Matlab

We provide the equivalent Matlab code for designing UKF on (parallelizable) manifolds, which is wholly independent from the python code.

Installation

The Matlab code has been tested with version R2019a without requiring any particular toolbox. To install:

  1. Download the repo:

    git clone https://github.com/CAOR-MINES-ParisTech/ukfm.git

2. Go to /my/directory/ukfm/matlab at the Matlab prompt and execute importukfm.

  1. You may save this path for your next Matlab sessions (via savepath).

Get Started

Follow the 2D robot localization example (tutorial): in the Matlab prompt execute:

main_localization

Usage

In contrast to Python, the code is implemented without class and has only functions.

The file for designing an UKF are given in the ukfm folder and useful geometry (Lie groups) functions are provided in the geometry folder.

We provide scripts for reproducing the examples and benchmarks respectively in the examples and benchmarks folders. Models functions are organized in suborder of the example folder: for e.g. the 2D robot localization model, see in examples/localization. You can use the Matlab publish tool for better rendering.

Function naming mimics the dot operator of class. To get the exponential of \(SE(3)\) or the propagation function of the localization example, call respectively se3_exp and localization_f.

Examples

The unscented Kalman filter on parallelizable manifolds has been implemented on the following examples, from which you can obtain published scripts:

Benchmarks

We simulate Monte-Carlo data and benchmark UKFs and EKFs based on different choices of uncertainty representation through accuracy and consistency metrics on the following benchmark: