• 제어로봇시스템학회:학술대회논문집
• /
• 1986.10a
• /
• pp.269-274
• /
• 1986

A continuous-time modified gain extended Kalman filter (MGEKF) is developed in an effort to extend the discrete-time results of 1) and 2). Used as an observer, it is globally exponentially convergent. For stochastic system, the stability of the MGEKF is proven under certain conditions. The performance of the MGEKF is compared with that of the EKF for a particular nonlinear system where the fininate dimensional optimal filter exists.

• Proceedings of the KIEE Conference
• /
• 2004.11c
• /
• pp.349-351
• /
• 2004

The classical dynamic backpropagation learning algorithm has the problems of learning speed and the determine of learning parameter. The Extend Kalman Filter(EKF) is used effectively for a state estimation method for a non linear dynamic system. This paper presents a learning algorithm using Dual Extended Kalman Filter(DEKF) for Fully Recurrent Neural Network(FRNN). This DEKF learning algorithm gives the minimum variance estimate of the weights and the hidden outputs. The proposed DEKF learning algorithm is applied to the system identification of a nonlinear SISO system and compared with dynamic backpropagation learning algorithm.

• Journal of Institute of Control, Robotics and Systems
• /
• v.12 no.8
• /
• pp.780-788
• /
• 2006

This paper deals with INS/GPS tightly coupled integration algorithms using extend Kalman filter (EKF) and unscented Kalman filter (UKF). In the tightly coupled approach, nonlinear pseudorange measurement models are used for the INS/GPS integration Kalman filter. Usually, an EKF is applied for this task, but it may diverge due to poor functional linearization of the nonlinear measurement. The UKF approximates a distribution about the mean using a set of calculated sigma points and achieves an accurate approximation to at least second-order. We introduce the generalized scaled unscented transformation which modifies the sigma points themselves rather than the nonlinear transformation. The generalized scaled method is used to transform the pseudo range measurement of the tightly coupled approach. To compare the performance of the EKF- and UKF-based tightly coupled approach, real van test and simulation have been carried out with feedforward and feedback indirect Kalman filter forms. The results show that the UKF and EKF have an identical performance in case of the feedback filter form, but the superiority of the UKF is demonstrated in case of the feedforward filer form.

• Proceedings of the KIEE Conference
• /
• 2008.10b
• /
• pp.323-324
• /
• 2008

In this paper, we present performance comparison of two methods to localize multiple robots. One is extended Kalman filter and the other is constraint propagation technique. Extended Kalman filter is conventional probabilistic method which gives the sub-optimal estimation rather than guarantee any boundary for true position of robot. In case of constraint propagation, it can give a boundary containing true robot position value. Especially, we deal with cooperative localization problem in outdoor environment for multiple robots equipped with GPS, gyro meter, wheel encoder. In simulation results, we present strength and weakness for localization methods based on extend Kalman filter and constraint propagation technique.

• Smart Structures and Systems
• /
• v.17 no.6
• /
• pp.903-915
• /
• 2016

The classical Kalman filter (KF) provides a practical and efficient state estimation approach for structural identification and vibration control. However, the classical KF approach is applicable only when external inputs are assumed known. Over the years, some approaches based on Kalman filter with unknown inputs (KF-UI) have been presented. However, these approaches based solely on acceleration measurements are inherently unstable which leads poor tracking and so-called drifts in the estimated unknown inputs and structural displacement in the presence of measurement noises. Either on-line regularization schemes or post signal processing is required to treat the drifts in the identification results, which prohibits the real-time identification of joint structural state and unknown inputs. In this paper, it is aimed to extend the classical KF approach to circumvent the above limitation for real time joint estimation of structural states and the unknown inputs. Based on the scheme of the classical KF, analytical recursive solutions of an improved Kalman filter with unknown excitations (KF-UI) are derived and presented. Moreover, data fusion of partially measured displacement and acceleration responses is used to prevent in real time the so-called drifts in the estimated structural state vector and unknown external inputs. The effectiveness and performance of the proposed approach are demonstrated by some numerical examples.

