1 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
2 |
|
|
// BSD 3-Clause License |
3 |
|
|
// |
4 |
|
|
// Copyright (C) 2019-2024, LAAS-CNRS, University of Edinburgh, |
5 |
|
|
// New York University, Max Planck Gesellschaft |
6 |
|
|
// Heriot-Watt University |
7 |
|
|
// Copyright note valid unless otherwise stated in individual files. |
8 |
|
|
// All rights reserved. |
9 |
|
|
/////////////////////////////////////////////////////////////////////////////// |
10 |
|
|
|
11 |
|
|
#ifndef CROCODDYL_CORE_NUMDIFF_DIFF_ACTION_HPP_ |
12 |
|
|
#define CROCODDYL_CORE_NUMDIFF_DIFF_ACTION_HPP_ |
13 |
|
|
|
14 |
|
|
#include <iostream> |
15 |
|
|
#include <vector> |
16 |
|
|
|
17 |
|
|
#include "crocoddyl/core/diff-action-base.hpp" |
18 |
|
|
|
19 |
|
|
namespace crocoddyl { |
20 |
|
|
|
21 |
|
|
/** |
22 |
|
|
* @brief This class computes the numerical differentiation of a differential |
23 |
|
|
* action model. |
24 |
|
|
* |
25 |
|
|
* It computes the Jacobian of the cost, its residual and dynamics via numerical |
26 |
|
|
* differentiation. It considers that the action model owns a cost residual and |
27 |
|
|
* the cost is the square of this residual, i.e., |
28 |
|
|
* \f$\ell(\mathbf{x},\mathbf{u})=\frac{1}{2}\|\mathbf{r}(\mathbf{x},\mathbf{u})\|^2\f$, |
29 |
|
|
* where \f$\mathbf{r}(\mathbf{x},\mathbf{u})\f$ is the residual vector. The |
30 |
|
|
* Hessian is computed only through the Gauss-Newton approximation, i.e., |
31 |
|
|
* \f{eqnarray*}{ |
32 |
|
|
* \mathbf{\ell}_\mathbf{xx} &=& \mathbf{R_x}^T\mathbf{R_x} \\ |
33 |
|
|
* \mathbf{\ell}_\mathbf{uu} &=& \mathbf{R_u}^T\mathbf{R_u} \\ |
34 |
|
|
* \mathbf{\ell}_\mathbf{xu} &=& \mathbf{R_x}^T\mathbf{R_u} |
35 |
|
|
* \f} |
36 |
|
|
* where the Jacobians of the cost residuals are denoted by \f$\mathbf{R_x}\f$ |
37 |
|
|
* and \f$\mathbf{R_u}\f$. Note that this approximation ignores the tensor |
38 |
|
|
* products (e.g., \f$\mathbf{R_{xx}}\mathbf{r}\f$). |
39 |
|
|
* |
40 |
|
|
* Finally, in the case that the cost does not have a residual, we set the |
41 |
|
|
* Hessian to zero, i.e., \f$\mathbf{L_{xx}} = \mathbf{L_{xu}} = \mathbf{L_{uu}} |
42 |
|
|
* = \mathbf{0}\f$. |
43 |
|
|
* |
44 |
|
|
* \sa `DifferentialActionModelAbstractTpl()`, `calcDiff()` |
45 |
|
|
*/ |
46 |
|
|
template <typename _Scalar> |
47 |
|
|
class DifferentialActionModelNumDiffTpl |
48 |
|
|
: public DifferentialActionModelAbstractTpl<_Scalar> { |
49 |
|
|
public: |
50 |
|
|
EIGEN_MAKE_ALIGNED_OPERATOR_NEW |
51 |
|
|
|
52 |
|
|
typedef _Scalar Scalar; |
53 |
|
|
typedef MathBaseTpl<Scalar> MathBase; |
54 |
|
|
typedef DifferentialActionModelAbstractTpl<Scalar> Base; |
55 |
|
|
typedef DifferentialActionDataNumDiffTpl<Scalar> Data; |
56 |
|
|
typedef DifferentialActionDataAbstractTpl<Scalar> |
