Visual laser-SLAM in large-scale indoor environments

Loop closure is a well-known problem in the research of laser based simultaneous localization and mapping, especially for applications in large-scale environments. The cumulative errors in the estimated pose and map make the loop detection difficult, no matter using particle filter-based or graph-based SLAM methods. Camera has the advantage of rich information but suffers from short distance and relative high computation burden. In this paper, we proposed a novel approach to address the loop closure problem in large-scale laser-SLAMs, where both laser and camera sensors are integrated. ORB features and bags-of-word were applied to obtain fast and robust performance of loop detection. The well-recognized LRGC SLAM framework and SPA optimization algorithm were then used to achieve the SLAM. Finally, several experiments in different large-scale environments were performed to verify the effectiveness of the proposed approach.

I. Introduction

Laser based simultaneous localization and mapping (SLAM) (e.g. [1]–[3]) is one of the most popular topics in robotic perception research, due to laser's advantages of high precision, long distance and obstacle avoidance ability. In many practical applications, robots need to simultaneously localize the pose and construct a map of the unexplored environments. As shown in Fig. 1, applications include service robots, industrial unmanned ground vehicles, and unmanned air vehicles, etc. In the past decades, many approaches have been developed to implement laser based SLAM, like EKF-SLAM[1], GMapping [2], and Graph-based SLAM [4]. These approaches are efficient in small-scale environments. However, they suffer from loop detection problem in large-scale environments, especially in environments with symmetries or geometrical simplicity.