BiCR-SLAM: A multi-source fusion SLAM system for biped climbing robots in truss environments

The low-texture, shape-similar, interconnected and mutual-occlusion nature of truss members poses challenges for simultaneous localization and mapping of biped climbing robots in truss environments. In this paper, we propose BiCR-SLAM, a multi-source fusion SLAM system, to estimate both the distinctive state of the robot and a parametric representation of the truss, going beyond traditional point cloud mapping. The proposed system comprises four modules such as encoder dead reckoning, LiDAR odometry, pole landmark mapping, and global optimization. To address the intricacies of truss environments, we present a pole landmark mapping module with dedicated operations including pole detection, data association, and parameterizations. In the back-end, we formulate the localization problem of biped climbing robots using a multi-source factor graph, encompassing factors including forward kinematics, LiDAR odometry, gripping, and points of poles. Experiments are conducted to evaluate the impact of various factors and to validate the effectiveness and accuracy of the proposed BiCR-SLAM system. A handheld LiDAR experiment in an outdoor large-scale truss environment demonstrates the generalization of our proposed approach.