基于PMCMC-RFS的自主车SLAM算法

Autonomous Vehicle SLAM Algorithm Based on PMCMC-RFS

针对无人自主车同时定位与地图构建(Simultaneous Localization And Mapping,SLAM)问题,采用随机有限集(Random Finite Set,RFS)方法对环境特征和车辆的位姿进行描述,将SLAM算法涉及到的多路标特征检测、跟踪、识别及相关等问题在一个统一的贝叶斯状态估计框架内表述,从而可以有效地解决后验估计、信息融合等算法严重依赖数据关联结果的问题。同时,为了计算复杂的联合后验分布,解决粒子滤波算法中提议分布选择困难问题,采用序贯蒙特卡罗(Sequential Monte Carlo,SMC)算法为马尔科夫链蒙特卡罗(Markov Chain Monte Carlo,MCMC)采样构建高维提议分布策略,提出了基于PMCMC-RFS(Particle MCMC based RFS)的SLAM问题求解方法。试验结果表明:PMCMC-RFS算法能动态估计感知范围内的特征数量,有效地避免了数据关联问题,从而提高了状态估计性能。

Aiming at the problem of Simultaneous Localization And mapping (SLAM)for autonomous ve-hicles,the environment feature as well as vehicle position and attitude are described by using Random Fi-nite Set (RFS),which integrates the aspects such as multi-feature detection,tracking,recognition and correlation related to SLAM algorithm into a single unified Bayesian framework so that the thorny problem that the algorithm of posterior estimation and information fusion relies on the result of association greatly can be effectively solved.Meanwhile,to calculate the complex joint posterior estimation distribution and resolve the proposal distribution selection difficulties in particle filtering algorithm,SLAM resolving meth-od based on Particle Markov Chain Monte Carlo (PMCMC)is proposed using Sequential Monte Carlo (SMC)and Markov Chain Monte Carlo (MCMC).The autonomous ground vehicle experimental results shows that the proposed method can estimate in dynamic manner the feature numbers of perspective scope and has better performance in states estimation without consideration of data association.