✗ |
(d->ddli_ddx[0] + Scalar(3.) * d->ddli_ddx[2]); |
| 274 |
|
✗ |
d->Luu.noalias() = time_step_ / Scalar(4.) * |
| 275 |
|
✗ |
(d->ddli_ddu[0] + Scalar(3.) * d->ddli_ddu[2]); |
| 276 |
|
✗ |
d->Lxu.noalias() = time_step_ / Scalar(4.) * |
| 277 |
|
✗ |
(d->ddli_dxdu[0] + Scalar(3.) * d->ddli_dxdu[2]); |
| 278 |
|
|
} else { |
| 279 |
|
✗ |
d->Fx.noalias() = time_step_ / Scalar(6.) * |
| 280 |
|
✗ |
(d->dki_dx[0] + Scalar(2.) * d->dki_dx[1] + |
| 281 |
|
✗ |
Scalar(2.) * d->dki_dx[2] + d->dki_dx[3]); |
| 282 |
|
✗ |
d->Fu.noalias() = time_step_ / Scalar(6.) * |
| 283 |
|
✗ |
(d->dki_du[0] + Scalar(2.) * d->dki_du[1] + |
| 284 |
|
✗ |
Scalar(2.) * d->dki_du[2] + d->dki_du[3]); |
| 285 |
|
✗ |
d->Lx.noalias() = time_step_ / Scalar(6.) * |
| 286 |
|
✗ |
(d->dli_dx[0] + Scalar(2.) * d->dli_dx[1] + |
| 287 |
|
✗ |
Scalar(2.) * d->dli_dx[2] + d->dli_dx[3]); |
| 288 |
|
✗ |
d->Lu.noalias() = time_step_ / Scalar(6.) * |
| 289 |
|
✗ |
(d->dli_du[0] + Scalar(2.) * d->dli_du[1] + |
| 290 |
|
✗ |
Scalar(2.) * d->dli_du[2] + d->dli_du[3]); |
| 291 |
|
✗ |
d->Lxx.noalias() = time_step_ / Scalar(6.) * |
| 292 |
|
✗ |
(d->ddli_ddx[0] + Scalar(2.) * d->ddli_ddx[1] + |
| 293 |
|
✗ |
Scalar(2.) * d->ddli_ddx[2] + d->ddli_ddx[3]); |
| 294 |
|
✗ |
d->Luu.noalias() = time_step_ / Scalar(6.) * |
| 295 |
|
✗ |
(d->ddli_ddu[0] + Scalar(2.) * d->ddli_ddu[1] + |
| 296 |
|
✗ |
Scalar(2.) * d->ddli_ddu[2] + d->ddli_ddu[3]); |
| 297 |
|
✗ |
d->Lxu.noalias() = time_step_ / Scalar(6.) * |
| 298 |
|
✗ |
(d->ddli_dxdu[0] + Scalar(2.) * d->ddli_dxdu[1] + |
| 299 |
|
✗ |
Scalar(2.) * d->ddli_dxdu[2] + d->ddli_dxdu[3]); |
| 300 |
|
|
} |
| 301 |
|
✗ |
d->Gx = k0_data->Gx; |
| 302 |
|
✗ |
d->Hx = k0_data->Hx; |
| 303 |
|
✗ |
d->Gu.conservativeResize(differential_->get_ng(), nu_); |
| 304 |
|
✗ |
d->Hu.conservativeResize(differential_->get_nh(), nu_); |
| 305 |
|
✗ |
control_->multiplyByJacobian(u0_data, k0_data->Gu, d->Gu); |
| 306 |
|
✗ |
control_->multiplyByJacobian(u0_data, k0_data->Hu, d->Hu); |
| 307 |
|
|
|
| 308 |
|
✗ |
state_->JintegrateTransport(x, d->dx, d->Fx, second); |
| 309 |
|
✗ |
state_->Jintegrate(x, d->dx, d->Fx, d->Fx, first, addto); |
| 310 |
|
✗ |
state_->JintegrateTransport(x, d->dx, d->Fu, second); |
| 311 |
|
|
} |
| 312 |
|
|
|
| 313 |
|
|
template <typename Scalar> |
| 314 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::calcDiff( |
| 315 |
|
|
const std::shared_ptr<ActionDataAbstract>& data, |
| 316 |
|
|
const Eigen::Ref<const VectorXs>& x) { |
