Similarity Analysis of Disturbance Observer and

Active Disturbance Rejection Control for Typical

Motor-driven System
Yutang Wang1,2 , Dapeng Tian∗1 ,Ming Dai1 , Honghai Shen1 and Ping Jia1
1
Key Laboratory of Airborne Optical Imaging and Measurement, Changchun Institute of Optics,
Fine Mechanics and Physics Chinese Academy of Sciences, Changchun, Jilin China
2
University of Chinese Academy of Sciences, Beijing China
∗
Corresponding author. Email: [email protected]

Abstract—High-precision motion control has attracted a lot has extensive practical engineering applications due to the
of attentions in the control fields of mechatronic system. Nev- simplicity of its design principle and efficiency [14], [15].
ertheless, the dynamic performance of a motor-driven system is The robustness and stability of a disturbance observer and a
limited due to system uncertainties and complex disturbances.
These challenges deteriorate the accuracy of the motion control reaction torque observer-based robust motion control systems
system. In recent years, Disturbance Observer (DOB) and Active is analyzed in [16].
Disturbance Rejection Control (ADRC) have earned widespread On the other hand, ADRC is a typical nonlinear-oriented
acceptance in the field of disturbance rejection control.This method which was proposed by J. Han on the idea of “active
paper presents a similarity analysis of DOB and ADRC for a disturbance rejection” [17]. The core idea is using a nonlinear
typical motor-driven system in frequency domain, similarities and
differences between them are analyzed. To confirm the analysis, estimator called Extended State Observer (ESO) to estimate
simulations is conducted. The analysis result could propose new and compensate all kinds of disturbance actively. It owns the
design ideas based on DOB and ADRC with better adaptablility advantage of “small error, big gain; big error, small gain”.
for kinds of complex systems. ADRC has widely applications. The stability of nonlinear
ADRC in frequency domain is analyzed in [18], root locus
I. I NTRODUCTION
analysis, describing function method and extend circle criteri-
High-precision motion control has attracted considerable on are applied. The stability, dynamic stiffness, and tracking
attention, especially in the fields of industrial automation, performance are analyzed for a second-order plant in [19]. The
robotics, and other mechatronic Systems [1]- [3]. However, analysis results show that the nonlinearity parameter plays a
system nonlinearity, model uncertainties, unknown dynamics crucial role in the system performance. The nonlinear ADRC
and external disturbances are significant challenges in practical has higher control efficiency than the linear ADRC but reduces
motor-driven control systems. These severely deteriorate the the stability margin of the system.
system control accuracy and dynamic performance. This is Both DOB and ADRC are very commonly used methods
an obstacle in the achievement of higher accuracy in motion for motor control system. The core principle of DOB and
control. ADRC resembles as the disturbance estimator and compen-
Various methods have been proposed to counter the above sator. In [20], DOB and ADRC are compared for estimating
challenges such as the disturbance observer (DOB) [4], [5], the uncertainties. A conclusion that ADRC provides a base
sliding-mode control (SMC) [6], [7], active disturbance rejec- for generalizing DOB for nonlinear system with mixed uncer-
tion control(ADRC) [8], [9], adaptive robust control (ARC) tainties is deduced. Nevertheless, the essence relationship of
[10], [11], repetitive control [12]. Because most uncertainties ADRC and DOB is still not clear. DOB and ESO have been
and disturbances are hard to measure, the disturbance esti- compared in a more general sense in [21]. However there is
mation via known information has become a significant and no similarity analysis on the equivalent gain of ADRC and
challenging problem. DOB. In this paper, the similarities and differences analysis
The DOB, which was proposed by Ohinishi, is an extremely of them are presented to evaluate their disturbance rejection
effective method compared to the other motion control meth- effects on a typical motor-driven system. The comparison
ods mentioned above. It estimates the equivalent disturbance performance of DOB and ADRC are analyzed in frequency
and then utilizes it as an offset input signal to compensate for domain. The simulation results further support the analysis.
the disturbance [13]. The closed-loop robust control method Such an analysis provides design reference for subsequent
based on DOB has a strong inhibitory effect on external redesign on disturbance rejection method.
disturbances and parameter variations in a small area and can This paper is organized as follows. we shall firstly introduce
obviously improve the control effect. The DOB has become the design principle of DOB and ADRC briefly in Section I
a basic control method for high precision servo control and and Section II , respectively. The similarity analysis of ADRC

l-))) 
 ƴ ƴ
ƴ
ƴ ƴ ƴ ƴ ƴ

Fig. 1. Equivalence disturbance

'LVWXUEDQFH2EVHUYHU

and DOB in frequency domain is proposed in Section III. The

Fig. 2. Block diagram of disturbance observer (DOB)
simulation is conducted in Section V. Finally, this paper is
concluded in Section VI.

