一种鲁棒闭环的增量式Graph SLAM算法

摘要

针对常规增量式Graph SLAM算法的后端优化无法高效排除错误闭环影响的问题，基于iSAM算法和SC算法，提出一种鲁棒闭环的增量式Graph SLAM算法R-iSAM。 R-iSAM在增量式过程中对当前时刻引入的闭环约束的转换变量进行初步近似计算，得到合理的机器人节点位姿。在离线式过程中对当前时期的所有闭环约束转换变量进行精确计算，判断当前时期闭环的正确性，并作为以后优化节点的基础。对公开的数据集进行的算法实验表明，在添加不同类型、不同数量的错误闭环条件下，所提算法对不同数据集具有良好适应性，且收敛速度满足增量式SLAM实时性要求，证明了算法的有效性。%In view of the problem that the back-end optimaziation for conventional incremental Graph SLAM alrgorithm cannot efficiently remove the influence of false close-loop, R-iSAM, a robust close-loop incremental Graph SLAM algorithm is put forward based on iSAM algorithm and SC algorithm. In the incremental process, R-iSAM makes a preliminary approximate calculation of the close-loop constrained conversion variables introduced at the current time, obtaining reasonable node positions of robot. In the off-line process, it makes a refined calculation of all close-loop constrained conversion variables in the current period, to judge the correctness of current-period close-loop, and lay a foundation for node optimization later on. The algorithm experiment of public data set ( DS) show that under the condition of adding different types and numbers of fault close-loops, the proposed algorithm is well adaptive to different DS, and the convergence rate satisfies the real-time requirement of incremental SLAM, proving the validity of this algorithm.