57 |
|
|
DifferentialActionDataAbstract; |
58 |
|
|
typedef typename MathBase::VectorXs VectorXs; |
59 |
|
|
typedef typename MathBase::MatrixXs MatrixXs; |
60 |
|
|
|
61 |
|
|
/** |
62 |
|
|
* @brief Initialize the numdiff differential action model |
63 |
|
|
* |
64 |
|
|
* @param[in] model Differential action model that we want to |
65 |
|
|
* apply the numerical differentiation |
66 |
|
|
* @param[in] with_gauss_approx True if we want to use the Gauss |
67 |
|
|
* approximation for computing the Hessians |
68 |
|
|
*/ |
69 |
|
|
explicit DifferentialActionModelNumDiffTpl( |
70 |
|
|
boost::shared_ptr<Base> model, const bool with_gauss_approx = false); |
71 |
|
|
virtual ~DifferentialActionModelNumDiffTpl(); |
72 |
|
|
|
73 |
|
|
/** |
74 |
|
|
* @brief @copydoc Base::calc() |
75 |
|
|
*/ |
76 |
|
|
virtual void calc( |
77 |
|
|
const boost::shared_ptr<DifferentialActionDataAbstract>& data, |
78 |
|
|
const Eigen::Ref<const VectorXs>& x, const Eigen::Ref<const VectorXs>& u); |
79 |
|
|
|
80 |
|
|
/** |
81 |
|
|
* @brief @copydoc Base::calc(const |
82 |
|
|
* boost::shared_ptr<DifferentialActionDataAbstract>& data, const |
83 |
|
|
* Eigen::Ref<const VectorXs>& x) |
84 |
|
|
*/ |
85 |
|
|
virtual void calc( |
86 |
|
|
const boost::shared_ptr<DifferentialActionDataAbstract>& data, |
87 |
|
|
const Eigen::Ref<const VectorXs>& x); |
88 |
|
|
|
89 |
|
|
/** |
90 |
|
|
* @brief @copydoc Base::calcDiff() |
91 |
|
|
*/ |
92 |
|
|
virtual void calcDiff( |
93 |
|
|
const boost::shared_ptr<DifferentialActionDataAbstract>& data, |
94 |
|
|
const Eigen::Ref<const VectorXs>& x, const Eigen::Ref<const VectorXs>& u); |
95 |
|
|
|
96 |
|
|
/** |
97 |
|
|
* @brief @copydoc Base::calcDiff(const |
98 |
|
|
* boost::shared_ptr<DifferentialActionDataAbstract>& data, const |
99 |
|
|
* Eigen::Ref<const VectorXs>& x) |
100 |
|
|
*/ |
101 |
|
|
virtual void calcDiff( |
102 |
|
|
const boost::shared_ptr<DifferentialActionDataAbstract>& data, |
103 |
|
|
const Eigen::Ref<const VectorXs>& x); |
104 |
|
|
|
105 |
|
|
/** |
106 |
|
|
* @brief @copydoc Base::createData() |
107 |
|
|
*/ |
108 |
|
|
virtual boost::shared_ptr<DifferentialActionDataAbstract> createData(); |
109 |
|
|
|
110 |
|
|
/** |
111 |
|
|
* @brief @copydoc Base::quasiStatic() |
112 |
|
|
*/ |
113 |
|
|
virtual void quasiStatic( |
114 |
|
|
const boost::shared_ptr<DifferentialActionDataAbstract>& data, |
115 |
|
|
Eigen::Ref<VectorXs> u, const Eigen::Ref<const VectorXs>& x, |
116 |
|
|
const std::size_t maxiter = 100, const Scalar tol = Scalar(1e-9)); |
117 |
|
|
|
118 |
|
|
/** |
119 |
|
|
* @brief Return the differential acton model that we use to numerical |
120 |
|
|
* differentiate |
121 |
|
|
*/ |
122 |
|
|
const boost::shared_ptr<Base>& get_model() const; |
123 |
|
|
