| Line |
Branch |
Exec |
Source |
| 1 |
|
|
#ifndef __quadruped_walkgen_quadruped_time_hxx__ |
| 2 |
|
|
#define __quadruped_walkgen_quadruped_time_hxx__ |
| 3 |
|
|
|
| 4 |
|
|
#include "crocoddyl/core/utils/exception.hpp" |
| 5 |
|
|
|
| 6 |
|
|
namespace quadruped_walkgen { |
| 7 |
|
|
template <typename Scalar> |
| 8 |
|
✗ |
ActionModelQuadrupedTimeTpl<Scalar>::ActionModelQuadrupedTimeTpl() |
| 9 |
|
|
: crocoddyl::ActionModelAbstractTpl<Scalar>( |
| 10 |
|
✗ |
boost::make_shared<crocoddyl::StateVectorTpl<Scalar> >(21), 1, 22) { |
| 11 |
|
|
// 4 costs |
| 12 |
|
✗ |
state_weights_ << Scalar(1.), Scalar(1.), Scalar(150.), Scalar(35.), |
| 13 |
|
✗ |
Scalar(30.), Scalar(8.), Scalar(20.), Scalar(20.), Scalar(15.), |
| 14 |
|
✗ |
Scalar(4.), Scalar(4.), Scalar(8.); |
| 15 |
|
✗ |
heuristic_weights_.setConstant(Scalar(1)); |
| 16 |
|
|
|
| 17 |
|
✗ |
dt_bound_weight_cmd = Scalar(100.); |
| 18 |
|
✗ |
dt_weight_cmd = Scalar(0.); |
| 19 |
|
|
|
| 20 |
|
✗ |
pheuristic_.setZero(); |
| 21 |
|
✗ |
gait_double_.setZero(); |
| 22 |
|
|
|
| 23 |
|
|
// Shoulder heuristic position |
| 24 |
|
|
// pshoulder_ << Scalar(0.1946) , Scalar(0.15005), Scalar(0.1946) , |
| 25 |
|
|
// Scalar(-0.15005) , |
| 26 |
|
|
// Scalar(-0.1946), Scalar(0.15005) , Scalar(-0.1946), |
| 27 |
|
|
// Scalar(-0.15005) ; |
| 28 |
|
|
// pshoulder_0 << Scalar(0.1946) , Scalar(0.1946) , Scalar(-0.1946), |
| 29 |
|
|
// Scalar(-0.1946) , |
| 30 |
|
|
// Scalar(0.15005) , Scalar(-0.15005) , Scalar(0.15005) , |
| 31 |
|
|
// Scalar(-0.15005) ; |
| 32 |
|
|
// pshoulder_tmp.setZero() ; |
| 33 |
|
|
// pcentrifugal_tmp_1.setZero() ; |
| 34 |
|
|
// pcentrifugal_tmp_2.setZero() ; |
| 35 |
|
|
// pcentrifugal_tmp.setZero() ; |
| 36 |
|
✗ |
centrifugal_term = true; |
| 37 |
|
✗ |
symmetry_term = true; |
| 38 |
|
✗ |
T_gait = Scalar(0.64); |
| 39 |
|
|
// R_tmp.setZero() ; |
| 40 |
|
|
|
| 41 |
|
|
// Cost relative to the command |
| 42 |
|
✗ |
rub_max_.setZero(); |
| 43 |
|
✗ |
rub_max_bool.setZero(); |
| 44 |
|
✗ |
dt_ref_.setConstant(Scalar(0.02)); |
| 45 |
|
✗ |
dt_min_.setConstant(Scalar(0.005)); |
| 46 |
|
✗ |
dt_max_.setConstant(Scalar(0.1)); |
| 47 |
|
|
|
| 48 |
|
|
// Log cost |
| 49 |
|
✗ |
cost_.setZero(); |
| 50 |
|
✗ |
log_cost = true; |
| 51 |
|
|
} |
| 52 |
|
|
|
| 53 |
|
|
template <typename Scalar> |
| 54 |
|
✗ |
ActionModelQuadrupedTimeTpl<Scalar>::~ActionModelQuadrupedTimeTpl() {} |
| 55 |
|
|
|
| 56 |
|
|
template <typename Scalar> |
| 57 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::calc( |
| 58 |
|
|
const boost::shared_ptr<crocoddyl::ActionDataAbstractTpl<Scalar> >& data, |
| 59 |
|
|
const Eigen::Ref<const typename MathBase::VectorXs>& x, |
| 60 |
|
|
const Eigen::Ref<const typename MathBase::VectorXs>& u) { |
| 61 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 62 |
|
✗ |
throw_pretty("Invalid argument: " |
| 63 |
|
|
<< "x has wrong dimension (it should be " + |
| 64 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 65 |
|
|
} |
| 66 |
|
✗ |
if (static_cast<std::size_t>(u.size()) != nu_) { |
| 67 |
|
✗ |
throw_pretty("Invalid argument: " |
| 68 |
|
|
<< "u has wrong dimension (it should be " + |
| 69 |
|
|
std::to_string(nu_) + ")"); |
| 70 |
|
|
} |
| 71 |
|
|
|
| 72 |
|
|
ActionDataQuadrupedTimeTpl<Scalar>* d = |
| 73 |
|
✗ |
static_cast<ActionDataQuadrupedTimeTpl<Scalar>*>(data.get()); |
| 74 |
|
|
|
| 75 |
|
✗ |
d->xnext.template head<12>() = x.head(12); |
| 76 |
|
✗ |
d->xnext.template segment<8>(12) = x.segment(12, 8); |
| 77 |
|
✗ |
d->xnext.template tail<1>() = u.cwiseAbs(); |
| 78 |
|
|
|
| 79 |
|
|
// Residual cost on the state and force norm |
| 80 |
|
|
// State : delta*||X-Xref|| |
| 81 |
|
✗ |
d->r.template head<12>() = state_weights_.cwiseProduct(x.head(12) - xref_); |
| 82 |
|
|
// Feet placement : delta*||P-Pref|| |
| 83 |
|
✗ |
d->r.template segment<8>(12) = |
| 84 |
|
✗ |
((heuristic_weights_.cwiseProduct(x.segment(12, 8) - pheuristic_)) |
| 85 |
|
|
.array() * |
| 86 |
|
✗ |
gait_double_.array()) |
| 87 |
|
|
.matrix(); |
| 88 |
|
|
// Dt command, used to fix the optimisation to dt_ref_ |
| 89 |
|
✗ |
d->r.template tail<1>() << dt_weight_cmd * (u.cwiseAbs() - dt_ref_); |
| 90 |
|
|
|
| 91 |
|
|
// Penalisation if dt out of lower/upper bound |
| 92 |
|
✗ |
rub_max_ << dt_min_ - u.cwiseAbs(), u.cwiseAbs() - dt_max_; |
| 93 |
|
✗ |
rub_max_bool = |
| 94 |
|
✗ |
(rub_max_.array() >= Scalar(0.)).matrix().template cast<Scalar>(); |
| 95 |
|
✗ |
rub_max_ = rub_max_.cwiseMax(Scalar(0.)); |
| 96 |
|
|
|
| 97 |
|
✗ |
d->cost = Scalar(0.5) * d->r.transpose() * d->r + |
| 98 |
|
✗ |
dt_bound_weight_cmd * Scalar(0.5) * rub_max_.squaredNorm(); |
| 99 |
|
|
|
| 100 |
|
✗ |
if (log_cost) { |
| 101 |
|
|
// Length of the cost similar to the Augmented cost... |
| 102 |
|
✗ |
cost_[0] = |
| 103 |
|
✗ |
Scalar(0.5) * d->r.head(12).transpose() * d->r.head(12); // state |
| 104 |
|
✗ |
cost_[1] = Scalar(0.5) * d->r.segment(12, 8).transpose() * |
| 105 |
|
✗ |
d->r.segment(12, 8); // feet placement |
| 106 |
|
✗ |
cost_[2] = Scalar(0.5) * d->r.tail(1).transpose() * |
| 107 |
|
✗ |
d->r.tail(1); // 0.5*\delta*|| U - dt_ref ||^2 |
| 108 |
|
✗ |
cost_[3] = dt_bound_weight_cmd * Scalar(0.5) * |
| 109 |
|
✗ |
rub_max_.squaredNorm(); // u/l bound weight |
| 110 |
|
✗ |
cost_[4] = Scalar(0); |
| 111 |
|
✗ |
cost_[5] = Scalar(0); |
| 112 |
|
✗ |
cost_[6] = Scalar(0); |
| 113 |
|
|
} |
| 114 |
|
|
} |
| 115 |
|
|
|
| 116 |
|
|
template <typename Scalar> |
| 117 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::calcDiff( |
| 118 |
|
|
const boost::shared_ptr<crocoddyl::ActionDataAbstractTpl<Scalar> >& data, |
| 119 |
|
|
const Eigen::Ref<const typename MathBase::VectorXs>& x, |
