Vehicle State Information Estimation with the Unscented Kalman Filter

Research Article

Hongbin Ren,1Sizhong Chen,1Gang Liu,1,2and Kaifeng Zheng1

1School of Mechanical Engineering,Beijing Institute of Technology,Beijing100081,China

2College of Mechanical&Electrical Engineering,Shenyang Aerospace University,Shenyang110136,China

Correspondence should be addressed to Sizhong Chen;chensz@https://www.360docs.net/doc/b511710222.html,

Received16August2013;Revised26November2013;Accepted26November2013;Published16January2014

Academic Editor:Yuan Zou

Copyright?2014Hongbin Ren et al.This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use,distribution,and reproduction in any medium,provided the original work is properly cited.

The vehicle state information plays an important role in the vehicle active safety systems;this paper proposed a new concept to estimate the instantaneous vehicle speed,yaw rate,tire forces,and tire kinemics information in real time.The estimator is based on the3DoF vehicle model combined with the piecewise linear tire model.The estimator is realized using the unscented Kalman filter (UKF),since it is based on the unscented transfer technique and considers high order terms during the measurement and update stage.The numerical simulations are carried out to further investigate the performance of the estimator under high friction and low friction road conditions in the MATLAB/Simulink combined with the Carsim environment.The simulation results are compared with the numerical results from Carsim software,which indicate that UKF can estimate the vehicle state information accurately and in real time;the proposed estimation will provide the necessary and reliable state information to the vehicle controller in the future.

1.Introduction

Active safety is all about technical solutions that help the drivers avoid accidents or significantly reduce the possibilities of the accidents.These include braking systems,like ABS, traction control systems,and electronic stability control systems that interpret the signals from various sensors to help drivers control the vehicle or give a warning to driver under various road conditions as well as the driving maneuvers.And the performances of the controller rely heavily on the accurate vehicle states information.Some of the required vehicle state information is easy to measure by the sensors which exist in model vehicle,such as the spin speed of four wheels,but others are difficult to detect due to the expense,complexity, and the technological limits.Reducing sensor is a potential approach to cut down the cost and improve the reliability and performance of the controller.State estimation is an algorithm which prides the internal state of a given system in real time;the algorithm uses the online measurements as inputs and obtains the estimated values.It provides the basis of many practical applications and attracts increasing attention in the field of automobile industry,especially in the vehicle active safety research field[1–4].

Accurate and real-time vehicle state information is nec-essary for the vehicle dynamic controller;there are a lot of researchers who have proposed some approaches to estimate the vehicle state and parameters.Zhao et al.[5]proposed the tire force estimation method by using sliding model observer, and its drawback is that the method calls for the high speed computational hardware because of the complicated and high order vehicle model.Tims[6]investigated the approach to reduce sensors used in active suspension control;he used a relative displacement sensor to estimate vehicle corner sprung mass velocity,but he did not consider the potential for phase advance or lag in the estimation of velocity,especially for the large pitch and roll motion;it may deteriorate the vehicle comfort caused by the signal’s time delay.Wenzel et al.

[7]used the dual extended Kalman filter to estimate vehicle states and parameters;the advantage is that two Kalman filters can run in parallel;which can also work as a signal EKF estimator by simply“turn off”one of them;however,the method is under the hypothesis that the tire-road friction coefficient is known.In[8],Zong et al.applied the dual extended Kalman filter to estimate the tire-road adhesion coefficient based on full vehicle model;nevertheless,it is

Hindawi Publishing Corporation Advances in Mechanical Engineering Volume 2014, Article ID 589397, 11 pages https://www.360docs.net/doc/b511710222.html,/10.1155/2014/589397

difficult to acquire the accurate Jacobi matrix,especially for the high order vehicle model or when the tire forces are working in the saturation zone.

The main contributions of this study are as follows:we proposed a UKF estimator based on a3DoF vehicle model for state estimation of a class of nonlinear systems;this estimation algorithm is to estimate the longitudinal and lateral velocity and the tire characteristics.And considering the limitations of the linear tire model and the complication of the Magic Formula tire model,the paper proposed a piece-wise linear tire model.The paper is organized as follows: firstly,piece-wise linear tire model and the vehicle model are introduced in Sections2and3,respectively.In Section4,the unscented Kalman filter is introduced in detail.In Section5 the estimation results of the KF with the linear bicycle model, UKF with linear bicycle model,and Carsim outputs are presented as the comparison.At last,the discussion and conclusions are given.

