目前开源算法中采取的scanMatching方法主要是以下四种:
- Gmapping:ICP(simple Gradient Descent)
- Hector:Guass-Newton(multi-resolution map)
- karto:Real-time CSM(multi-resolution + 三维窗口遍历寻优)
- cartographer:Fast CSM(multi-resolution + branch and bound)
scanMatcher主要涉及两个评价函数,一个score用于优化调整粒子pose作为参考,一个likelihoodAndScore用于更新粒子权重:
粒子权重根据地图的匹配度更新: