Randomly Weighted CKF for Multisensor Integrated Systems

Abstract

The cubature Kalman filter (CKF) is an estimation method for nonlinear Gaussian systems. However, its filtering solution is affected by system error, leading to biased or diverged system state estimation. This paper proposes a randomly weighted CKF (RWCKF) to handle the CKF limitation. This method incorporates random weights in CKF to restrain system error’s influence on system state estimation by dynamic modification of cubature point weights. Randomly weighted theories are established to estimate predicted system state and system measurement as well as their covariances. Simulation and experimental results as well as comparison analyses demonstrate the presented RWCKF conquers the CKF problem, leading to enhanced accuracy for system state estimation.

A novel improved cubature Kalman filter with adaptive generation of cubature points and weights for target tracking

“…In step 3, instead of √ n in equation ( 14), the cubature points can be calculated by the coefficients Σ/λρ i or Σ/(2 − λ)ρ i in equation (19), which can reduce the estimation error caused by the nonlocal sampling problem. The value of the cosine similarity i indicates that the cubature point is closer to or farther from the center point, and we can assign different weights to different cubature points based on the value.…”

“…Remark 2. In equation (19), the parameter λ distributes the probability mass corresponding to over two sets of volume points, doubling the number of sigma points around the mean. When λ = 1, the two sets of cubature points (4n) will overlap so that only 2n cubature points remain, and the NCR degenerates into the classical CKF.…”

“…Besides, in table 1, the determination of the cubature points (sampling points) and weights has two main problems: (a) since the cubature points are calculated by mul-tiplying, when the dimension is high, the distance between the points and the center point will become larger, which makes the filter prone to non-local sampling problems; (b) all cubature points have the same weight (1/2n), and they have the same contribution to the evaluations of prediction errors. In this case, the prediction and error covariance obtained from all cubature points cannot accurately reflect the actual statistical characteristics of the system error, which affects the estimation accuracy of the filter [17][18][19].…”

Modified Sage‐Husa Adaptive Kalman Filter‐Based SINS/DVL Integrated Navigation System for AUV

“…To solve this problem, the interacting multiple model algorithm which uses more than one model is proposed in [11], where a variable model set based on the model probability weighted average of the model parameter is generated to pursue the real model. Considering the nonlinearity of the model of SINS/DVL, the square root unscented information filter is designed in [12], the randomly weighted cubature Kalman filter is discussed in [13], and the unscented Kalman filter (UKF) is employed in [14][15][16]. Furthermore, considering the influence of unknown environment and the inexact error model caused by model simplification, various adaptive Kalman filters are proposed in [17][18][19][20], where statistical characteristics of noises are online estimated in [17][18][19] and the recursive filtering gain is adaptively adjusted in [20].…”

Combination of IMM Algorithm and ASTRWCKF for Maneuvering Target Tracking

“…In Ref. [20], a randomly weighted cubature Kalman filter (RWCKF) is proposed by using randomly weighted factor instead of original factor to reduce the influence of the system noise. Compared with CKF, RWCKF has higher tracking accuracy for nonlinear systems, which provides a new way to solve the tracking problem of nonlinear maneuvering targets [21][22][23].…”