| 120 |
|
|
const Eigen::Ref<const typename MathBase::VectorXs>& u) { |
| 121 |
|
✗ |
if (static_cast<std::size_t>(x.size()) != state_->get_nx()) { |
| 122 |
|
✗ |
throw_pretty("Invalid argument: " |
| 123 |
|
|
<< "x has wrong dimension (it should be " + |
| 124 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 125 |
|
|
} |
| 126 |
|
✗ |
if (static_cast<std::size_t>(u.size()) != nu_) { |
| 127 |
|
✗ |
throw_pretty("Invalid argument: " |
| 128 |
|
|
<< "u has wrong dimension (it should be " + |
| 129 |
|
|
std::to_string(nu_) + ")"); |
| 130 |
|
|
} |
| 131 |
|
|
|
| 132 |
|
|
ActionDataQuadrupedTimeTpl<Scalar>* d = |
| 133 |
|
✗ |
static_cast<ActionDataQuadrupedTimeTpl<Scalar>*>(data.get()); |
| 134 |
|
|
|
| 135 |
|
|
// Cost derivatives : Lx |
| 136 |
|
✗ |
d->Lx.template head<12>() = |
| 137 |
|
✗ |
(state_weights_.array() * d->r.template head<12>().array()).matrix(); |
| 138 |
|
✗ |
d->Lx.template segment<8>(12) = |
| 139 |
|
✗ |
(heuristic_weights_.array() * d->r.template segment<8>(12).array()) |
| 140 |
|
|
.matrix(); // * gait_double in d->r |
| 141 |
|
|
|
| 142 |
|
✗ |
d->Lu << dt_bound_weight_cmd * std::copysign(1., u(0)) * |
| 143 |
|
✗ |
(-rub_max_[0] + rub_max_[1]); |
| 144 |
|
✗ |
d->Lu += dt_weight_cmd * std::copysign(1., u(0)) * d->r.template tail<1>(); |
| 145 |
|
|
|
| 146 |
|
|
// Hessian : Lxx |
| 147 |
|
✗ |
d->Lxx.diagonal().head(12) = |
| 148 |
|
✗ |
(state_weights_.array() * state_weights_.array()).matrix(); |
| 149 |
|
✗ |
d->Lxx.diagonal().segment(12, 8) = |
| 150 |
|
✗ |
(gait_double_.array() * heuristic_weights_.array() * |
| 151 |
|
✗ |
heuristic_weights_.array()) |
| 152 |
|
|
.matrix(); |
| 153 |
|
|
|
| 154 |
|
✗ |
d->Luu.diagonal() << dt_weight_cmd * dt_weight_cmd + |
| 155 |
|
✗ |
dt_bound_weight_cmd * rub_max_bool[0] + |
| 156 |
|
✗ |
dt_bound_weight_cmd * rub_max_bool[1]; |
| 157 |
|
|
|
| 158 |
|
|
// Dynamic derivatives |
| 159 |
|
✗ |
d->Fx.setIdentity(); |
| 160 |
|
✗ |
d->Fx(20, 20) = Scalar(0.); |
| 161 |
|
✗ |
d->Fu.block(20, 0, 1, 1) << std::copysign(1., u(0)); |
| 162 |
|
|
} |
| 163 |
|
|
|
| 164 |
|
|
template <typename Scalar> |
| 165 |
|
|
boost::shared_ptr<crocoddyl::ActionDataAbstractTpl<Scalar> > |
| 166 |
|
✗ |
ActionModelQuadrupedTimeTpl<Scalar>::createData() { |
| 167 |
|
✗ |
return boost::make_shared<ActionDataQuadrupedTimeTpl<Scalar> >(this); |
| 168 |
|
|
} |
| 169 |
|
|
|
| 170 |
|
|
//////////////////////////////// |
| 171 |
|
|
// get & set state weight ////// |
| 172 |
|
|
//////////////////////////////// |
| 173 |
|
|
|
| 174 |
|
|
template <typename Scalar> |
| 175 |
|
|
const typename Eigen::Matrix<Scalar, 12, 1>& |
| 176 |
|
✗ |
ActionModelQuadrupedTimeTpl<Scalar>::get_state_weights() const { |
| 177 |
|
✗ |
return state_weights_; |
| 178 |
|
|
} |
| 179 |
|
|
template <typename Scalar> |
