Improved cubature Kalman filtering-based estimation of the Jacobian
matrix for environment mutation
Xiuqian Jia
1
, Haixia Wang
1
, Chunyang Sheng*
,1
, Zhiguo Zhang
1
, Wei Cui
1
, Xiao Lu
1
1. College of Electrical Engineering and Automation, Shandong University of Science and Technology, Qingdao10424
E-mail: scy@sdust.edu.cn
Abstract: To improve the accuracy of online estimation of Jacobian matrix (JCM) in image-based visual servo, an improved
cubature Kalman filtering (CKF) algorithm is proposed to estimate the Jacobian matrix online. The key idea of the proposed
method is to introduce the fading factor into the measurement update process, which can improve the capability when coping
with the environment mutation. In addition, to avoid the emergence of the non-positive covariance matrix, the Cholesky
decomposition of the estimation is replaced by the singular value decomposition (SVD). To verify the effect of the proposed
method, two sets of experiments are carried out. Three-dimensional motion tracking is employed to prove tracking effect and the
eye in hand manipulator positioning based on the Puma560 simulation is adopted to testify stability of this method. Simulation
results show that the proposed method is feasible and effective.
Key Words: Image-based visual servo, Jacobian Matrix, Improved CKF, environment mutation
1 Introduction
Comparing with the calibrated visual servo method, the
uncalibrated one has attracted extensive attention, since it
avoids the complexity of calibration calculation, simplifies
the visual servo process and improves the accuracy and
flexibility of the servo system
[1]
. As for image-based visual
servo (IBVS), the given information and feedback
information are both defined in the image feature space. The
control rule is designed based on the difference between
these two images, which can further improve the robustness
of the system.
The essence of IBVS method is how to approximate the
interaction matrix between image feature space and
workspace. And the online estimation of system interaction
matrix (also called Jacobian matrix, JCM) based on filtering
algorithm is one of the most important research branches
[2-5]
.
A modified Kalman filter is reported in [3] for online
estimation of JCM, which makes up for the errors caused by
multiple iterative operations. In addition, a method based on
Sage-Husa adaptive Kalman filtering is proposed in [4] to
modify the measured noise online, which also achieves good
results. However, it is necessary to approximate IBVS, a
highly nonlinear system, to a time-varying linear system to
estimate JCM online by Kalman filtering, which limits the
estimation accuracy. Therefore, aiming for the tracking
problem of eye-in-hand configuration moving target, an
unscented Kalman filter (UKF) is designed for the JCM
online estimation in [5].However, the stability factor of
UKF increases linearly with the increase of dimension when
the dimension is greater than three
[10]
, which leads to the
introduction of truncation error and the decrease of accuracy.
Since CKF has better real-time performance in multiple
estimation, this paper puts forward an improved cubature
Kalman filter algorithm for estimating the JCM online.
Aiming at the environmental mutation that may exist in
IBVS, the fading factor is introduced into the measurement
*
This work was supported by the National Natural Science Foundation
of China ( 61773245, 61603068, 61806113), Shandong Province Natural
Science Foundation (ZR2018ZC0436, ZR2018PF011, ZR2018BF020)
update process, and the estimated covariance matrix is
corrected online with the idea of strong tracking filter under
the framework of CKF. Secondly, for high-order systems,
the covariance matrix might not be decomposed by
Cholesky decomposition due to the loss of positive
characteristics, which affects the convergence and stability
of the algorithm. Seriously, it will lead to the decrease of
estimation accuracy and even the divergence of filtering. In
order to solve the problem that the state covariance matrix of
standard CKF is easy to be non-positive in the process of
recursion, the singular value decomposition is replaced by
the Cholesky decomposition to improve the filtering
stability and simplify the calculation process.
In this paper, a JCM online estimator based on improved
CKF algorithm is designed, which achieves the goal of
fast-tracking ability in the case of system state mutation.
SVD is employed in this study for the covariance matrix in
the iteration process. Finally, two sets of experiments are
carried out to verify the effect of the proposed method.
Three-dimensional motion tracking is employed to prove
tracking effect and the eye in hand manipulator positioning
based on the Puma560 simulation is adopted to testify
stability of this method. Simulation results show that the
proposed method is feasible and effective.
2 Image Jacobian Matrix
The input and output signal are defined in the image
feature space for the IBVS system. The error between the
current image feature and the designed one is taken as the
control object, which forms a closed-loop feedback control
system. The purpose of IBVS control scheme is to minimize
the target error
e
(t)
[6]
, which is usually defined as:
() () ()
dc
ttt ef f
where
()
d
tf
denotes the desired image feature, and
()
c
tf
is
the image feature of the current time t.
In practice, the servo task can be completed by controlling
the manipulator to satisfy the condition that the current
Proceedings of the 38th Chinese Control Conference
Jul
27-30, 2019, Guan
zhou, China
4433