Multisensor Out of Sequence Data Fusion for Estimating the State of Discrete Control Systems

Abstract

Abstract-The fusion center of a complex control system estimates its state with the information provided by different sensors. Physically distributed sensors, communication networks, pre-processing algorithms, multitasking, etc, introduce non-systematic delays in the arrival of information to the fusion center, making the information available Out-Of-Sequence (OOS). For real-time control systems, the state has to be efficiently estimated with all the information received so far. So, several solutions of the OOS problem for dynamic Multiple-Input Multiple-Output (MIMO) discrete control systems traditionally solved by the Kalman Filter (KF) have been proposed recently. This paper presents two new streamlined algorithms for the linear and non-linear case. IFAsyn, the linear algorithm, is equivalent to other optimal solutions but more general, efficient and easy to implement. EIFAsyn, the nonlinear one, is a new solution of the OOS problem in the Extended KF (EKF) framework.

Particle filter for nonlinear systems with multi-sensor asynchronous random delays

“…The delay probabilities are set as follows: p 0 t,1 = p 0 t,2 = 1 when t = 1; p 0 t,1 = p 0 t,2 = 0.7, p 1 t,1 = p 1 t,2 = 0.3 when t = 2; p 0 t,1 = 0.5, p 1 t,1 = 0.3, p 2 t,1 = 0.2, p 0 t,2 = 0.6, p 1 t,2 = 0.3, p 2 t,2 = 0.1 when t > 2. Four algorithms are used to estimate the state sequence for t = 1, 2, .…”

“…Generally speaking, the filtering problems for systems with measurement delay can be divided into two categories according to whether the measurement packet is time-stamped or not. For time-stamped cases, it is exactly known when the current received measurement is collected by the sensor, and many filters have been developed [1,5,6].…”

Localization of Non-Linearly Modeled Autonomous Mobile Robots Using Out-of-Sequence Measurements

“…Although this approach is applied with success in [ 26 ], where the OOS 1-step lag [ 5 ]-A1 KF is used for tracking autonomous vehicles with visual information only delayed 1 time step, it can produce erroneous results when the robot dynamic and sensorial models have strong non-linearities. OOS Extended KF/IF ([ 27 – 31 ]): They extend some OOS KF/IF to make them deal with the non-linearities of the problem: - [ 27 , 28 ]-EB1 present the extension of the retrodiction OOS KF in [ 32 ]-B1 to the non-linear case. This OOS EKF can work with non-linear measurement functions h s ( x t k , t k ), although its transition function f ( x t i , u t k , t i , t k , t i ) has to be linear ( F t k , t i x t i + u t k , t i ), due to the retrodiction (backward propagation) operation used in [ 32 ]-Bl.…”

“…Besides, the Extended OOS KFs presented in [ 29 ], and called [ 29 ]-EAl1, [ 29 ]-EBl1, [ 29 ]-EFPFD hereafter, can produce erroneous results when the robot dynamic and sensorial models have strong non-linearities. - [ 30 , 31 ]-EIFAsyn extends the forward OOS IF presented in [ 30 , 31 ]-IFAsyn to work with systems with linear and/or non-linear transition and sensorial models. Besides, in order to overcome the difficulties that the former non-linear OOS algorithms face in systems with strong non-linearities, it distinguish two types of non-linear sensors: those whose information needs to be recalculated when older measurements arrive at the localization module and those whose information do not need to be recalculated.…”

Gaussian particle filtering algorithm for out-of-sequence-measurement problem

“…out of sequence measurement OOSM [1][2] OOSM OOSMs OOSM OOSM OOSM 1993 Hilton OOSM B1 [3] Bar-shalom A1 [2] B1 OOSM OOSM [4][5][6] Zhang Mahendra Zl Ml Bar-shalom [5] A1 B1 Al1 Bl1 Zl Ml OOSM [7][8][9] Shen [7] Zl Bl1 Zhang [5] Besada-Portas [8] IFAsyn Challa [9] OOSM OOSM OOSMs [10] OOSMs…”