Autonomous Localization and Mapping Method of Mobile Robot in Underground Coal Mine Based on Edge Computing

Abstract

When applying visual SLAM systems to underground coal mines, several challenges arise. First, there are non-ideal texture areas in the scene, which make feature extraction and matching difficult and reduce the accuracy of positioning and mapping. Second, the limited computing resources of mobile robots prevent the real-time execution of complex algorithms. To address these challenges, this paper proposes an edge computing-based SLAM system that fuses point and line features. The visual odometer of point and line feature fusion solves the problem of insufficient feature extraction in texture sparse areas and incorrect feature matching in texture repetitive areas, thereby improving the accuracy of visual positioning and mapping. The distributed deployment strategy of edge computing enables the algorithm to be executed in real-time on the underground coal mine mobile robot. The experiment demonstrated that using the visual odometer method with ORB-SLAM 2 reduced the absolute trajectory error by 8.87% in dense repetitive texture areas and 9.96% in low texture areas.