基于果蝇算法优化的移动机器人RBPF-SLAM研究

Research on RBPF-SLAM for mobile robot based on fruit fly optimization algorithm

针对基于传统Rao-Blackwellized粒子滤波(Rao-Blackwellised Particle Filter,RBPF)算法的移动机器人在进行同时定位与地图构建(Simultaneous Location and Mapping,SLAM)时易发生粒子退化导致移动机器人位姿估计不准确以及地图一致性较差的问题,提出一种基于果蝇优化算法的RBPF-SLAM算法。该算法将果蝇种群觅食过程中果蝇具有的趋味特性引入RBPF算法,将粒子视为果蝇个体,粒子的适应度值作为空气中食物味道的浓度,利用果蝇优化算法的高寻优能力使粒子向高似然区域移动并不断迭代寻优,以优化粒子种群的整体分布。同时,在果蝇寻优后的新种群中引入自适应交叉变异操作,以增加种群多样性。根据粒子的适应度值确定交叉概率,对配对好的粒子进行自适应交叉操作,再根据变异概率对当前种群的最优粒子进行变异操作,选取适应度值更高的粒子作为当前最优解。采用指数函数步长公式更新粒子状态,增加寻优过程中的搜索距离,有效提高算法的收敛效率。基于ACES building和MIT Killian Court数据集的仿真实验以及移动机器人实机测试结果显示,基于果蝇优化算法的RBPF-SLAM算法在比传统RBPF-SLAM算法在粒子数减少50%以上的情况下仍可以得到效果更佳的栅格地图,并且CPU占用率更低。仿真和实验结果表明基于果蝇优化算法的RBPF-SLAM算法有效提高了滤波器的估计性能,是一种提高移动机器人位姿估计和建图精度的有效方法。

To solve the problem of mobile robot position estimation inaccuracy and poor map consistency due to particle degradation in simultaneous location and mapping(SLAM)of mobile robot based on traditional Rao-Blackwellized particle filter(RBPF)algorithm,a RBPF-SLAM algorithm based on fruit fly optimization algorithm was proposed.According to the algorithm,the taste-tending characteristics of fruit flies during foraging were introduced into RBPF algorithm,where each particle was treated as a fruit fly individual and the particle’s fitness value was considered as food flavour concentration in the air.The particles were moved towards high likelihood region and iteratively optimised depending on the high search capability of fruit fly optimization algorithm to optimize the population distribution.Meanwhile,an adaptive cross-variation operation was introduced in the new population after optimization to increase the population diversity.Initially,the crossover probability was determined by particle’s fitness value to perform an adaptive crossover operation on the paired particle,and then the optimal particle of current population was subjected to a mutation operation based on the mutation probability,and the particle with a higher fitness value was selected as the current optimal solution.In the end,the exponential function step formula was adopted to update particle's states and enhance the search distance in seeking superiority,which effectively boosted convergence efficiency of the algorithm.The results of simulation experiments based on the ACES building and MIT Killian Court datasets as well as real-world tests on mobile robot show that compared with traditional RBPF-SLAM algorithm,the RBPF-SLAM algorithm based on fruit fly optimization algorithm can generate better raster maps with more than 50%reduction in the number of particles and a lower CPU usage.Simulation and experimental results indicate that the RBPF-SLAM algorithm based on fruit fly optimization algorithm can effectively enhance the filter estimation performance,which is an alternative method to improve the accuracy of mobile robot pose estimation and mapping.