2.Piecewise Linear Tire Model

In the simulation of the vehicle dynamic response,the control performances are strongly depending on the precision and accuracy of tire forces.Longitudinal and lateral tire forces are calculated from the tire model.The popularly used tire mod-els are“Dugoff tire model”[9],“Magic Formula”[10],“LuGre tire model”[11],and“Unitire”[12].The Magic Formula tire model is a semiempirical model which is mathematically fitted to experimental data.The approach yields realistic tire behavior;it requires many experimental coefficients and is complicated.When the slip angle or the slip ratio is small, the longitudinal and lateral tire forces are linear to the tire deformation,and most normal driving conditions are in the linear region.The slops of the pure slip curves are defined as the longitudinal and lateral slip stiffness respectively.And when the slip angle or the slip ratio is larger than a certain value,the tire forces are saturated,and the slop tends to0. So the paper proposed the piece-wise function to describe the tire model;it just needs a few coefficients.First the pure longitudinal and lateral sliding tire model is proposed,and then the combined working tire forces are calculated from the pure tire forces.The pure longitudinal and lateral slip working conditions could be fitted by using the piece-wise linear function

F x=F x(s,F z),

F y=F y(α,F z),

(1)

where s is the slip ratio;αis the slip angle;and F z is the vertical tire force.

The longitudinal wheel slip ratio for each individual tire is given by

s wi=1?u wi

w wi

,u wi≤R wωwi,

s wi=R wωwi

u wi

?1,u wi>R wωwi.

(2)

The calculation of the wheel longitudinal slip ratio and

sideslip angle requires the wheel longitudinal speed.These

wheel longitudinal speeds,considering the vehicle lateral

dynamics,are calculated as

u wfl=(u?

)cosδ+(V+ar)sinδ,

u wfr=(u+

)cosδ+(V+ar)sinδ,

u wrl=(u?

u wrr=(u+

(3)

The calculation of the wheel lateral slip angle requires the

wheel lateral speed.These wheel lateral speeds are calculated

V wfl=?(u?

)sinδl f+(V+ar)cosδ,

V wfr=?(u+

)sinδr f+(V+ar)cosδ,

V wrl=V?br,

V wrr=V?br.

(4)

Assuming the small lateral velocity and yaw angular velocity,

the slip angle can be approximately obtained by

αfi=σf?

u wfi

,αri=?

V wrl

u wrl

(i=l,r).(5)

Considering the pitch load transfer and the roll load transfer

is ignored,the normal forces F z at the front and rear tires are

calculated by

F zfl=

m s gb

2(a+b)

m s a x?cg

2(a+b)

m s a y?cg

a+b

F zfr=

m s gb

2(a+b)

m s a x?cg

2(a+b)

m s a y?cg

a+b

F zrl=

m s ga

2(a+b)

m s a x?cg

2(a+b)

m s a y?cg

a+b

F zrr=

m s ga

2(a+b)

m s a x?cg

2(a+b)

m s a y?cg

a+b

(6)

In the beginning,the longitudinal and lateral tire forces F X0

and F Y0are calculated under the assumption that the tire is

in pure longitudinal and lateral slip condition.The coupling

slip condition is taken into account in second step.

2.1.Pure Longitudinal Slip.The longitudinal tire force F X0 under the pure slip condition is calculated by

F X0={{{

{{{

{

C x?s,|s|

C x?s1?sgn(s),s1≤|s|

[C x?s1?

C x?s1?F end

s2?1

?(|s|?s2)]?sgn(s),|s|>s2.

(7)

Here,s1=7s1nom;C x=C x

nom (1+((F z?F z nom)/F z nom));

s2=s2

nom (1+((F z?F z nom)/6F z nom))?2;F end=((F end nom/

F z nom)+((F z?F z nom)/4F z nom))0.8μF z;F end nom=3000 (1+((F z?F z nom)/6F z nom)).

2.2.Pure Lateral Slip.The lateral tire force F Y0under the pure slip condition is calculated by

F Y0={{{

{{{

{

{{{

{

?C y?α,|α|<0.85α0,

C y

?(|α|+4.25α0)

?sng(α),0.85α0≤|α|<1.75α0,

F peak?sng(α),|α|>1.75α0.