• Smart Structures and Systems
• /
• v.18 no.3
• /
• pp.375-387
• /
• 2016

The classical Kalman filter (KF) can provide effective state estimation for structural identification and vibration control, but it is applicable only when external inputs are measured. So far, some studies of Kalman filter with unknown inputs (KF-UI) have been proposed. However, previous KF-UI approaches based solely on acceleration measurements are inherently unstable which leads to poor tracking and fictitious drifts in the identified structural displacements and unknown inputs in the presence of measurement noises. Moreover, it is necessary to have the measurements of acceleration responses at the locations where unknown inputs applied, i.e., with collocated acceleration measurements in these approaches. In this paper, it aims to extend the classical KF approach to circumvent the above limitations for general real time estimation of structural state and unknown inputs without using collocated acceleration measurements. Based on the scheme of the classical KF, an improved Kalman filter with unknown excitations (KF-UI) and without collocated acceleration measurements is derived. Then, data fusion of acceleration and displacement or strain measurements is used to prevent the drifts in the identified structural state and unknown inputs in real time. Such algorithm is not available in the literature. Some numerical examples are used to demonstrate the effectiveness of the proposed approach.

• Journal of the Korean Society for Aviation and Aeronautics
• /
• v.21 no.2
• /
• pp.33-39
• /
• 2013

Multilateration(MLAT) may complement secondary surveillance radar and also act as a real-time backup for the ADS-B system. This System is using time difference of arrival (TDOA) and based on triangulation principle. Each TDOA measurement defines a hyperbola describing possible aircraft locations. The accuracy in MLAT system depends on the positional relationship of the receiver and aircraft. There are various algorithms to localize aircraft based on TOA estimation. In this paper, we use least square method and extended Kalman filter and compare their results. Study results show that the extend Kalman filter provides a better performance than the least square method.

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• v.1
• /
• pp.103-107
• /
• 2006

At present, most efforts tend to develop a INS which is only based linear accelerometers, because of the low cost micro-machining gyroscopes lack of the accuracy needed for precise navigation application and possible achieving the required levels of precise for micro-machining accelerometer. Although it was known in theory that a minimum of six accelerometers are required for a complete description of a rigid body motion, and any configuration of six accelerometers (except for a "measure zero " set of six-accelerometer schemes) will work. Studies on the feasible configuration of GF-INS indicate that the errors of angular velocity resolved from the six accelerometers scheme are diverged with time or have multi solutions. The angular velocity errors are induced by the biases together with the position vectors of the accelerometers, therefore, in order to treat with the problem just mentioned, researchers have been doing many efforts, such as the extra three accelerometers or the magnetometers may be taken as the reference information, the extended Kalman filter (EKF) involved to make the angular velocity errors bound and be estimated, and so on. In this paper, the typical configurations of GF-INS are introduced; for each type GF-INS described, the solutions to the angular velocity and the specific force are derived and the characteristic is indicated; one of the corresponding extend Kalman filters are introduced to estimate the angular errors; parts of the simulation results are presented to verify the validity of the equations of angular velocity and specific force and the performance of extend Kalman filter.

• International Journal of Aeronautical and Space Sciences
• /
• v.17 no.2
• /
• pp.222-231
• /
• 2016

This paper proposes a new method achieving computationally efficient attitude reference system for low cost strapdown sensors and microprocessor platform. The main idea in this method is to define and compare velocity differential vectors, geometrically computed from INS and GPS data with different update rate, for generating attitude error measurements which is further used for filter construction. A quaternion based Kalman filter configuration is applied for the attitude estimation with the adapted measurement model of differential vector comparison. Linearized model for Extended Kalman Filter and low pass filtered characteristics of measurement greatly extend the affordability of the proposed algorithm to the field of simple low cost embedded systems. For performance verification, experiment are done employing a practical low cost MEMS IMU and GPS receiver specification. Performance comparison with a high grade navigation system demonstrated good estimation result.

• Proceedings of the KIEE Conference
• /
• 2005.10b
• /
• pp.86-88
• /
• 2005

An autonomous semantic-map building method is proposed, with the robot localized in the semantic-map. Our semantic-map is organized by objects represented as SIFT features and vision-based relative localization is employed as a process model to implement extended Kalman filters. Thus, we expect that robust SLAM performance can be obtained even under poor conditions in which localization cannot be achieved by classical odometry-based SLAM