KISS - Keep It Static SLAMMOT

This is the repository to accompany code for the paper titled KISS - Keep-It-Static-SLAMMOT, which integrates moving landmarks into an EKF-SLAM algorithm. Think of this analogy: you are on a ship and want to navigate at night. There is a cliff in front of you, with one lighthouse and a single car parked on the cliff. In traditional SLAM algorithms, you would discard the car as a moving object. However, your information would be incomplete and you would be unable to uniquely determine your position. Even though the car moves, at that instant it yields information which may contribute to the localisation of your ship. This is what this repository does. A cliff at night with a lighthouse and a car parked on the cliff. A ship in front of it.

Installation

This package relies on the robotics-toolbox as a basis. Please install the conda environment from conda env create -f environment.yml, activate the environment with conda activate mrekf and then pip-install the files locally editable through pip install -e . For further reference, see the Good Research Code Handbook

Usage

All executable scripts are in the scripts subfolder and make use of source files in src. To run a script, please activate the conda environment first with conda activate mrekf. arguments.py is the main script, which will execute using hydra. The config files for hydra can be found in the config folder

Re-running Filters

Filters can be re-run with the rerun-script. This loads a ground-truth history, with odometry and observations, which are the input to the filter. Also, ground truth values for all robot tracks are available. If desired, stored filter values can be loaded as well. The map can be loaded from the simulation dictionary, which allows re-calculation of all transforms and map values.

Data

Data files are available on the repository of the University of Lubeck

submisson_results.xlsx is the spreadsheet for analysis with eval_csv.py
histories.zip includes the Groundtruth histories as .pkl files for 1996 cases included in the evaluation and can be used to re-run all filters deterministically, see rerun-script. plot_case.py can be used to plot single case results.
files prefixed with r5 and r10 are for sensor settings, which use reduced sensor ranges.
files prefixed with dynamic use a dynamic driver, which changes velocities depending on the heading angle.

Feedback

please use the issues of this repository. or provide direct feedback to nicolas.mandel@uni-luebeck.de