(8)

Here,C y=C y

nom (1+(F z/F z

nom

))/2.5;F peak=(0.9?0.182

((F z/F z

nom

)?1))0.5;α0=F peak/C y.

https://www.360docs.net/doc/b511710222.html,bined Working Condition.The longitudinal and lat-eral tire forces in the combined driving and steering condition are given by[13]

δx=?

s+1

,δ?x=

δx

δx max

δy=tan(α)

s+1,δ?y=

δy

δy max

δtotal=√δ2x+δ2y,δ?total=√δ?2x+δ?2y,

s =

δ?total?δx max?sign(δx)

1+δ?

total

?δx max?sign(δx)

μ0

α =arctan(δ?total?δy max?sign(δx))?μ0μ,

F x=F x0?μ?

δx

δtotal

?sign(s ),

F y=F y0?μ?

δy

δtotal

?sign(α ).

(9)

To well approach the Magic Formula tire model,in the linear region,the slope of the piece-wise linear tire model should be similar to the Magic Formula tire model;and in the saturation region,the results of the piece-wise tire model should be approximated to the Magic formulation model.And it should be suitable for the pure condition as well as the combined condition.

00.20.40.60.81

0.2

0.4

0.6

0.8

Slip

(

)

?0.2

?0.4

?0.6

?0.8

?0.2

?0.4

?0.6

?0.8

Piecewise linear tire

Magic Formula tire

×104

Figure1:The longitudinal tire force for various vertical tire forces of2kN,4kN,6kN,and8kN.

05101520 0

2000

4000

6000

8000

aerfa (deg)

(

)

?20?15?10?5

?2000

?4000

?6000

?8000

Piecewise linear tire

Magic Formula tire

Figure2:The lateral tire force for various vertical tire forces of2kN, 4kN,6kN,and8kN.

To validate the preciseness of the piece-wise linear tire model,the comparison results are plotted in Figures1–4. Figures1-2show the comparison of Magic Formula tire forces and the piece-wise tire forces under different vertical forces.It can be found that during the pure longitudinal sliding and pure lateral sliding,the piece-wise linear tire model is accurate(piece-wise linear tire model parameters are shown in Table1).Figures3-4show the comparison of Magic Formula tire forces and the piece-wise tire forces at different slip ratios and slip angles.There are still some errors compared with the Magic Formula tire model,especially

F x (N )

8°,F y (N )

Magic Formula tire

Figure 4:The lateral tire force for various slip ratios of 0,0.2,0.4,0.6,and 0.8(F z =4000N).

when the slip angle or slip ratio is increasing,but it describes well the tire model and can be used in the estimation algorithm due to the small calculation load.

3.Vehicle Model

Figure 5shows the schematic of the vehicle model;it is 3DoF

vehicle model,which consists of the longitudinal,the lateral,and the yaw motion,front wheel steering,and considering the load transfer,and the four-wheel spin speeds can be measured by the equipped sensors.In the paper,the subscripts of

m (V +ur)=∑F yi

(i =fl,fr,rl,rr)

I zz r =[[∑i=l f ,lr

F xi ,∑i=r f ,rr F xi ,∑i=l f ,r f F yi ,∑i=lr,rr

F yi ]

]×[?B f 2,B r

,a,?b]Τ

r =ψ,

(10)

where m =m s +2(m uf +m ur );Ψis the yaw angle.Consider

a x =u ?r V ,a y =V +ru,

(11)

where a x is the longitudinal acceleration;a y is the lateral acceleration.

The forces F xi and F yi can be computed by the longitu-dinal tire force F wxi,lateral force F wyi,and the steer angleδ, from the following equations:

F xi=F xwi cosδi?F ywi sinδi,

F yi=F xwi sinδi+F ywi cosδi,

(i=fl,fr,rl,rr),

(12)

where front steering inputδf=δ;δr=0.

Based on(1)–(12),all presented submodels are merged into one model for the estimator design,

x(t)=f(x(t),u(t),w(t)),

y(t)=?(x(t),u(t),V(t)),

(13)

where state vector x=[V x,V y,r] ;the control input and mea-surement input u=[δ,w fl,w fr,w rl,w rr] ;the observations y=[a x,a y] ;w(t)is the process noise which is assumed to be drawn from a zero mean with covariance Q;and V(t)is the observation noise which is assumed to be zero mean Gaussian white noise with covariance of R.

Assuming the zero holds for the system input u(t)during the sampling timeΔT,the discrete form of the system could be obtained as follows:

x(k+1)=x(k)+ΔT?f(x(k),u(k),w(k)),

y(k)=?(x(k),u(k),V(k)),

(14) where k is the discrete time step.

4.Unscented Kalman Filter

Extend Kalman filter(EKF)is an optimal estimation method for the nonlinear systems.The nonlinear system is approxi-mated to a linear system model around the last state estimate. It has been widely used in a number of nonlinear estimation and machine learning applications.These include estimating the state of a nonlinear dynamic system,parameters identifi-cation for nonlinear system,and dual estimation[15]where both states and parameters are estimated simultaneously. The EKF is based on the Taylor Expansion theory and first order approximation to linearize the nonlinear system.This [16]can introduce large errors in the true posterior mean and covariance,which may lead to suboptimal performance and sometimes divergence of the filter.And it is difficult to calculate the Jacobi matrix especially for the strong nonlinear system.For overcoming this defect,the unscented Kalman filter(UKF)[17]is proposed.

Unscented Kalman filter(UKF)[18]is based on the unscented transform(UT)theory and statistical linearization technique.This technique is used to linearize a nonlinear function of a random variable through a linear regression between n points drawn from the prior distribution of the random variable[19].And the UKF uses the sigma points and could consider the high order terms during the state prediction and correction stage[20].The UKF is better than the EKF in terms of robustness and speed of convergence[21]; the computational load in applying the UKF is comparable to the EKF.And the absence of the linearization error further allows us to execute the filter with lager sampling times.The UKF is as in Figure6and achieved by the following five steps.

(a)Sigma Points Calculation.In the paper,we use the symmetrical sampling method to pick up the sigma points. For the random variable state vector X,the mean is x and the covariance of X is denoted by P x.The sigma points are chosen so that their mean and covariance are exactly x and P x.Let X k be a set of2n+1sigma points(where n is the dimension of the state vector x),

X k=

{{{

{

x,k=0,

x+(√(n+λ)P x)

,k=1,2,...,n,

x?(√(n+λ)P x)

k?n

,k=n+1,...,2n.

(15)

Here,λ=α2(n+κ)?n is a scaling parameter.The constant αdetermines the spread of the sigma around the x and is usually set to a small value(1≤α≤10?4);κ≥0,which makes sure that the covariance matrix is positive definite. (b)These Sigma Vectors Are Propagated through the Nonlinear System Function.Consider

X k+1=X k+1+f(X k+1,u k,w k)?ΔT.(16) (c)Measurement Update.One has

Y k+1=?(x k+1,u k+1,V k+1).(17)

The mean of the state vector x k+1and the measurement value can be approximated by using a weighted sample of the sigma vectors as follows:

x k+1|k≈

∑

i=0

(W m i(X k+1)i),

y k+1≈

∑

i=0

(W m i(Y k+1)i).

(18)

(d)Covariance Update.The covariance of the state vector and the measurement value can be calculated by

P x k+1|k=

∑

i=0

[W c i((X k+1)i?x k+1)((X k+1)i?x k+1) ]+Q, P y k+1=

∑

i=0

[W c i((Y k+1)i?y k+1)((Y k+1)i?y k+1) ]+R, P xy k+1=

∑

i=0

[W c i((X k+1)i?x k+1)((Y k+1)i?y k+1) ].

(19)

Figure6:The block diagram of UKF. The weight can be acquired by

W m i=[[

[

(n+λ)

2/(n+λ)

,...,

2/(n+λ)

???????????????????????????????????????????????

]]

]

W c i=[[

[

(n+λ)

+1?α2+β,

2/(n+λ)

,...,

2/(n+λ)

???????????????????????????????????????????????

]]

]

(20)

Here,assume that the system noise and the measurement noise are white Gaussian noise and covariance is Q and R respectively.βconsider the high order moment of the prior distribution,for Gaussian distribution,β=2is optimal. (e)Correction.Consider

K k+1=P xy k+1(P y k+1)?1,

x k+1|k+1=x k+1|k+K k+1(z k+1?y k+1),

P x k+1|k+1=P x k+1|k?K k+1P y k+1(K k+1) .

(21)

Here,z k+1is the measurement value from the sensors,z= [a x,a y].

Table3:Vehicle system parameters.

