融合遗传优化的粒子滤波器算法

Particle Filter Algorithm Based on Genetic Optimization Method

为了解决基于Rao-Blackwellized粒子滤波器的同时定位与地图创建算法需要大量的采样粒子,而且频繁重采样可能导致粒子耗尽的问题,提出了融合遗传优化的粒子滤波器算法。设计了一种变异的遗传算法来兼顾粒子的权值和粒子集的多样性,取代原有的重采样步骤。在计算采样的提议分布时考虑了里程计信息和距离传感器信息,并且通过遗传算法来维持粒子集的多样性。实验结果表明,融合遗传优化的粒子滤波器算法在估计精度和一致性方面都具有较好的性能,所创建的地图具有更高的精度。

In order to solve the problem that simultaneous localization and mapping （SLAM） algorithm based on Rao-Blackwellized particle filters needs a large number of particles and the frequent resampling might lead to the particle impoverishment, this paper proposes a kind of particle filter based on genetic optimization method. In order to combine the particle weight and the diversity of samples, the paper designs an improved genetic algorithm to replace resampling. It takes into account both the odometer and the observed information when computing the proposal distribution, through the genetic algorithm to keep the diversity of samples. The experimental results show that the new particle filter algorithm performs well on both estimation accuracy and consistency as well as builds maps with higher accuracy.