A real-time map merging strategy for robust collaborative reconstruction of unknown environments

Abstract

The development of collaborative techniques for exploring and mapping environments has been rising in the last decade. These techniques, known as multi-robot SLAM (MRSLAM), aim to extend the use of autonomous mobile robots to autonomous multi-agent systems. The MRSLAM technique presented here consists mainly of a robust map merging algorithm and a decision-making algorithm that controls agents in the field. On the one hand, the proposed merging algorithm performs a consistent and robust map fusion in real time. It consists of an own corner detector, a cylindrical descriptor, a matching technique and the RANSAC algorithm. On the other hand, once the fusion of maps is performed, the decision-making algorithm is responsible for controlling the robot operation in the field, based on the general current state of the multi-robot system. The main contribution of this MRSLAM technique is the robust map merging algorithm, since it was implemented and validated in simulated and real scenarios, resulting in collaborative maps that are consistent with the environment and obtained in less than 280 ms. This technique also achieves a significant decrease in reconstruction time when two or three robots are used: up to 35% in a simulated scenario and up to 49% in a real one. The proposed MRSLAM technique shows important similarities to expert multi-agent systems, as it is able to control and organize a team of robots in order to collaboratively explore and map an unknown environment. This approach was developed under the ROS framework to be used and tested by the scientific and academic community.