Template Function mrpt::graphslam::optimize_graph_spa_levmarq

Function Documentation

template<class GRAPH_T, class FEEDBACK_CALLABLE = typename graphslam_traits<GRAPH_T>::TFunctorFeedback>
void mrpt::graphslam::optimize_graph_spa_levmarq(GRAPH_T &graph, TResultInfoSpaLevMarq &out_info, const std::set<mrpt::graphs::TNodeID> *in_nodes_to_optimize = nullptr, const mrpt::containers::yaml &extra_params = {}, FEEDBACK_CALLABLE functor_feedback = FEEDBACK_CALLABLE())

Optimize a graph of pose constraints using the Sparse Pose Adjustment (SPA) sparse representation and a Levenberg-Marquardt optimizer. This method works for all types of graphs derived from CNetworkOfPoses (see its reference mrpt::graphs::CNetworkOfPoses for the list). The input data are all the pose constraints in graph (graph.edges), and the gross first estimations of the “global” pose coordinates (in graph.nodes).

Note that these first coordinates can be obtained with mrpt::graphs::CNetworkOfPoses::dijkstra_nodes_estimate().

The method implemented in this file is based on this work:

  • ”Efficient Sparse Pose Adjustment for 2D Mapping”, Kurt Konolige et al., 2010. , but generalized for not only 2D but 2D and 3D poses, and using on-manifold optimization.

List of optional parameters by name in “extra_params”:

  • ”verbose”: (default=0) If !=0, produce verbose ouput.

  • ”max_iterations”: (default=100) Maximum number of Lev-Marq. iterations.

  • ”initial_lambda”: (default=0) <=0 means auto guess, otherwise, initial lambda value for the lev-marq algorithm.

  • ”tau”: (default=1e-6) Initial tau value for the lev-marq algorithm.

  • ”e1”: (default=1e-6) Lev-marq algorithm iteration stopping criterion #1: |gradient| < e1

  • ”e2”: (default=1e-6) Lev-marq algorithm iteration stopping criterion #2: |delta_incr| < e2*(x_norm+e2)

See also

The example “graph_slam_demo”

Note

The following graph types are supported: mrpt::graphs::CNetworkOfPoses2D, mrpt::graphs::CNetworkOfPoses3D, mrpt::graphs::CNetworkOfPoses2DInf, mrpt::graphs::CNetworkOfPoses3DInf

Note

Implementation can be found in file levmarq_impl.h

Parameters:

  • graph – [inout] The input edges and output poses.

  • out_info – [out] Some basic output information on the process.

  • nodes_to_optimize – [in] The list of nodes whose global poses are to be optimized. If nullptr (default), all the node IDs are optimized (but that marked as root in the graph).

  • extra_params – [in] Optional parameters, see below.

  • functor_feedback – [in] Optional: a pointer to a user function can be set here to be called on each LM loop iteration (eg to refresh the current state and error, refresh a GUI, etc.)

Template Parameters:

GRAPH_T – Normally a mrpt::graphs::CNetworkOfPoses<EDGE_TYPE,MAPS_IMPLEMENTATION>. Users won’t have to write this template argument by hand, since the compiler will auto-fit it depending on the type of the graph object.