优化的多假设FastSLAM数据关联方法

摘要

Aiming at data association problems about robot Simultaneous Localization and Mapping(SLAM), this paper presents an optimized multiple hypothesis data association approach to SLAM problem based on FastSLAM. This method uses dynamic threshold to reduce the number of probable data association solution. After that, FastSLAM is combined with multiple hypothesis data association method, and a new particle weight is composed with the product of FastSLAM particle weight and multiple hypothesis data association posterior distribution probability. FastSLAM resample is used to reduce relevance trees compounding. Experimental result indicates that it can modify error data association solution in certain degree compared with JCBB, complete data association well, and improve the accuracy for localization and mapping.%针对机器人同时定位与地图创建(SLAM)中的数据关联问题，基于FastSLAM提出一种多假设数据关联方法，采用动态阈值滤波方法确定每个观测的可能数据关联解，结合多假设数据关联方法和 FastSLAM方法，将 FastSLAM的权重与多数据关联假设数据关联的后验概率乘积构成新的粒子权重，用 FastSLAM 的重采样进行假设组合的关联树剪枝，实现多假设数据关联。实验结果表明，与JCBB方法相比，该方法能修正错误的数据关联，提高机器人定位和建图的精度。