| 180 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_state_weights( |
| 181 |
|
|
const typename MathBase::VectorXs& weights) { |
| 182 |
|
✗ |
if (static_cast<std::size_t>(weights.size()) != 12) { |
| 183 |
|
✗ |
throw_pretty("Invalid argument: " |
| 184 |
|
|
<< "Weights vector has wrong dimension (it should be 12)"); |
| 185 |
|
|
} |
| 186 |
|
✗ |
state_weights_ = weights; |
| 187 |
|
|
} |
| 188 |
|
|
|
| 189 |
|
|
template <typename Scalar> |
| 190 |
|
|
const typename Eigen::Matrix<Scalar, 8, 1>& |
| 191 |
|
✗ |
ActionModelQuadrupedTimeTpl<Scalar>::get_heuristic_weights() const { |
| 192 |
|
✗ |
return heuristic_weights_; |
| 193 |
|
|
} |
| 194 |
|
|
template <typename Scalar> |
| 195 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_heuristic_weights( |
| 196 |
|
|
const typename MathBase::VectorXs& weights) { |
| 197 |
|
✗ |
if (static_cast<std::size_t>(weights.size()) != 8) { |
| 198 |
|
✗ |
throw_pretty("Invalid argument: " |
| 199 |
|
|
<< "Weights vector has wrong dimension (it should be 8)"); |
| 200 |
|
|
} |
| 201 |
|
✗ |
heuristic_weights_ = weights; |
| 202 |
|
|
} |
| 203 |
|
|
|
| 204 |
|
|
////////////////////////////////////////////////////////// |
| 205 |
|
|
// Param relative to the heuristic position // |
| 206 |
|
|
////////////////////////////////////////////////////////// |
| 207 |
|
|
|
| 208 |
|
|
template <typename Scalar> |
| 209 |
|
✗ |
const bool& ActionModelQuadrupedTimeTpl<Scalar>::get_symmetry_term() const { |
| 210 |
|
✗ |
return symmetry_term; |
| 211 |
|
|
} |
| 212 |
|
|
template <typename Scalar> |
| 213 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_symmetry_term( |
| 214 |
|
|
const bool& sym_term) { |
| 215 |
|
|
// The model need to be updated after this changed |
| 216 |
|
✗ |
symmetry_term = sym_term; |
| 217 |
|
|
} |
| 218 |
|
|
|
| 219 |
|
|
template <typename Scalar> |
| 220 |
|
✗ |
const bool& ActionModelQuadrupedTimeTpl<Scalar>::get_centrifugal_term() const { |
| 221 |
|
✗ |
return centrifugal_term; |
| 222 |
|
|
} |
| 223 |
|
|
template <typename Scalar> |
| 224 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_centrifugal_term( |
| 225 |
|
|
const bool& cent_term) { |
| 226 |
|
|
// The model need to be updated after this changed |
| 227 |
|
✗ |
centrifugal_term = cent_term; |
| 228 |
|
|
} |
| 229 |
|
|
|
| 230 |
|
|
template <typename Scalar> |
| 231 |
|
✗ |
const Scalar& ActionModelQuadrupedTimeTpl<Scalar>::get_T_gait() const { |
| 232 |
|
|
// The model need to be updated after this change |
| 233 |
|
✗ |
return T_gait; |
| 234 |
|
|
} |
| 235 |
|
|
template <typename Scalar> |
| 236 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_T_gait(const Scalar& T_gait_) { |
| 237 |
|
|
// The model need to be updated after this change |
| 238 |
|
✗ |
T_gait = T_gait_; |
| 239 |
|
|
} |
