摘要： 针对扩展卡尔曼滤波(EKF)算法在移动机器人同时定位和环境建模(SLAM)中的缺点,即非线性系统简单线性化所导致的系统状态方程的不准确性、雅克比矩阵的计算所导致的计算复杂化以及噪声模型不确定性所导致的滤波稳定性降低等问题,提出一种对噪声自适应的UKF-SLAM 算法。该算法通过对噪声缩放进而改变噪声模型,利用观测残差序列准确估计观测噪声模型协方差,运用预测的新息协方差和IAE 开窗法求其系统状态噪声缩放因子,从而准确估计系统状态噪声模型协方差,实现对不确定的噪声模型能够自适应UKF-SLAM 算法。UKF的Sigma 点采样策略是比例对称采样。实验结果证明,该方法相对EKF 算法和UKF 算法具有较高的定位精度和自适应能力。
关键词:
同时定位和环境建模,
无迹卡尔曼滤波,
噪声缩放,
在线自适应,
比例对称采样,
开窗法
Abstract: For Extend Kalman Filtering (EKF) algorithm disadvantage on the Simultaneous Location and Mapping
(SLAM),that is simple linearization of nonlinear systems resulting from the inaccuracy of the system state equation,Jacobimatrix calculation resulting from computational complexity,and noise uncertainty caused by the filtering reduced stability and other issues,this paper proposes a noise adaptive UKF-SLAM algorithm. In order to achieve adaptive UKF-SLAM algorithm,the paper scales the noise to change the noise model. Using the observed innovation sequence to accurately estimate the covariance of the measurement noise model. And using a new message convariance and IAE fenestration to find the system noise scaling factor,and thus accurately setimates the convariance of the system state noise model,it achieves a adaptive UKF-SLAM algorithm. The sampling strategy of UKF Sigma points is scaling symmetric sampling. Experimental results show that the algorithm has a high accuracy on SLAM compared with the EKF-SLAM and UKF-SLAM.
Key words:
Simultaneous Location and Mapping(SLAM),
Unscented Kalman Filtering(UKF),
noise scaling,
online
adaptive,
scale symmetric sampling,
windowing method
中图分类号: