Skip to content

core: Core Gaussian Process Models and Transformations

This module contains the fundamental building blocks for modeling and transforming sensor data.

  • AugmentedSGPR and AugmentedGPR: These are extensions of GPflow's SGPR and GPR models. They are "augmented" to incorporate custom Transformations on the inducing points, which is a key feature of this library for modeling complex sensor setups.

  • Transformations: This is a crucial part of the library, defining how inducing points in the SGP are manipulated to represent different physical sensing scenarios.

    • Transform: The base class for all transformations.

    • IPPTransform: A versatile transform for Informative Path Planning (IPP). It can model continuous sensing paths (by interpolating points between waypoints), handle multi-robot scenarios, and enforce distance constraints on the paths. It also supports online IPP where some waypoints are fixed.

    • SquareTransform and SquareHeightTransform: These transforms model non-point, 2D fields of view (FoV). SquareTransform creates a square FoV with a fixed size and optimizable orientation, while SquareHeightTransform models a FoV whose size depends on the sensor's height from the ground (the z-dimension).

  • osgpr: This module provides an implementation of an Online Sparse Variational GP regression model (OSGPR_VFE), which is designed for streaming data scenarios where the model is updated sequentially with new data batches; used for adaptive IPP. The init_osgpr function helps in setting up this model.