Lightweight Levenberg Marquardt

This is a malloc-free Levenberg-Marquardt optimizer for nonlinear least squares regression. This means that heap objects are never allocated during the optimization phase. It comes with a demonstration fitting an arbitrary nonlinear function. This is a subproject for GTSAM, a smoothing and mapping library as a part of my undergraduate research.

How does it work?

As a consumer of the optimizer engine, you simply need to implement the DataManipulator class. The class is used to fill out a jacobian and a residual matrix belonging to the optimizer. The optimizer does not care what data it's fitting, just that the manipulator fills the aforementioned matrices.

Dependencies

• Eigen 3.3

Running the demo

From the root of the project:

1. mkdir build/
2. cd build
3. cmake ..
4. make && ./LMSolver