AKAZE Feature-Based Map Merging for Multi-Robot SLAM with Unknown Initial Pose

Abstract

Map merging is an important issue of multi-robot SLAM system, especially in the case of unknown initial pose. This paper presents an efficient algorithm that enables teams of robots to build occupancy grid maps without initial pose. The relative pose transformation between pairs of robots are estimated by the AKAZE features of the overlapping area, which are descripted by Modified Local Difference Binary (MLDB) descriptor. The proposed algorithm reduces the computational complexity and improves the robustness of the search process. The experimental results obtained from two robots under the simulation environment and real environment validated the effectiveness of the proposed method.