基于动态种群分布的双变异优化FastSLAM算法的改进

Improvement of FastSLAM Algorithm of Double Mutation Optimization Based on Dynamic Population Distribution

针对自适应粒子群FastSLAM算法的特点,设计了一种改进的FastSLAM算法。采用动态种群分布策略来改善粒子分布,并且引入动态双变异算子,将它应用到无人车的FastSLAM算法中,形成一种基于动态种群分布双变异优化算法的FastSLAM算法。通过对粒子种群的实时状态估计来“增减”粒子,并根据大概率的最差适应度变异对惰性粒子重新初始化,增加搜索空间,增加粒子多样性的同时有利于算法的全局收敛。小概率的最优适应度变异加强最优解附近范围的搜索性能,增加种群多样性,使算法有效跳出局部极值。最后在MATLAB仿真平台通过测试函数确定参数并将FastSLAM算法、基于自适应粒子群优化的FastSLAM算法和改进后的算法以无人车的路径偏差和运行时长两个参考量进行了对比,实验结果表明:基于动态种群分布的双变异优化FastSLAM算法在估计精度、搜索性能和运算效率方面在无人车的实际运用中都较为优秀。

Considering FastSLAM algorithm’s characteristics of adaptive particle swarm optimization(APSO), an improved FastSLAM algorithm was designed, which adopts a dynamic population distribution strategy to improve particle distribution and employs a dynamic double mutation operator and applies it to the FastSLAM algorithm of unmanned vehicles to form a new FastSLAM algorithm of the double mutation optimization based on dynamic population distribution. In which, particles were "incremented" by real-time state estimation of the particle population, and the worst-case fitness mutation with a high probability was based to reinitialize the inert particles so as to increase the search space, and this benefitted the global convergence of the algorithm while increasing the particle diversity. The compilation of optimal fitness with small probability strengthened the search performance in the vicinity of the optimal solution and increased the diversity of the population. Finally, in the MATLAB simulation platform, the parameter selection was determined through the test function and the FastSLAM algorithm, the APSO-based FastSLAM algorithm and the improved algorithm were compared with the reference values of the path deviation and running time of the unmanned vehicle. The experimental results showed that, based on dynamic population distribution,the double mutation optimization Fast SLAM algorithm is excellent in the practical application of unmanned vehicles in terms of estimation accuracy, search performance, and computing efficiency.