| 240 |
|
|
|
| 241 |
|
|
// dt_ref to fix the dt optimised |
| 242 |
|
|
template <typename Scalar> |
| 243 |
|
✗ |
const Scalar& ActionModelQuadrupedTimeTpl<Scalar>::get_dt_ref() const { |
| 244 |
|
✗ |
return dt_ref_[0]; |
| 245 |
|
|
} |
| 246 |
|
|
template <typename Scalar> |
| 247 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_dt_ref(const Scalar& dt) { |
| 248 |
|
✗ |
dt_ref_[0] = dt; |
| 249 |
|
|
} |
| 250 |
|
|
|
| 251 |
|
|
////////////////////////////////////////////////////////// |
| 252 |
|
|
// Upper and lower bound dt // |
| 253 |
|
|
////////////////////////////////////////////////////////// |
| 254 |
|
|
template <typename Scalar> |
| 255 |
|
✗ |
const Scalar& ActionModelQuadrupedTimeTpl<Scalar>::get_dt_min() const { |
| 256 |
|
✗ |
return dt_min_[0]; |
| 257 |
|
|
} |
| 258 |
|
|
template <typename Scalar> |
| 259 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_dt_min(const Scalar& dt) { |
| 260 |
|
✗ |
dt_min_[0] = dt; |
| 261 |
|
|
} |
| 262 |
|
|
template <typename Scalar> |
| 263 |
|
✗ |
const Scalar& ActionModelQuadrupedTimeTpl<Scalar>::get_dt_max() const { |
| 264 |
|
✗ |
return dt_max_[0]; |
| 265 |
|
|
} |
| 266 |
|
|
template <typename Scalar> |
| 267 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_dt_max(const Scalar& dt) { |
| 268 |
|
✗ |
dt_max_[0] = dt; |
| 269 |
|
|
} |
| 270 |
|
|
///////////////////////////////////////////////////////////// |
| 271 |
|
|
// Get access and modify to the upper/lower bound for dt // |
| 272 |
|
|
///////////////////////////////////////////////////////////// |
| 273 |
|
|
template <typename Scalar> |
| 274 |
|
✗ |
const Scalar& ActionModelQuadrupedTimeTpl<Scalar>::get_dt_bound_weight_cmd() |
| 275 |
|
|
const { |
| 276 |
|
✗ |
return dt_bound_weight_cmd; |
| 277 |
|
|
} |
| 278 |
|
|
template <typename Scalar> |
| 279 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_dt_bound_weight_cmd( |
| 280 |
|
|
const Scalar& weight_) { |
| 281 |
|
✗ |
dt_bound_weight_cmd = weight_; |
| 282 |
|
|
} |
| 283 |
|
|
|
| 284 |
|
|
/////////////////////////////////////////////////////////////// |
| 285 |
|
|
// Get access and modify to the \weight*||U-dt_ref|| weight // |
| 286 |
|
|
/////////////////////////////////////////////////////////////// |
| 287 |
|
|
template <typename Scalar> |
| 288 |
|
✗ |
const Scalar& ActionModelQuadrupedTimeTpl<Scalar>::get_dt_weight_cmd() const { |
| 289 |
|
✗ |
return dt_weight_cmd; |
| 290 |
|
|
} |
| 291 |
|
|
template <typename Scalar> |
| 292 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::set_dt_weight_cmd( |
| 293 |
|
|
const Scalar& weight_) { |
| 294 |
|
✗ |
dt_weight_cmd = weight_; |
| 295 |
|
|
} |
| 296 |
|
|
|
| 297 |
|
|
/////////////////////// |
| 298 |
|
|
// Logging cost // |
| 299 |
|
|
/////////////////////// |
| 300 |
|
|
template <typename Scalar> |