Sprung mass m s1111kg Front unsprung mass m uf60kg Rear unsprung mass m ur60kg Yaw moment of inertia I z2031.41kgm2 c.g.distance to front wheels a 1.04m c.g.distance to rear wheels b 1.56m Sprung c.g.height?cg0.542m Front track width B f1481m Rear track width B r 1.486m The wheel radius R w0.28m 5.Simulation and Analysis

To validate the performance of the UKF estimator,the algorithm is implemented through cosimulation between Carsim software and MATLAB/Simulink software.Carsim is vehicle dynamics software widely used in the automotive industry.During the simulation,Carsim model is assumed to be a real vehicle due to its high accuracy;it provides the control inputs,measurement outputs,and the estimate state information as comparison values.

The simulation scenario includes two situations,the high road adhesion with high speed and the low road adhesion with low speed.The vehicle parameters are list in Table3;

12345Time (s)

S t e e r i n g i n p u t (d e g )

?2?3?4Figure 7:Steering angle input.

456

0.20.40.6Time (s)

Y a w r a t e (r a d /s )

UKF with linear tire UKF with piecewise linear tire Carsim

?0.2

?0.4?0.6?0.8

Figure 8:Yaw rate comparison.

and the ISO-3888/1double line change (DLC)test is used in the simulation for validation of the proposed estimation algorithm.In the simulation,the estimation results of the Kalman filter with the linear tire model and the UKF with linear tire model are also presented as the comparison of the different estimation methods.

5.1.High Road Adhesion Estimation.In the simulation,the initial velocity is 100km/h,open loop throttle;and the tire-road friction coefficient is constant,1.The front wheel steering input is plotted in Figure 7.The simulation results are presented in Figures 8–10.

Figures 8–10show the yaw rate,lateral and longitu-dinal velocity estimation results using different estimation methods.It can be concluded that compared with the linear Kalman filter algorithm,the results of UKF estimator with the piece-wise linear tire model are more close to the real values.

456

0.511.52Time (s)

L a t e r a l v e l o c i t y (m /s )

UKF with linear tire UKF with piecewise linear tire Carsim

?0.5

?1?1.5

Figure 9:Lateral velocity comparison.

0123

45670

51015202530Time (s)

L o n g i t u d i n a l v e l o c i t y (m /s )

UKF with linear tire UKF with piecewise linear tire Carsim

Figure 10:Longitudinal velocity comparison.

And during the estimation of the vehicle body lateral velocity,the UKF also has a big error when the lateral acceleration is bigger,as in Figure 9,because the 3DoF vehicle model does not consider the load transfer caused by the suspension motion.

Figure 11shows the estimation results of tire slip angles (see Figure 11(a))and the tire forces (see Figure 11(b))of four individual wheels.It can be found that the UKF estimator can estimate the tire force well.We can also find that the estimation errors will increase when the lateral slip angle is large;these errors are partially due to the deficiency of the tire model.And performance of the tire model can be further improved by using the tire test data.From the above simulation results,it clearly indicates the effectiveness of

02468

0510

Time (s)

S l i p a n g l e (d e g )

?10

468

510Time (s)S l i p a n g l e (d e g )

?10

468

510Time (s)

S l i p a n g l e (d e g )

?10

468

510Time (s)

S l i p a n g l e (d e g )

?10

(a)

20004000LF ?2000

?4000

468

Time (s)

02000

4000?2000

?4000

468

Time (s)

02000

4000?2000

?40000

Time (s)

02000

4000?2000

?4000

468

Time (s)

F y (N )

(b)

Figure 11:Slip angles estimation and lateral tire forces results of four wheels (the dotted line is UKF estimation results;continuous line is Carsim results).

the proposed UKF algorithm when vehicle is driving on high friction road.

5.2.Low Road Adhesion Estimation.The DLC test was then carried out on the low friction road;the initial velocity is 40km/h,open loop throttle;and the tire-road friction coefficient is 0.2.The front wheel steering input is as in Figure 12;and the simulation results are as in Figures 13,14,15,and 1

6.The simulation results indicate that the UKF with the piece-wise linear tire can also have a good performance for the low road adhesion.It can be found that the UKF could estimate the vehicle state very well,even though the tire-road friction coefficient is very low.Figure 16illustrates the estimated tire forces (see Figure 16(a))and the tire kinemics

2460

152025

Time (s)

S t e e r i n g i n p u t (d e g )

?4?6

Figure 12:Steering angle input.

