多策略人工兔算法优化粒子滤波的SLAM精度研究

Research on SLAM accuracy of multi-strategy artificial rabbits algorithm optimized particle filter

针对传统粒子滤波算法(particle filter,PF)重采样导致粒子贫乏,以及需增加粒子数提高估计精度的问题,提出一种基于多策略人工兔算法优化的粒子重组滤波算法。引入中垂线算法提高人工兔算法收敛速度,通过其觅食与隐藏机制,使得最优粒子引导粒子集向高似然区域移动,以此提高估计精度;实时计算最优粒子附近的粒子密度,当密度大于设置的阈值时,自适应调整迭代次数,当大于最大密度值时,引入自扰动策略避免陷入局部最优以及增加样本多样性;重采样阶段,将筛选后保留的粒子与剩余粒子重新组合成新的粒子,以此增加粒子多样性。通过仿真检验改进算法在SLAM中的性能,结果表明:该算法与其他3种算法相比,位姿与路标估计精度更高,鲁棒性更佳。

The resampling of traditional particle filter algorithm causes particle scarcity and needs to increase the number of particles to improve the estimation accuracy.In this paper,a particle recombination particle filter algorithm based on multi-strategy artificial rabbits optimization algorithm is proposed.First,the introduction of the median line algorithm improves the convergence speed of the artificial rabbit algorithm,and through its foraging and hiding mechanism,the optimal particle guides the particle set to move towards the high likelihood region,thereby improving estimation accuracy.Second,the particle density near the optimal particle is calculated in real-time.Adaptive adjustment is made to the number of iterations when the density exceeds the set threshold whereas a self disturbance strategy is introduced to increase sample diversity when it exceeds the maximum density value.Finally,the resampling stage recombines the retained particles and the remaining particles into new particles to increase particle diversity.The performance of the improved algorithm in SLAM is verified through simulation experiments.The results show the algorithm achieves higher accuracy and better robustness in pose and landmark estimation compared with other three algorithms.