Template Class CNetworkOfPoses

Nested Relationships

Nested Types

Inheritance Relationships

Base Type

Class Documentation

template<class CPOSE, class MAPS_IMPLEMENTATION = mrpt::containers::map_traits_stdmap, class NODE_ANNOTATIONS = mrpt::graphs::detail::TNodeAnnotationsEmpty, class EDGE_ANNOTATIONS = mrpt::graphs::detail::edge_annotations_empty>
class CNetworkOfPoses : public mrpt::graphs::CDirectedGraph<CPOSE, mrpt::graphs::detail::edge_annotations_empty>

A directed graph of pose constraints, with edges being the relative poses between pairs of nodes identified by their numeric IDs (of type mrpt::graphs::TNodeID). A link or edge between two nodes “i” and “j”, that is, the pose \( p_{ij} \), holds the relative position of “j” with respect to “i”. These poses are stored in the edges in the format specified by the template argument CPOSE. Users should employ the following derived classes depending on the desired representation of edges:

Two main members store all the information in this class:

  • edges (in the base class mrpt::graphs::CDirectedGraph::edges): A map from pairs of node ID -> pose constraints.

  • nodes : A map from node ID -> estimated pose of that node (actually, read below on the template argument MAPS_IMPLEMENTATION).

Graphs can be loaded and saved to text file in the format used by TORO & HoG-man (more on the format here), using loadFromTextFile and saveToTextFile.

This class is the base for representing networks of poses, which are the main data type of a series of SLAM algorithms implemented in the library mrpt-slam, in the namespace mrpt::graphslam.

The template arguments are:

See also

mrpt::graphslam

Typedef’s

using BASE = mrpt::graphs::CDirectedGraph<CPOSE, EDGE_ANNOTATIONS>

The base class “CDirectedGraph<CPOSE,EDGE_ANNOTATIONS>”

using self_t = CNetworkOfPoses<CPOSE, MAPS_IMPLEMENTATION, NODE_ANNOTATIONS, EDGE_ANNOTATIONS>

My own type

using constraint_t = CPOSE

The type of PDF poses in the contraints (edges) (=CPOSE template argument)

using node_annotations_t = NODE_ANNOTATIONS

The extra annotations in nodes, apart from a constraint_no_pdf_t

using edge_annotations_t = EDGE_ANNOTATIONS

The extra annotations in edges, apart from a constraint_t

using maps_implementation_t = MAPS_IMPLEMENTATION

The type of map’s implementation (=MAPS_IMPLEMENTATION template argument)

using constraint_no_pdf_t = typename CPOSE::type_value

The type of edges or their means if they are PDFs (that is, a simple “edge” value)

using global_poses_pdf_t = typename MAPS_IMPLEMENTATION::template map<mrpt::graphs::TNodeID, CPOSE>

A map from pose IDs to their global coordinate estimates, with uncertainty

using global_poses_t = typename MAPS_IMPLEMENTATION::template map<mrpt::graphs::TNodeID, global_pose_t>

A map from pose IDs to their global coordinate estimates, without uncertainty (the “most-likely value”)

static inline constexpr auto getClassName()

Data members

global_poses_t nodes

The nodes (vertices) of the graph, with their estimated “global” (with respect to root) position, without an associated covariance.

See also

dijkstra_nodes_estimate

mrpt::graphs::TNodeID root = {0}

The ID of the node that is the origin of coordinates, used as reference by all coordinates in nodes. By default, root is the ID “0”.

bool edges_store_inverse_poses = {false}

False (default) if an edge i->j stores the normal relative pose of j as seen from i: \( \Delta_i^j = j \ominus i \) True if an edge i->j stores the inverse relateive pose, that is, i as seen from j: \( \Delta_i^j = i \ominus j \)

Utility methods

static inline void connectGraphPartitions(self_t *sub_graph, const std::set<TNodeID> &groupA, const std::set<TNodeID> &groupB)

Add an edge between the last node of the group with the lower nodeIDs and the first node of the higher nodeIDs.

Given groups of nodes should only contain consecutive nodeIDs and there should be no overlapping between them

Note

It is assumed that the sets of nodes are already in ascending order (default std::set behavior.

inline void getAs3DObject(mrpt::viz::CSetOfObjects::Ptr object, const mrpt::containers::yaml &viz_params) const

Return 3D Visual Representation of the edges and nodes in the network of poses.

Method makes the call to the corresponding method of the CVisualizer class instance.

inline void dijkstra_nodes_estimate(std::optional<std::reference_wrapper<std::map<TNodeID, size_t>>> topological_distances = std::nullopt)

Spanning tree computation of a simple estimation of the global coordinates of each node just from the information in all edges, sorted in a Dijkstra tree based on the current “root” node.

See also

node, root

Note

The “global” coordinates are with respect to the node with the ID specified in root.

Note

This method takes into account the value of edges_store_inverse_poses

inline size_t collapseDuplicatedEdges()

Look for duplicated edges (even in opposite directions) between all pairs of nodes and fuse them. Upon return, only one edge remains between each pair of nodes with the mean & covariance (or information matrix) corresponding to the Bayesian fusion of all the Gaussians.

