levmarq_impl.h

Go to the documentation of this file.

/* +------------------------------------------------------------------------+
   |                     Mobile Robot Programming Toolkit (MRPT)            |
   |                          https://www.mrpt.org/                         |
   |                                                                        |
   | Copyright (c) 2005-2020, Individual contributors, see AUTHORS file     |
   | See: https://www.mrpt.org/Authors - All rights reserved.               |
   | Released under BSD License. See: https://www.mrpt.org/License          |
   +------------------------------------------------------------------------+ */
#pragma once

#include <Eigen/Dense>
#include <vector>

namespace mrpt
{
namespace graphslam
{
/// Internal auxiliary classes
namespace detail
{
using namespace mrpt;
using namespace mrpt::poses;
using namespace mrpt::graphslam;
using namespace mrpt::math;
using namespace std;

// An auxiliary struct to compute the pseudo-ln of a pose error, possibly
// modified with an information matrix.
//  Specializations are below.
template <class EDGE, class gst>
struct AuxErrorEval;

// For graphs of 2D constraints (no information matrix)
template <class gst>
struct AuxErrorEval<CPose2D, gst>
{
    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJtLambdaJ(
        const MAT& J1, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        MRPT_UNUSED_PARAM(edge);
        JtJ.matProductOf_AtA(J1);
    }

    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJ1tLambdaJ2(
        const MAT& J1, const MAT& J2, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        MRPT_UNUSED_PARAM(edge);
        JtJ.asEigen() = J1.transpose() * J2.asEigen();
    }

    template <class JAC, class EDGE_ITERATOR, class VEC1, class VEC2>
    static inline void multiply_Jt_W_err(
        const JAC& J, const EDGE_ITERATOR& edge, const VEC1& ERR, VEC2& OUT)
    {
        MRPT_UNUSED_PARAM(edge);
        const auto grad_incr = (J.transpose() * ERR.asEigen()).eval();
        OUT.asEigen() += grad_incr;
    }
};

// For graphs of 3D constraints (no information matrix)
template <class gst>
struct AuxErrorEval<CPose3D, gst>
{
    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJtLambdaJ(
        const MAT& J1, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        MRPT_UNUSED_PARAM(edge);
        JtJ.matProductOf_AtA(J1);
    }

    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJ1tLambdaJ2(
        const MAT& J1, const MAT& J2, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        MRPT_UNUSED_PARAM(edge);
        JtJ.asEigen() = J1.transpose() * J2.asEigen();
    }

    template <class JAC, class EDGE_ITERATOR, class VEC1, class VEC2>
    static inline void multiply_Jt_W_err(
        const JAC& J, const EDGE_ITERATOR& edge, const VEC1& ERR, VEC2& OUT)
    {
        MRPT_UNUSED_PARAM(edge);
        OUT.asEigen() += J.transpose() * ERR.asEigen();
    }
};

// For graphs of 2D constraints (with information matrix)
template <class gst>
struct AuxErrorEval<CPosePDFGaussianInf, gst>
{
    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJtLambdaJ(
        const MAT& J1, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        JtJ.asEigen() =
            (J1.transpose() * edge->second.cov_inv.asEigen()) * J1.asEigen();
    }
    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJ1tLambdaJ2(
        const MAT& J1, const MAT& J2, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        JtJ.asEigen() =
            (J1.transpose() * edge->second.cov_inv.asEigen()) * J2.asEigen();
    }

    template <class JAC, class EDGE_ITERATOR, class VEC1, class VEC2>
    static inline void multiply_Jt_W_err(
        const JAC& J, const EDGE_ITERATOR& edge, const VEC1& ERR, VEC2& OUT)
    {
        OUT.asEigen() +=
            (J.transpose() * edge->second.cov_inv.asEigen()) * ERR.asEigen();
    }
};

// For graphs of 3D constraints (with information matrix)
template <class gst>
struct AuxErrorEval<CPose3DPDFGaussianInf, gst>
{
    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJtLambdaJ(
        const MAT& J1, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        JtJ.asEigen() =
            (J1.transpose() * edge->second.cov_inv.asEigen()) * J1.asEigen();
    }

    template <class MAT, class EDGE_ITERATOR>
    static inline void multiplyJ1tLambdaJ2(
        const MAT& J1, const MAT& J2, MAT& JtJ, const EDGE_ITERATOR& edge)
    {
        JtJ.asEigen() =
            (J1.transpose() * edge->second.cov_inv.asEigen()) * J2.asEigen();
    }

    template <class JAC, class EDGE_ITERATOR, class VEC1, class VEC2>
    static inline void multiply_Jt_W_err(
        const JAC& J, const EDGE_ITERATOR& edge, const VEC1& ERR, VEC2& OUT)
    {
        OUT.asEigen() +=
            (J.transpose() * edge->second.cov_inv.asEigen()) * ERR.asEigen();
    }
};

}  // namespace detail

// Compute, at once, jacobians and the error vectors for each constraint in
// "lstObservationData", returns the overall squared error.
template <class GRAPH_T>
double computeJacobiansAndErrors(
    const GRAPH_T& graph,
    const std::vector<typename graphslam_traits<GRAPH_T>::observation_info_t>&
        lstObservationData,
    typename graphslam_traits<GRAPH_T>::map_pairIDs_pairJacobs_t& lstJacobians,
    std::vector<typename graphslam_traits<GRAPH_T>::Array_O>& errs)
{
    MRPT_UNUSED_PARAM(graph);
    using gst = graphslam_traits<GRAPH_T>;

    lstJacobians.clear();
    errs.clear();

    const size_t nObservations = lstObservationData.size();

    for (size_t i = 0; i < nObservations; i++)
    {
        const typename gst::observation_info_t& obs = lstObservationData[i];
        auto e = obs.edge;
        const typename gst::graph_t::constraint_t::type_value* EDGE_POSE =
            obs.edge_mean;
        typename gst::graph_t::constraint_t::type_value* P1 = obs.P1;
        typename gst::graph_t::constraint_t::type_value* P2 = obs.P2;

        const auto& ids = e->first;
        // const auto& edge = e->second;

        // Compute the residual pose error of these pair of nodes + its
        // constraint:
        // DinvP1invP2 = inv(EDGE) * inv(P1) * P2 = (P2 \ominus P1) \ominus EDGE
        typename gst::graph_t::constraint_t::type_value DinvP1invP2 =
            ((*P2) - (*P1)) - *EDGE_POSE;

        // Add to vector of errors:
        errs.resize(errs.size() + 1);
        errs.back() = gst::SE_TYPE::log(DinvP1invP2);

        // Compute the jacobians:
        alignas(MRPT_MAX_STATIC_ALIGN_BYTES)
            std::pair<mrpt::graphs::TPairNodeIDs, typename gst::TPairJacobs>
                newMapEntry;
        newMapEntry.first = ids;
        gst::SE_TYPE::jacob_dDinvP1invP2_de1e2(
            -(*EDGE_POSE), *P1, *P2, newMapEntry.second.first,
            newMapEntry.second.second);

        // And insert into map of jacobians:
        lstJacobians.insert(lstJacobians.end(), newMapEntry);
    }

    // return overall square error:  (Was:
    // std::accumulate(...,mrpt::squareNorm_accum<>), but led to GCC
    // errors when enabling parallelization)
    double ret_err = 0.0;
    for (size_t i = 0; i < errs.size(); i++)
        ret_err += mrpt::square(errs[i].norm());
    return ret_err;
}

}  // namespace graphslam
}  // namespace mrpt