Practical Formation Acquisition Mechanism for Nonholonomic Leader-Follower Networks

Abstract

A grand challenge lying ahead of the realization of multi-robot systems is the lack of an adequate coordination mechanism with reliable localization solutions. In some workspaces, external infrastructure needed for precise localization may not be always available to the MRS, e.g., GPS-denied environments, and the robots may need to rely on their onboard resources without explicit communication. We address the practical formation control of nonholonomic ground robots where external localization aids are not available. We propose a systematic framework for the formation maintenance problem that is composed of a localization module and a control module. The onboard localization module relies on heterogeneity in sensing modality comprised of ultrawideband, 2D LIDAR, and camera sensors. Particularly, we apply deep learning-based object detection algorithm to detect the bearing between robots and fuse the outcome with ultrawideband distance measurements for precise relative localization. Integration of the localization outcome into a distributed formation acquisition controller yields high performance. Furthermore, the proposed framework can eliminate the magnetometer sensor which is known to produce unreliable heading readings in some environments. We conduct several realistic simulations and real world experiments whose results validate the competency of the proposed solution.