| 301 |
|
|
const typename Eigen::Matrix<Scalar, 7, 1>& |
| 302 |
|
✗ |
ActionModelQuadrupedTimeTpl<Scalar>::get_cost() const { |
| 303 |
|
✗ |
return cost_; |
| 304 |
|
|
} |
| 305 |
|
|
|
| 306 |
|
|
//////////////////////// |
| 307 |
|
|
// Update current model |
| 308 |
|
|
//////////////////////// |
| 309 |
|
|
|
| 310 |
|
|
template <typename Scalar> |
| 311 |
|
✗ |
void ActionModelQuadrupedTimeTpl<Scalar>::update_model( |
| 312 |
|
|
const Eigen::Ref<const typename MathBase::MatrixXs>& l_feet, |
| 313 |
|
|
const Eigen::Ref<const typename MathBase::MatrixXs>& xref, |
| 314 |
|
|
const Eigen::Ref<const typename MathBase::VectorXs>& S) { |
| 315 |
|
✗ |
if (static_cast<std::size_t>(l_feet.size()) != 12) { |
| 316 |
|
✗ |
throw_pretty("Invalid argument: " |
| 317 |
|
|
<< "l_feet matrix has wrong dimension (it should be : 3x4)"); |
| 318 |
|
|
} |
| 319 |
|
✗ |
if (static_cast<std::size_t>(xref.size()) != 12) { |
| 320 |
|
✗ |
throw_pretty("Invalid argument: " |
| 321 |
|
|
<< "Weights vector has wrong dimension (it should be " + |
| 322 |
|
|
std::to_string(state_->get_nx()) + ")"); |
| 323 |
|
|
} |
| 324 |
|
✗ |
if (static_cast<std::size_t>(S.size()) != 4) { |
| 325 |
|
✗ |
throw_pretty("Invalid argument: " |
| 326 |
|
|
<< "S vector has wrong dimension (it should be 4x1)"); |
| 327 |
|
|
} |
| 328 |
|
|
// l_feet and S useless, kept for now, to be consistent with others models |
| 329 |
|
|
|
| 330 |
|
✗ |
xref_ = xref; |
| 331 |
|
✗ |
for (int i = 0; i < 4; i = i + 1) { |
| 332 |
|
✗ |
gait_double_[2 * i] = S[i]; |
| 333 |
|
✗ |
gait_double_[2 * i + 1] = S[i]; |
| 334 |
|
✗ |
pheuristic_[2 * i] = l_feet(0, i); |
| 335 |
|
✗ |
pheuristic_[2 * i + 1] = l_feet(1, i); |
| 336 |
|
|
} |
| 337 |
|
|
|
| 338 |
|
|
// To compute heuristic here |
| 339 |
|
|
// R_tmp << cos(xref(5,0)) ,-sin(xref(5,0)) , Scalar(0), |
| 340 |
|
|
// sin(xref(5,0)), cos(xref(5,0)), Scalar(0), |
| 341 |
|
|
// Scalar(0),Scalar(0),Scalar(1.0) ; |
| 342 |
|
|
|
| 343 |
|
|
// // Centrifual term |
| 344 |
|
|
// pcentrifugal_tmp_1 = xref.block(6,0,3,1) ; |
| 345 |
|
|
// pcentrifugal_tmp_2 = xref.block(9,0,3,1) ; |
| 346 |
|
|
// pcentrifugal_tmp = 0.5*std::sqrt(xref(2,0)/9.81) * |
| 347 |
|
|
// pcentrifugal_tmp_1.cross(pcentrifugal_tmp_2) ; |
| 348 |
|
|
|
| 349 |
|
|
// for (int i=0; i<4; i=i+1){ |
| 350 |
|
|
// pshoulder_tmp.block(0,i,2,1) = |
| 351 |
|
|
// R_tmp.block(0,0,2,2)*(pshoulder_0.block(0,i,2,1) + symmetry_term * |
| 352 |
|
|
// 0.25*T_gait*xref.block(6,0,2,1) + centrifugal_term * |
| 353 |
|
|
// pcentrifugal_tmp.block(0,0,2,1) ); pshoulder_[2*i] = pshoulder_tmp(0,i) + |
| 354 |
|
|
// xref(0,0); pshoulder_[2*i+1] = pshoulder_tmp(1,i) + xref(1,0); |
| 355 |
|
|
// } |
| 356 |
|
|
} |
| 357 |
|
|
} // namespace quadruped_walkgen |
| 358 |
|
|
|
| 359 |
|
|
#endif |
| 360 |
|
|
|