Returns:

Overall number of removed edges.

inline double chi2() const

Returns the total chi-squared error of the graph. Shortcut for getGlobalSquareError(false).

inline double getGlobalSquareError(bool ignoreCovariances = true) const

Evaluates the graph total square error (ignoreCovariances=true) or chi2 (ignoreCovariances=false) from all the pose constraints (edges) with respect to the global poses in nodes.

See also

getEdgeSquareError

Throws:

std::exception – On global poses not in nodes

inline void extractSubGraph(const std::set<TNodeID> &node_IDs, self_t *sub_graph, const TNodeID root_node_in = INVALID_NODEID, bool auto_expand_set = true) const

Find the edges between the nodes in the node_IDs set and fill given graph pointer accordingly.

Parameters:

  • node_IDs – [in] Set of nodes, between which, edges should be found and inserted in the given sub_graph pointer

  • root_node_in – [in] Node ID to be used as the root node of sub_graph. If this is not given, the lowest nodeID is to be used.

  • CNetworkOfPoses – pointer that is to be filled.

  • auto_expand_set – [in] If true and in case the node_IDs set contains non-consecutive nodes the returned set is expanded with the in-between nodes. This makes sure that the final graph is always connected. If auto_expand_set is false but there exist non-consecutive nodes, virtual edges are inserted in the parts that the graph is not connected

inline double getEdgeSquareError(const typename BASE::edges_map_t::const_iterator &itEdge, bool ignoreCovariances = true) const

Computes the square error of one pose constraints (edge) with respect to the global poses in nodes If ignoreCovariances is false, the squared Mahalanobis distance will be computed instead of the straight square error.

Throws:

std::exception – On global poses not in nodes

inline double getEdgeSquareError(const mrpt::graphs::TNodeID from_id, const mrpt::graphs::TNodeID to_id, bool ignoreCovariances = true) const

Computes the square error of one pose constraints (edge) with respect to the global poses in nodes If ignoreCovariances is false, the squared Mahalanobis distance will be computed instead of the straight square error.

Throws:

std::exception – On edge not existing or global poses not in nodes

inline void clear()

Empty all edges, nodes and set root to ID 0.

inline size_t nodeCount() const

Return number of nodes in the list nodes of global coordinates (may be different that all nodes appearing in edges)

See also

mrpt::graphs::CDirectedGraph::countDifferentNodesInEdges

I/O methods

inline void saveToTextFile(const std::string &fileName) const

Saves to a text file in the format used by TORO, HoG-man, G2O. See: https://www.mrpt.org/Graph-SLAM_maps

See also

saveToBinaryFile, loadFromTextFile, writeAsText

Throws:

On – any error

inline void writeAsText(std::ostream &o) const

Writes as text in the format used by TORO, HoG-man, G2O. See: https://www.mrpt.org/Graph-SLAM_maps

See also

saveToBinaryFile, loadFromTextFile, saveToTextFile, readAsText

Throws:

On – any error

inline void loadFromTextFile(const std::string &fileName, bool collapse_dup_edges = true)

Loads from a text file in the format used by TORO & HoG-man (more on the format here) Recognized line entries are: VERTEX2, VERTEX3, EDGE2, EDGE3, EQUIV. If an unknown entry is found, a warning is dumped to std::cerr (only once for each unknown keyword). An exception will be raised if trying to load a 3D graph into a 2D class (in the opposite case, missing 3D data will default to zero).

See also

loadFromBinaryFile, saveToTextFile

Parameters:

  • fileName – [in] The file to load.

  • collapse_dup_edges – [in] If true, collapseDuplicatedEdges will be called automatically after loading (note that this operation may take significant time for very large graphs).

Throws:

On – any error, as a malformed line or loading a 3D graph in a 2D graph.

inline void readAsText(std::istream &i)

Reads as text in the format used by TORO, HoG-man, G2O. See: https://www.mrpt.org/Graph-SLAM_maps

See also

saveToBinaryFile, loadFromTextFile, saveToTextFile

Throws:

On – any error

struct global_pose_t : public constraint_no_pdf_t, public NODE_ANNOTATIONS

The type of each global pose in nodes: an extension of the constraint_no_pdf_t pose with any optional user-defined data

Unnamed Group

global_pose_t() = default

Potential class constructors.

template<typename ARG1>
inline global_pose_t(const ARG1 &a1)
template<typename ARG1, typename ARG2>
inline global_pose_t(const ARG1 &a1, const ARG2 &a2)

Public Types

using self_t = typename CNetworkOfPoses<CPOSE, MAPS_IMPLEMENTATION, NODE_ANNOTATIONS, EDGE_ANNOTATIONS>::global_pose_t

Public Functions

inline bool operator==(const global_pose_t &other) const
inline bool operator!=(const global_pose_t &other) const

Public Static Functions

static inline constexpr auto getClassName()

Friends

inline friend std::ostream &operator<<(std::ostream &o, const self_t &global_pose)