In the past decades, fleets of mobile robots have increasingly entered the sector of industrial production and warehousing, replacing conventional logistic systems like conveyor belts and human-controlled vehicles. To achieve the reliability needed for these industrial applications, current navigation solutions commonly rely on additional infrastructure (like magnetic wires or retro-reflective markers) or are limited to highly structured, non-changing environments. This circumstance, however, limits their flexibility with respect to modifications of the environment or altered transportation tasks as well as their efficiency when operating in shared workspaces with humans or other dynamic objects. Moreover, this forbids a further exploration into new applications of highly dynamic and changing environments. In order to overcome these limitations, navigation solutions with an increased level of autonomy without decreasing reliability or precision of current solutions are needed. This thesis tackles this issue by leveraging current advances in the fields of cloud and networked robotics for the particular application of mobile robot navigation. We propose a cloud-based cooperative navigation architecture which enables knowledge sharing and remote computing for the mobile robots. The main concept of this architecture consists in keeping basic navigation functionalities on the mobile robots to make them temporal independent of the cloud server and provide long-term navigation capabilities through globally coherent navigation solutions running on the server side. Thereby, the mobile robots do not rely on low-latency or high-frequency server information and are able to maintain their operational capability in the presence of network disruptions. Since the availability of upto-date map information is of crucial importance for both localization and path planning when facing dynamic and highly changing environments, the thesis’ main focus consists in leveraging the shared sensor observations to build and maintain an up-to-date global map. Consequently, the thesis proposes a cooperative long-term simultaneous localization and mapping (LT-SLAM) approach where each mobile robot shares its detected map changes with the cloud-based LT-SLAM server, which fuses the incoming map information into a consistent global map and provides map updates to the robots. A further emphasize of this work consists in deriving a map representation specifically tailored for this application and its manifold requirements. More concretely, we introduce low-resolution dynamic occupancy grid maps in combination with a cell-wise continuous representation of the object within the cell modeling its contour and reflectivity using mixture models. We then show how this map representation is integrated into a local LT-SLAM approach running on each robot and providing high-frequency localization estimates as well as local map updates. Additionally, we demonstrate how the cooperative functionalities can be seamlessly added in terms of cooperative map updating and cooperative localization by mutual detection. Numerous simulative and real-world experiments demonstrate both effectiveness and practicability of the cooperative LT-SLAM approach in terms of increased localization robustness and accuracy, improved navigation efficiency with reduced travel times as well as reasonable network loads.
Cloud-based Cooperative Long-Term SLAM for Mobile Robots in Industrial Applications
2020-01-01
Fraunhofer IPA
Theses
Electronic Resource
English
DDC: | 629 |
An Adaptive Square Root Cubature Kalman Filter based SLAM Algorithm for Mobile Robots
British Library Conference Proceedings | 2015
|2D Lidar-Based SLAM and Path Planning for Indoor Rescue Using Mobile Robots
DOAJ | 2020
|A Fast and Adaptive SLAM Algorithm for Mobile Robots in Three-dimensional Space
British Library Online Contents | 2012
|