ForschungBig Data und Machine Learning
Localization and mapping using maximum consensus

Localization and mapping using maximum consensus

Team:  Axmann, Brenner
Jahr:  2020

The long-term goal of this research topic is the creation of a localization and mapping algorithm, which is robust against outliers and disturbances. The research project is embedded in the Research Training Group “Integrity and Collaboration in Dynamic Sensor Networks (i.c.sens)” and primarily aims at improving integrity measures. The research is devided into two steps. In the first step, the localization considering the map as known is examined. In the second step, the problem will be extended treating the map as unknown as well.

Often, both for the pose of the autonomous mobile system and for the map, uncertainties were modelled using parametric distributions. In localization and mapping applications, wrong associations induced by erroneous measurements, false assumptions about the ego position, or altering environments cause deviations from the distribution function. Those deviations can’t be modelled and result in a wrong estimate of the system state, if not detected as outliers before. Such erroneous estimates are hard to ascertain leading to integrity concerns.

To approach these drawbacks, the newly established localization and mapping algorithm is based on maximum consensus techniques. The single state estimation as well as the applicability of these techniques for problems of filtering and smoothing are investigated. Furthermore, quality measures for the evaluation of the estimated system state using conventional or deep learning methods are developed. The high computational effort, which comes along with maximum consensus techniques, is handled by means of parallel computation using parallel clusters or GPUs.

If you are interested in this research project (e.g. for a Master's Thesis), please contact Jeldrik Axmann (jeldrik.axmann@ikg.uni-hannover.de).