Probability Density Functions (PDFs) over spatial transformations

These distributions can be used to represent the robot positioning belief and the map uncertainty in many localization and SLAM algorithms.

They include unimodal Gaussians (in covariance or information form), sums of Gaussians, sets of particles, and grid representations, together with methods to convert between them and to draw an arbitrary number of samples from any kind of distribution.

The abstract base classes, and their implementations, are:

• 2D points:

  • mrpt::poses::CPoint2DPDF: CPoint2DPDFGaussian.

• 3D points:

  • mrpt::poses::CPointPDF: CPointPDFGaussian, CPointPDFSOG, CPointPDFParticles.

• SE(2) poses:

  • mrpt::poses::CPosePDF: CPosePDFGaussian, CPosePDFGaussianInf, CPosePDFSOG, CPosePDFParticles, CPosePDFGrid.

• SE(3) poses, with the rotation as yaw/pitch/roll angles:

  • mrpt::poses::CPose3DPDF: CPose3DPDFGaussian, CPose3DPDFGaussianInf, CPose3DPDFSOG, CPose3DPDFParticles, CPose3DPDFGrid.

• SE(3) poses, with the rotation as a quaternion:

  • mrpt::poses::CPose3DQuatPDF: CPose3DQuatPDFGaussian, CPose3DQuatPDFGaussianInf.

The technical report [BC21] contains most of the derivations of the implemented Jacobians.