gtsam  3.2.0
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages

Creating new factor and variable types

GTSAM comes with a set of variable and factor types typically used in SFM and SLAM. Geometry variables such as points and poses are in the geometry subdirectory and module. Factors such as BetweenFactor and BearingFactor are in the gtsam/slam directory.

To use GTSAM to solve your own problems, you will often have to create new factor types, which derive either from NonlinearFactor or NoiseModelFactor, or one of their derived types. Here is an outline of the options:

  1. The number of variables your factor involves is unknown at compile time - derive from NoiseModelFactor and implement NoiseModelFactor::unwhitenedError()
    • This is a factor expressing the sum-of-squares error between a measurement \( z \) and a measurement prediction function \( h(x) \), on which the errors are expected to follow some distribution specified by a noise model (see noiseModel).
  2. The number of variables your factor involves is known at compile time and is between 1 and 6 - derive from NonlinearFactor1, NonlinearFactor2, NonlinearFactor3, NonlinearFactor4, NonlinearFactor5, or NonlinearFactor6, and implement evaluateError()
    • This factor expresses the same sum-of-squares error with a noise model, but makes the implementation task slightly easier than with NoiseModelFactor.
  3. Derive from NonlinearFactor
    • This is more advanced and allows creating factors without an explicit noise model, or that linearize to HessianFactor instead of JacobianFactor.