| 317 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 318 |
|
✗ |
throw_pretty( |
| 319 |
|
|
"Invalid argument: " << "x has wrong dimension (it should be " + |
| 320 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 321 |
|
|
} |
| 322 |
|
✗ |
Data* d = static_cast<Data*>(data.get()); |
| 323 |
|
|
|
| 324 |
|
|
const std::shared_ptr<DifferentialActionDataAbstract>& k0_data = |
| 325 |
|
✗ |
d->differential[0]; |
| 326 |
|
✗ |
differential_->calcDiff(k0_data, x); |
| 327 |
|
✗ |
d->Lx = k0_data->Lx; |
| 328 |
|
✗ |
d->Lxx = k0_data->Lxx; |
| 329 |
|
✗ |
d->Gx = k0_data->Gx; |
| 330 |
|
✗ |
d->Hx = k0_data->Hx; |
| 331 |
|
|
} |
| 332 |
|
|
|
| 333 |
|
|
template <typename Scalar> |
| 334 |
|
|
std::shared_ptr<ActionDataAbstractTpl<Scalar> > |
| 335 |
|
✗ |
IntegratedActionModelRKTpl<Scalar>::createData() { |
| 336 |
|
✗ |
return std::allocate_shared<Data>(Eigen::aligned_allocator<Data>(), this); |
| 337 |
|
|
} |
| 338 |
|
|
|
| 339 |
|
|
template <typename Scalar> |
| 340 |
|
|
template <typename NewScalar> |
| 341 |
|
✗ |
IntegratedActionModelRKTpl<NewScalar> IntegratedActionModelRKTpl<Scalar>::cast() |
| 342 |
|
|
const { |
| 343 |
|
|
typedef IntegratedActionModelRKTpl<NewScalar> ReturnType; |
| 344 |
|
✗ |
if (control_) { |
| 345 |
|
✗ |
ReturnType ret(differential_->template cast<NewScalar>(), |
| 346 |
|
✗ |
control_->template cast<NewScalar>(), rk_type_, |
| 347 |
|
✗ |
scalar_cast<NewScalar>(time_step_), with_cost_residual_); |
| 348 |
|
✗ |
return ret; |
| 349 |
|
✗ |
} else { |
| 350 |
|
✗ |
ReturnType ret(differential_->template cast<NewScalar>(), rk_type_, |
| 351 |
|
✗ |
scalar_cast<NewScalar>(time_step_), with_cost_residual_); |
| 352 |
|
✗ |
return ret; |
| 353 |
|
|
} |
| 354 |
|
|
} |
| 355 |
|
|
|
| 356 |
|
|
template <typename Scalar> |
| 357 |
|
✗ |
bool IntegratedActionModelRKTpl<Scalar>::checkData( |
| 358 |
|
|
const std::shared_ptr<ActionDataAbstract>& data) { |
| 359 |
|
✗ |
std::shared_ptr<Data> d = std::dynamic_pointer_cast<Data>(data); |
| 360 |
|
✗ |
if (data != NULL) { |
| 361 |
|
✗ |
for (std::size_t i = 0; i < ni_; ++i) { |
| 362 |
|
✗ |
if (!differential_->checkData(d->differential[i])) { |
| 363 |
|
✗ |
return false; |
| 364 |
|
|
} |
| 365 |
|
|
} |
| 366 |
|
✗ |
return true; |
| 367 |
|
|
} else { |
| 368 |
|
✗ |
return false; |
| 369 |
|
|
} |
| 370 |
|
|
} |
| 371 |
|
|
|
| 372 |
|
|
template <typename Scalar> |
| 373 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::quasiStatic( |
| 374 |
|
|