|
124 |
|
|
/** |
125 |
|
|
* @brief Return the disturbance constant used in the numerical |
126 |
|
|
* differentiation routine |
127 |
|
|
*/ |
128 |
|
|
const Scalar get_disturbance() const; |
129 |
|
|
|
130 |
|
|
/** |
131 |
|
|
* @brief Modify the disturbance constant used in the numerical |
132 |
|
|
* differentiation routine |
133 |
|
|
*/ |
134 |
|
|
void set_disturbance(const Scalar disturbance); |
135 |
|
|
|
136 |
|
|
/** |
137 |
|
|
* @brief Identify if the Gauss approximation is going to be used or not. |
138 |
|
|
*/ |
139 |
|
|
bool get_with_gauss_approx(); |
140 |
|
|
|
141 |
|
|
/** |
142 |
|
|
* @brief Print relevant information of the action numdiff model |
143 |
|
|
* |
144 |
|
|
* @param[out] os Output stream object |
145 |
|
|
*/ |
146 |
|
|
virtual void print(std::ostream& os) const; |
147 |
|
|
|
148 |
|
|
protected: |
149 |
|
|
using Base::has_control_limits_; //!< Indicates whether any of the control |
150 |
|
|
//!< limits |
151 |
|
|
using Base::nr_; //!< Dimension of the cost residual |
152 |
|
|
using Base::nu_; //!< Control dimension |
153 |
|
|
using Base::state_; //!< Model of the state |
154 |
|
|
using Base::u_lb_; //!< Lower control limits |
155 |
|
|
using Base::u_ub_; //!< Upper control limits |
156 |
|
|
|
157 |
|
|
private: |
158 |
|
|
void assertStableStateFD(const Eigen::Ref<const VectorXs>& x); |
159 |
|
|
boost::shared_ptr<Base> model_; |
160 |
|
|
bool with_gauss_approx_; |
161 |
|
|
Scalar e_jac_; //!< Constant used for computing disturbances in Jacobian |
162 |
|
|
//!< calculation |
163 |
|
|
Scalar e_hess_; //!< Constant used for computing disturbances in Hessian |
164 |
|
|
//!< calculation |
165 |
|
|
}; |
166 |
|
|
|
167 |
|
|
template <typename _Scalar> |
168 |
|
|
struct DifferentialActionDataNumDiffTpl |
169 |
|
|
: public DifferentialActionDataAbstractTpl<_Scalar> { |
170 |
|
|
EIGEN_MAKE_ALIGNED_OPERATOR_NEW |
171 |
|
|
|
172 |
|
|
typedef _Scalar Scalar; |
173 |
|
|
typedef MathBaseTpl<Scalar> MathBase; |
174 |
|
|
typedef DifferentialActionDataAbstractTpl<Scalar> Base; |
175 |
|
|
typedef typename MathBase::VectorXs VectorXs; |
176 |
|
|
typedef typename MathBase::MatrixXs MatrixXs; |
177 |
|
|
|
178 |
|
|
/** |
179 |
|
|
* @brief Construct a new ActionDataNumDiff object |
180 |
|
|
* |
181 |
|
|
* @tparam Model is the type of the ActionModel. |
182 |
|
|
* @param model is the object to compute the numerical differentiation from. |
183 |
|
|
*/ |
184 |
|
|
template <template <typename Scalar> class Model> |
185 |
|
85 |
explicit DifferentialActionDataNumDiffTpl(Model<Scalar>* const model) |
186 |
|
|
: Base(model), |
187 |
|
85 |
Rx(model->get_model()->get_nr(), |
188 |
|
85 |
model->get_model()->get_state()->get_ndx()), |
189 |
|
170 |
