传统粒子滤波算法的单次迭代过程以及小权值粒子在重采样中被删除都使得机器人位姿的历史信息不能充分利用,因而会出现粒子的退化现象,从而导致滤波算法的估计精度较低.本文提出基于精确稀疏扩展信息滤波的粒子滤波SLAM算法,利用精确稀疏扩展信息滤波的信息矩阵反映机器人位姿相对变化的同时,也对应于状态后验概率的条件概率的性质,应用Gibbs采样直接从SLAM完全后验分布产生样本,充分利用了信息矩阵包含的不确定信息,粒子分布均匀,且保持了多样性,缓解了粒子退化现象.实验结果表明所提算法的粒子集能够更好地描述真实后验分布,显著提高了SLAM算法的估计精度.%Historical information can not be fully utilized because of the small weight particle removed in resampling and the single iteration of the particle filter algorithm, thus there is degradation of the particles, and the estimation accuracy of the filtering algorithm is low. A SLAM algorithm based on exactly sparse extended information filter is put forward, the nonzero elements in a natural sparse information matrix of the exactly sparse extended information filter not only reflect the relative variations, but also correspond to the conditional probability of posterior probability related to the robot's state. And with the help of the Gibbs sampling, a new sample occurs from the SLAM complete posterior distribution. Then uncertain information included in the information matrix is made full advantage of to lower any degradation possibility of the samples, keep the diversity of particle, and ease particle degradation. The results show that, the particle set gained from the above can describe the real posterior distribution in detail and improve the accuracy of the calculation of our SLAM algorithm.
展开▼
