Robust loop closure method for multi-robot map fusion

Abstract

For multi-robot systems to efficiently coordinate and collaborate during missions, it is essential for the robots to estimate their poses and map the surrounding environment. Such an approach is commonly known as multi-robot Simultaneous Localization and Mapping (SLAM). However, depending on the nature of the mission, it is possible that the relative poses between the local coordinates of the robots are unknown. In such scenarios, the robots have to build maps in their local coordinates first and then merge the maps by inferring the relative poses between the local coordinates during or after the exploration. The inference is achieved through inter-robot loop closures, which are measurement constraints between two robot trajectories. These are perception-derived measurements that can be obtained from two robots observing the same place. However, as perception-derived measurements rely on the similarity of two instances of sensor data, two different places could be wrongly identified as the same location if they exhibit similar appearances. This phenomenon, called perceptual aliasing, can be observed in environments with repetitive patterns, such as indoor settings, and it produces false loop closures, thereby leading to catastrophic failure to the SLAM system. In this thesis, we propose a robust inter-robot loop closure selection method that rejects outlier measurements through maximum weight clique considering both the consistency of the loop closures and the similarity between the sensor data associated with the loop closure. The algorithm is tested on an experimental dataset to validate the performance of the method and the result is discussed in comparison to a state-of-the-art algorithm.