Add relative pose to pose graph
creates a node based on the input
measurement that connects to
the last pose node in the pose graph. To add landmark nodes, see the
also specifies the information matrix as part of the edge constraint, which
represents the uncertainty of the pose measurement.
creates a new pose node and connects it to the specific node specified by
creates an edge by specifying a relative pose measurement between existing nodes
edge is called a loop closure. If a loop closure already
exists, the function appends the new measurement. Calling the
optimizePoseGraph function combines multiple appended measurements
into a single edge. This syntax does not support adding edges to a landmark
Identify and Remove Spurious Loop Closures from Pose Graph
This example shows how to identify and remove spurious loop closures from pose graph. To do this, you can modify the relative pose of a loop closure edge and try optimizing the pose graph with and without removing the auto spurious loop closure and compare the results.
Load the Intel Research Lab Dataset that contains a 2-D pose graph. Optimize the pose graph. Plot the pose graph with IDs off. Red lines indicate loop closures identified in the dataset.
load intel-2d-posegraph.mat pg optimizedPG = optimizePoseGraph(pg); show(optimizedPG,IDs="off"); title("Optimized Pose Graph")
Modify the relative pose of the loop closure edge
1386 to some random values.
loopclosureId = 1386; nodePair = edgeNodePairs(optimizedPG,loopclosureId); [relPose,infoMat] = edgeConstraints(optimizedPG,loopclosureId); relPose(2) = -5; relPose(3) = 1.5; addRelativePose(optimizedPG,relPose,infoMat,nodePair(1),nodePair(2));
Optimize the pose graph without auto loop closure trimming. Plot the optimized pose graph to see the poor adjustment of the nodes with loop closures.
[updatedPG,solutionInfo] = optimizePoseGraph(optimizedPG); show(updatedPG,IDs="off"); title("Updated Pose Graph")
Certain loop closures should be trimmed from the pose graph. Use the
trimLoopClosures function to trim these bad loop closures. Set the truncation threshold and maximum iterations for the trimmer parameters.
trimParams = struct("TruncationThreshold",0.5,"MaxIterations",100);
Generate solver options.
solverOptions = poseGraphSolverOptions("g2o-levenberg-marquardt");
trimLoopClosures function with the trimmer parameters and solver options. Plot the new pose graph to see the bad loop closures were removed.
[newPG,trimInfo] = trimLoopClosures(updatedPG,trimParams,solverOptions); show(newPG,IDs="off"); title("New Pose Graph")
poseGraph — Pose graph
poseGraph object |
Pose graph, specified as a
measurement — Relative pose between nodes
[x y theta] vector |
[x y z qw qx qy qz] vector
Relative pose between nodes, specified as one of the following:
poseGraph (2-D), the pose is a
[x y theta]
vector, which defines a xy-position and orientation
poseGraph3D, the pose is a
[x y z qw qx qy
qz] vector, which defines by an
xyz-position and quaternion orientation,
Many other sources for 3-D pose graphs, including
.g2o formats, specify the quaternion orientation
in a different order, for example,
[qx qy qz qw].
Check the source of your pose graph data before adding nodes to your
infoMat — Information matrix
six-element vector | 21-element vector
Information matrices, specified in compact form as a six-element vector or 21-element vector.
Each vector is the compact form of the upper triangle of the square
information matrix. An information matrix represents the uncertainty of the
measurement. The matrix is calculated as the inverse of the covariance. If
the measurement is an
[x y theta] vector, the covariance
matrix is a 3-by-3 of pairwise covariance calculations. Typically, the
uncertainty is determined by the sensor model.
poseGraph (2-D), each information matrix is a six-element
vector. The default is
[1 0 0 1 0 1]. For landmark nodes,
the last three elements are returned as
poseGraph3D, each information matrix is a 21-element vector.
The default is
[1 0 0 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 1 0
toNodeID — Node to attach to
Node to attach to, specified as a positive integer. This integer
corresponds to the node ID of a node in
addRelativePose adds an edge between this node and
nodePair — Edge node pair in pose graph
Edge node pairs in pose graph, returned as two-element vector that lists the IDs of the two nodes that each edge connects. Multiple edges may exist between the same pair of nodes.
edgeID — ID of added edge
ID of added edge, returned as a positive integer.
C/C++ Code Generation
Generate C and C++ code using MATLAB® Coder™.
Use this syntax when constructing
poseGraph3D objects for code generation:
poseGraph('MaxNumEdges',maxEdges,'MaxNumNodes',maxNodes) specifies an
upper bound on the number of edges and nodes allowed in the pose graph when
generating code. This limit is only required when generating code.
Introduced in R2019b