ISBN 9783843950466

978-3-8439-5046-6, Reihe Informatik

Cornelia Schulz
Probabilistic Multi-Distribution Mapping for Mobile Robots

200 Seiten, Dissertation Eberhard-Karls-Universität Tübingen (2021), Softcover, A5

Zusammenfassung / Abstract

Nowadays, autonomous mobile robots are increasingly finding their way into our everyday lives. Especially for applications like lawn-mowing or vacuum-cleaning, robotic solutions have become very popular. For such autonomous applications, the deployed robot needs to be able to model its environment precisely, as well as recognize previously visited places. In this work, we focus on a data representation called Normal Distributions Transform (NDT). Based on a regular grid, this model fuses environmental data to one normal distribution per grid cell in order to track the exact location of an obstacle in sub-pixel precision.

To introduce our mapping approach, we first of all present the modular architecture of our NDT-based map representations, as well as the design concepts we chose to efficiently incorporate additional information such as occupancy probabilities. Besides that, we also investigate various techniques in order to find a suitable tradeoff between the runtime of our mapping algorithms and the accuracy of the resulting maps.

Then, we present the advances we made with these map types in basic robotic algorithms. These include Monte Carlo Localization (MCL), a method for localizing within a given map, and Simultaneous Localization and Mapping (SLAM), where the map is instead constructed during the localization process from integrated sensor measurements that are matched to previous observations.

Finally, we present a simple distributed mapping approach, with which we investigate the applicability of these NDT-based maps to multi-robot scenarios. In our experiments, we deployed multiple robots that autonomously moved through a known indoor environment, and at the same time transmitted incremental map updates to the other robots via simple radio frequency (RF) modules, so that each robot was able to construct a fused map of the complete environment.