Swarm robots based SLAM in weak environmental information applications

Abstract

In SLAM, weak environmental information refers to the lack of indications of distinctive landmarks in sensor data. This brings difficulties to SLAM as it relies on the indications. Various approaches of dealing with weak environmental information have been reported. Single-robot-based ones require either large datasets (for running machine learning) or artificial landmarks both of which need the prior knowledge of the environments where robots explore. Multiple-robots-based/swarm-based approaches seem more promising as they rely on information sharing among robots rather than the prior knowledge. They are more robust when facing the failure of individual robots. In the collaborative swarm-based SLAM, individual robots run their own SLAM at local level and contribute to the development of a global map and to the estimation of their locations in the global map. The key to success is the collaboration among the robots. The existing collaboration mechanisms leave SLAM at both local and global levels completely to individual robots, leading to inconsistent global maps. They also require individual robots to have sufficient computational power. This research develops a novel collaborative swarm robot system tailored for global SLAM in weak environmental information scenarios. The system introduces four key advancements: (1) designing a coordination strategy to guide robots toward unexplored regions, improving exploration efficiency and coverage; (2) implementing cross-validation mechanisms to identify and eliminate redundant landmarks from local maps, ensuring data accuracy; (3) enabling individual robots to extend their perception by leveraging peer robots' sensory data, allowing relative localization and integration of local maps in sparse environments; and (4) creating a centralized global map framework to consolidate data from all swarm members, ensuring consistency in the overall mapping process.