152025

Time (s)

00.050.10.150.2Y a w r a t e (r a d /s )

?0.05?0.1?0.15

UKF with linear tire UKF with piecewise linear tire Carsim

Figure 13:Yaw rate comparison.

(see Figure 16(b))in comparison with the Carsim results.It can be concluded that the UKF could estimate the tire state information effectively under the low friction condition.From the above estimation results,it is obvious that per-formance of the UKF with the linear tire model is nearly the same as the Kalman filter based on the linear bicycle model,but the estimate performances are not good;the estimator errors are large,and these errors are mainly caused by the saturation tire forces;it means that during the linear region,the estimator’s performance is good,but the estimation errors are increasing obviously during the nonlinear region.

6.Conclusion

(1)The paper is based on the 3DoF vehicle model to estimate the vehicle state information which is

00.050.10.150.2L a t e r a l v e l o c i t y (m /s )

?0.05?0.1

?0.15

?0.2

152025

Time (s)

UKF with linear tire UKF with piecewise linear tire Carsim

Figure 14:Lateral velocity comparison.

24681012

L o n g i t u d i n a l v e l o c i t y (m /s )

0510

1520253035

Time (s)

UKF with linear tire UKF with piecewise linear tire Carsim

Figure 15:Longitudinal velocity comparison.

necessary for the design of active safety controller,such as longitudinal velocity,lateral velocity,yaw rate and tire state information.

(2)Considering the computation efficiency,the piece-wise linear tire model is proposed,and compared with the Magic Formula tire model,the simplified tire model is accurate.(3)According to the unscented transfer theory,com-bined with the Kalman filter,the UKF is presented to estimate the vehicle state information in real time.(4)Compared with the reference data from the Carsim,the estimation results of the vehicle state information are precise.

010

203040

0.51Time (s)S l i p a n g l e (d e g )

?0.5

010

203040

0.51Time (s)S l i p a n g l e (d e g )

?0.5

010203040

0.51Time (s)

S l i p a n g l e (d e g )

?0.5

010203040

0.51Time (s)

S l i p a n g l e (d e g )

?0.5

(a)

500

2030

Time (s)?500LR

500

1000010

203040

Time (s)?1000

?500

20040000

203040

Time (s)

?400

?200RR

20040000

203040

Time (s)

?400

?200F y (N )

F y (N )

(b)

Figure 16:Slip angles estimation and lateral tire forces results of four wheels (the dotted line is UKF estimation results;continuous line is Carsim results).

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments

The authors acknowledge that this paper has been supported by China Scholarship Council(CSC).The authors would also like to express their sincere thanks to Professor Taehyun Shim from University of Michigan and Professor Murat Barut from Nigde University,for the guidance and help during the research.

References

[1]J.Yi,L.Alvarez,and R.Horowitz,“Adaptive emergency braking

control with underestimation of friction coefficient,”IEEE Transactions on Control Systems Technology,vol.10,no.3,pp.

381–392,2002.

[2]S.Zhu,X.Gao,and Z.Yu,“Vehicle sideslip angle estimation

under extreme driving condition,”Journal of Tongji University, vol.37,no.8,pp.1070–1074,2009.

[3]N.Ding,G.Yu,and W.Wang,“Estimation of brake pressure and

tyre-road friction during ABS activation,”International Journal of Vehicle Design,vol.58,no.1,pp.33–45,2012.

[4]B.-C.Chen and F.-C.Hsieh,“Sideslip angle estimation using

extended Kalman filter,”Vehicle System Dynamics,vol.46,no.

1,pp.353–364,2008(Chinese).

[5]H.Zhao,H.Chen,H.Su,H.Ding,and H.Guo,“Estimation

for tyre force based on sliding mode observer,”in Proceedings of the27th Chinese Control Conference(CCC’08),pp.691–695, Kunming,China,July2008.

[6]H.Tims,Vehicle active suspension system sensor reduction[Ph.D.

thesis],University of Texas,Austin,Tex,USA,2005.

[7]T.A.Wenzel,K.J.Burnham,M.V.Blundell,and R.A.Williams,

“Dual extended Kalman filter for vehicle state and parameter estimation,”Vehicle System Dynamics,vol.44,no.2,pp.153–171, 2006.

[8]C.Zong,D.Hu,and H.Zheng,“Dual extended Kalman filter

for combined estimation of vehicle state and road friction,”

Chinese Journal of Mechanical Engineering,vol.26,no.2,pp.

