模拟退火方差缩减的FastSLAM算法在AUV导航定位中的应用

摘要

The FastSLAM algorithm based on variance reduction of particle weight was presented in order to solve the decrease of estimated accuracy of AUV ( autonomous underwater vehicle) , location due to particles degeneracy and the sample impoverishment as a result of resampling in standard FastSLAM. The variance of particle weight was decreased by generating an adaptive exponential fading factor, which came from the thought of cooling function in simulated annealing. The effective particle number was increased by application of FastSLAM based on simulated annealing variance reduction in navigation and localization of AUV. Resampling in standard FastSLAM was replaced with it. Establish the kinematic model of AUV, feature model and measurement models of sensors, and make feature extraction with Hough transform. The experiment of AUV's simultaneous localization and mapping u-sing simulated annealing variance reduction FastSLAM was based on trial data. The results indicate that the method described in this paper maintains the diversity of the particles, however, weakens the degeneracy, while at the same time enhances the accuracy stability of AUV's navigation and localization system.%由于标准FastSLAM中存在粒子退化及重采样引起的粒子贫化,导致自主水下航行器(AUV)位置估计精度严重下降的问题,提出了一种基于粒子权值方差缩减的FastSLAM算法.利用模拟退火的降温函数产生自适应指数渐消因子来降低粒子权值的方差,进而增加有效粒子数,以此取代标准FastSLAM中的重采样步骤.建立AUV的运动学模型、特征模型及传感器的测量模型,通过霍夫变换进行特征提取.利用方差缩减FastSLAM算法,基于海试数据进行了AUV同步定位与构图仿真试验,结果表明所提方法能够保证粒子的多样性,并且降低粒子的退化程度,提高了AUV定位与地图构建系统的准确性及稳定性.