다중 로봇 시스템 지도통합을 위한 최대 가중 클릭 기반 강인한 루프 폐쇄 기법

Abstract

For efficient collaborations of multi-robot systems during missions, robots must estimate their poses and map the surrounding environment, which can be achieved through multi-robot SLAM (Simultaneous Localization and Mapping). Depending on the mission, the relative poses between the local coordinates of the robots, which must be inferred to generate a global map, may be unknown. The inference is made using the measurement constraints between the robot trajectories, which are often perception-derived measurements that rely on the similarity of two instances of sensor data. Due to this dependence, perceptual aliasing, which is a phenomenon of wrongly identifying two different places as the same location, may occur and produce false loop closures that lead to a catastrophic failure of the SLAM system. This study proposes a robust inter-robot loop closure selection method that reject outlier measurements by checking both consistency of the loop closure and the similarity between the sensor data associated with the loop closure. By considering these two properties, the correct loop closures can be found despite the lack of prior knowledge regarding the relative poses among robots. We demonstrate herein how this problem can be formulated as a maximum weight clique problem, in which the degree of data similarity associated with the loop closures is considered in the form of weight in the objective function. A simulation was executed to validate the method performance and the results showed that the proposed method outperforms existing methods.