基于改进多新息理论的EKF-SLAM算法

EKF-SLAM algorithm based on improved multi-innovation theory

扩展卡尔曼滤波(EKF)是移动机器人实时定位和地图构建(SLAM)的经典方法。但是当系统为强非线性时,EKF将违背局部线性化假设,导致估计误差增加,并最终降低移动机器人定位和建图的准确性。为了解决上述问题,提出了一种改进MI-EKF-SLAM算法,在每个滤波过程中,将当前时刻的新息向量扩展为过去多个时刻的新息向量,并且将扩展新息向量中每个时刻的观测值都替换为上一时刻的观测值。仿真结果表明,改进算法减少了状态估计误差,实现了更高的定位和建图精度,并且得到改进MI-EKF-SLAM算法中最佳新息长度为3。

Extended Kalman filtering(EKF)is a classic method in simultaneous localization and mapping(SLAM)for mobile robots.However,when the system is strongly nonlinear,EKF will violate the local linear assumption,cause the error to increase,and ultimately reduce the accuracy of positioning and mapping of mobile robots.To solve the above problems,a multi-innovation with Forgetting factor based EKF-SLAM(MI-EKF-SLAM)is investigated.At each filtering step,the MI-EKF-SLAM expand the scalar innovation at current step to an innovation vector containing innovation of current and previous steps,and And the observations at each moment in the extended innovation vector are replaced with the observations from the previous moment.The simulation results show that the explored MIEKF-SLAM reduces the state estimation errors,and achieves higher accuracy in positioning and mapping.And the best innovation length in the improved MI-EKF-SLAM algorithm is 3.