Ru(model->get_model()->get_nr(), model->get_model()->get_nu()), |
190 |
|
85 |
dx(model->get_model()->get_state()->get_ndx()), |
191 |
|
85 |
du(model->get_model()->get_nu()), |
192 |
✓✗✓✗ ✓✗✓✗ ✓✗ |
255 |
xp(model->get_model()->get_state()->get_nx()) { |
193 |
✓✗ |
85 |
Rx.setZero(); |
194 |
✓✗ |
85 |
Ru.setZero(); |
195 |
✓✗ |
85 |
dx.setZero(); |
196 |
✓✗ |
85 |
du.setZero(); |
197 |
✓✗ |
85 |
xp.setZero(); |
198 |
|
|
|
199 |
|
85 |
const std::size_t ndx = model->get_model()->get_state()->get_ndx(); |
200 |
|
85 |
const std::size_t nu = model->get_model()->get_nu(); |
201 |
✓✗ |
85 |
data_0 = model->get_model()->createData(); |
202 |
✓✓ |
4601 |
for (std::size_t i = 0; i < ndx; ++i) { |
203 |
✓✗✓✗
|
4516 |
data_x.push_back(model->get_model()->createData()); |
204 |
|
|
} |
205 |
✓✓ |
2005 |
for (std::size_t i = 0; i < nu; ++i) { |
206 |
✓✗✓✗
|
1920 |
data_u.push_back(model->get_model()->createData()); |
207 |
|
|
} |
208 |
|
85 |
} |
209 |
|
|
|
210 |
|
|
Scalar x_norm; //!< Norm of the state vector |
211 |
|
|
Scalar |
212 |
|
|
xh_jac; //!< Disturbance value used for computing \f$ \ell_\mathbf{x} \f$ |
213 |
|
|
Scalar |
214 |
|
|
uh_jac; //!< Disturbance value used for computing \f$ \ell_\mathbf{u} \f$ |
215 |
|
|
Scalar xh_hess; //!< Disturbance value used for computing \f$ |
216 |
|
|
//!< \ell_\mathbf{xx} \f$ |
217 |
|
|
Scalar uh_hess; //!< Disturbance value used for computing \f$ |
218 |
|
|
//!< \ell_\mathbf{uu} \f$ |
219 |
|
|
Scalar xh_hess_pow2; |
220 |
|
|
Scalar uh_hess_pow2; |
221 |
|
|
Scalar xuh_hess_pow2; |
222 |
|
|
MatrixXs Rx; |
223 |
|
|
MatrixXs Ru; |
224 |
|
|
VectorXs dx; |
225 |
|
|
VectorXs du; |
226 |
|
|
VectorXs xp; |
227 |
|
|
boost::shared_ptr<Base> data_0; |
228 |
|
|
std::vector<boost::shared_ptr<Base> > data_x; |
229 |
|
|
std::vector<boost::shared_ptr<Base> > data_u; |
230 |
|
|
|
231 |
|
|
using Base::cost; |
232 |
|
|
using Base::Fu; |
233 |
|
|
using Base::Fx; |
234 |
|
|
using Base::g; |
235 |
|
|
using Base::Gu; |
236 |
|
|
using Base::Gx; |
237 |
|
|
using Base::h; |
238 |
|
|
using Base::Hu; |
239 |
|
|
using Base::Hx; |
240 |
|
|
using Base::Lu; |
241 |
|
|
using Base::Luu; |
242 |
|
|
using Base::Lx; |
243 |
|
|
using Base::Lxu; |
244 |
|
|
using Base::Lxx; |
245 |
|
|
using Base::r; |
246 |
|
|
using Base::xout; |
247 |
|
|
}; |
248 |
|
|
|
249 |
|
|
} // namespace crocoddyl |
250 |
|
|
|
251 |
|
|
/* --- Details -------------------------------------------------------------- */ |
252 |
|
|
/* --- Details -------------------------------------------------------------- */ |
253 |
|
|
/* --- Details -------------------------------------------------------------- */ |
254 |
|
|
#include "crocoddyl/core/numdiff/diff-action.hxx" |
255 |
|
|
|
256 |
|
|
#endif // CROCODDYL_CORE_NUMDIFF_DIFF_ACTION_HPP_ |