313–324,2013.

[9]R.Guntur and S.Sankar,“Friction circle concept for dugoff’s

tyre friction model,”International Journal of Vehicle Design,vol.

1,no.4,pp.373–377,1980.

[10]H.B.Pacejka and I.J.M.Besselink,“Magic formula tire model

with transient properties,”Vehicle System Dynamics,vol.27, supplement1,pp.234–249,1997.

[11]P.Tsiotras,E.Velenis,and M.Sorine,“A LuGre tire friction

model with exact aggregate dynamics,”Vehicle System Dynam-ics,vol.42,no.3,pp.195–210,2004.

[12]K.Guo and L.Ren,“A unified semi-empirical tire model with

higher accuracy and less parameters,”SAE Technical Paper Series,paper1999-01-0785,1999.

[13]H.B.Pacejka,Tyre and Vehicle Dynamics,Butterworth Heine-

mann,Oxford,UK,2002.

[14]J.Yi,J.Li,J.Lu,and Z.Liu,“On the stability and agility

of aggressive vehicle maneuvers:a pendulum-turn maneuver example,”IEEE Transactions on Control Systems Technology,vol.

20,no.3,pp.663–676,2012.[15]T.A.Wenzel,K.J.Burnham,M.V.Blundell,and R.A.Williams,

“Dual extended Kalman filter for vehicle state and parameter estimation,”Vehicle System Dynamics,vol.44,no.2,pp.153–171, 2006.

[16]E.A.Wan and R.Van der Merwe,“The unscented Kalman filter

for nonlinear estimation,”in Proceedings of the Adaptive Systems for Signal Processing,Communications,and Control Symposium (AS-SPCC’00),pp.153–158,Alberta,Canada,October2000.

[17]R.Van Der Merwe, E. A.Wan,and S.Julier,“Sigma-

point Kalman filters for nonlinear estimation and sensor-fusion:applications to integrated navigation,”in Proceedings of the AIAA Guidance,Navigation,and Control Conference and Exhibit(AIAA’04),pp.1735–1764,Providence,RI,USA, Aughust2004.

[18]E.Wan and R.van der Merwe,“Kalman filtering and neural

networks,”in The Unscented Kalman Filter,S.Haykin,Ed., chapter7,John Wiley and Sons,New Y ork,NY,USA,2001. [19]L.Meng,L.Li,and S.M.Veres,“Comparison of linear and

nonlinear aerodynamic parameter estimation approaches for an unmanned aerial vehicle using unscented Kalman filter,”

Journal of Beijing Institute of Technology,vol.20,no.3,pp.339–344,2011.

[20]R.Kandepu,B.Foss,and L.Imsland,“Applying the unscented

Kalman filter for nonlinear state estimation,”Journal of Process Control,vol.18,no.7-8,pp.753–768,2008.

[21]S.Antonov,A.Fehn,and A.Kugi,“Unscented Kalman filter for

vehicle state estimation,”Vehicle System Dynamics,vol.49,no.

9,pp.1497–1520,2011.

VLSI Design Hindawi Publishing Corporation https://www.360docs.net/doc/b511710222.html,

Volume 2014

International Journal of

R otating

Machinery

Hindawi Publishing Corporation https://www.360docs.net/doc/b511710222.html, Volume 2014

Hindawi Publishing Corporation Journal of Engineering Hindawi Publishing Corporation https://www.360docs.net/doc/b511710222.html, Volume 2014Shock and Vibration Mechanical Engineering Advances in Civil Engineering

Advances in Acoustics and Vibration Advances in Electrical and Computer Engineering

Journal of Distributed S ensor Networks The Scientific World Journal Hindawi Publishing Corporation Volume 2014Sensors Modelling &

Simulation

in Engineering

https://www.360docs.net/doc/b511710222.html, A ctive and Passive Electronic Components Advances in OptoElectronics

Hindawi Publishing Corporation https://www.360docs.net/doc/b511710222.html,

Volume 2014Robotics Journal of Volume 2014Chemical Engineering International Journal of Hindawi Publishing Corporation Volume 2014Control Science and Engineering Journal of

Hindawi Publishing Corporation https://www.360docs.net/doc/b511710222.html, Volume 2014 ntennas and Propagation Hindawi Publishing Corporation https://www.360docs.net/doc/b511710222.html, Volume 2014Navigation and O bservation International Journal of