Objectives  The current railway maintenance method is unitary and simple, which is a human force intensive work, requiring tedious and interminable manual check from professional technicians. As a multi-scale, multi-probability and long-term digital representation approach, the digital-twin has seen a rising popularity in engineering construction. Inspired by the convenience of digital-twin, this paper designs a simultaneous localization and mapping (SLAM) system for railway locomotives.

  Methods  Unlike traditional mobile mapping system which demands high precision three-dimensional laser scanning, inertial measurement unit (IMU) / real time kinematic (RTK), as well as complicated post-processing methods, our solution is capable of real time mapping and odometry visualization through constructed factor graph. The factor graph is a bipartite graph with two node types: Factor nodes and variables nodes, and they are always connected by edges. A new variable node is added to the graph when the pose displacements exceed a certain threshold, then the factor graph is optimized upon the insertion. We use three types of factors for graph construction: IMU preintegration factors, light detection and ranging (LiDAR) odometry factors, RTK factors.

  Results and Conclusions  The real time performance is further achieved through sliding window optimization. Under more than 300 km field test in various environments, our approach can achieve centimeter-lever positioning accuracy in feature rich and low speed cases, and the structures are clearly visible in the real time mapping result.