const std::shared_ptr<ActionDataAbstract>& data, Eigen::Ref<VectorXs> u, |
| 375 |
|
|
const Eigen::Ref<const VectorXs>& x, const std::size_t maxiter, |
| 376 |
|
|
const Scalar tol) { |
| 377 |
|
✗ |
if (static_cast<std::size_t>(u.size()) != nu_) { |
| 378 |
|
✗ |
throw_pretty( |
| 379 |
|
|
"Invalid argument: " << "u has wrong dimension (it should be " + |
| 380 |
|
|
std::to_string(nu_) + ")"); |
| 381 |
|
|
} |
| 382 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 383 |
|
✗ |
throw_pretty( |
| 384 |
|
|
"Invalid argument: " << "x has wrong dimension (it should be " + |
| 385 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 386 |
|
|
} |
| 387 |
|
|
|
| 388 |
|
✗ |
Data* d = static_cast<Data*>(data.get()); |
| 389 |
|
|
const std::shared_ptr<ControlParametrizationDataAbstract>& u0_data = |
| 390 |
|
✗ |
d->control[0]; |
| 391 |
|
✗ |
u0_data->w *= Scalar(0.); |
| 392 |
|
✗ |
differential_->quasiStatic(d->differential[0], u0_data->w, x, maxiter, tol); |
| 393 |
|
✗ |
control_->params(u0_data, Scalar(0.), u0_data->w); |
| 394 |
|
✗ |
u = u0_data->u; |
| 395 |
|
|
} |
| 396 |
|
|
|
| 397 |
|
|
template <typename Scalar> |
| 398 |
|
✗ |
std::size_t IntegratedActionModelRKTpl<Scalar>::get_ni() const { |
| 399 |
|
✗ |
return ni_; |
| 400 |
|
|
} |
| 401 |
|
|
|
| 402 |
|
|
template <typename Scalar> |
| 403 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::print(std::ostream& os) const { |
| 404 |
|
✗ |
os << "IntegratedActionModelRK {dt=" << time_step_ << ", " << *differential_ |
| 405 |
|
✗ |
<< "}"; |
| 406 |
|
|
} |
| 407 |
|
|
|
| 408 |
|
|
template <typename Scalar> |
| 409 |
|
✗ |
void IntegratedActionModelRKTpl<Scalar>::set_rk_type(const RKType rktype) { |
| 410 |
|
✗ |
switch (rktype) { |
| 411 |
|
✗ |
case two: |
| 412 |
|
✗ |
ni_ = 2; |
| 413 |
|
✗ |
rk_c_.resize(ni_); |
| 414 |
|
✗ |
rk_c_[0] = Scalar(0.); |
| 415 |
|
✗ |
rk_c_[1] = Scalar(0.5); |
| 416 |
|
✗ |
break; |
| 417 |
|
✗ |
case three: |
| 418 |
|
✗ |
ni_ = 3; |
| 419 |
|
✗ |
rk_c_.resize(ni_); |
| 420 |
|
✗ |
rk_c_[0] = Scalar(0.); |
| 421 |
|
✗ |
rk_c_[1] = Scalar(1. / 3.); |
| 422 |
|
✗ |
rk_c_[2] = Scalar(2. / 3.); |
| 423 |
|
✗ |
break; |
| 424 |
|
✗ |
case four: |
| 425 |
|
✗ |
ni_ = 4; |
| 426 |
|
✗ |
rk_c_.resize(ni_); |
| 427 |
|
✗ |
rk_c_[0] = Scalar(0.); |
| 428 |
|
✗ |
rk_c_[1] = Scalar(0.5); |
| 429 |
|
✗ |
rk_c_[2] = Scalar(0.5); |
| 430 |
|
✗ |
rk_c_[3] = Scalar(1.); |
| 431 |
|
✗ |
break; |
| 432 |
|
|
} |
| 433 |
|
|
} |
| 434 |
|
|
|
| 435 |
|
|
} // namespace crocoddyl |
| 436 |
|
|
|