II. D ESIGN P RINCIPLES OF THE D ISTURBANCE O BSERVER A typical low-pass filter of Q(s) is given by (6):
A. Equivalence of Model Uncertainty and External Distur-
bance 
M
N !(τ s)k
The error between the actual model and the nominal model (N − k)!k!
k=0
is named as the model uncertainty ΔM (s). The summation QN M (s) = , M = 0, 1, ..., N − 1, (6)
of the model uncertainty ΔM (s) and external disturbance (τ s + 1)N
Dext can be transferred into the equivalence disturbance D where N is the denominator order, M is the molecular order,
as shown in Fig. 1. Gp and Gn are the practical and nominal and τ is the filter time constant.
models, respectively. Ur is the expected control input. The
III. D ESIGN P RINCIPLES OF THE ACTIVE D ISTURBANCE
transformation is shown as follows.
R EJECTION C ONTROL
Fig. 1 can be described as
The structure of ADRC with a second order ESO is shown
Y (s) = Gn (s)(1 + ΔM (s))(Ur + Dext ) in Fig. 3. ESO plays a role as an active observation of the
= Gn (s)(Ur + Dext + ΔM (s)Ur + ΔM (s)Dext ). output information and its differential, it provides real-time
(1) estimation and compensation of equivalence disturbance.
Assuming that
D(s) = Dext + ΔM (s)(Ur + Dext ), (2)
ƴ ƴ

the system output Y (s) is given by

Y (s) = Gn (s)(Ur (s) + D(s)), (3)
6HFRQGRUGHU(62
where Ur (s) represents the expected control input.
A typical motor control system is considered here for the
sake of simplicity and clarity. The nominal model of its $'5&
velocity loop can be expressed as (4).
1 Fig. 3. Block diagram of simplified ADRC
Gn (s) = (4)
Jn s + Bn
where Jn is the equivalence of the nominal inertia and Bn is The control plant Eq.(4) is re-written into state space as
the damping coefficient. If the mechanical structure is designed ⎧
⎨ ẋ = 1 (D + u)
and finely machined, the bearing friction of a direct-drive 1
Jn (7)
motor is sufficiently small to be ignored. In this paper, Bn ⎩
y = x1 .
is set to zero.
Supposing that x2 = J1n D is related to equivalence distur-
B. Disturbance Observer
bance of system and the derivation of x2 is h(t), then the first-
The structure of the DOB is represented in Fig.2. It is order system Eq.(7) is re-written into a second-order system
composed of a low-pass filter Q(s) and the inverse of the as Eq.(8). x2 is called extended state.
nominal model G−1
n (s). D̂ is the estimated disturbance of D ⎧
and U is the compensated input in the Laplace domain given ⎪ 1
⎨ ẋ1 = x2 + Jn u
⎪
by (5).
ẋ2 = h(t) (8)
⎪
⎪
⎩
U (s) = Ur (s) − D̂(s) (5) y = x1


 Based on the output y and input u, Eq.(9) is constructed. z 
is estimator of state variable x in Eq.(8), e is the state error of 4 π/2
B1 = f al(e, α, δ))sin(ωt)dωt. (15)
the two systems. β01 and β02 are adjustable parameters tuned π 0
according to practical experience.
For the specific design in Eq.(10), the describing function
⎧ N (A) can be calculated as Eq.(16), where τ1 = arcsin(δ/A)
⎪ e = z1 − x1
⎪
⎨ is the demarcation point of piecewise function f al(e, α, δ) and
1
ż1 = z2 + u − β01 e (9) 0 < τ1 < π/2.
⎪
⎪ J
⎩ n
ż2 = −β02 f al(e, α, δ)  τ1
4
f al(e, α, δ) is a non-smooth function. A specific piecewise N (A) = Asin(ωt)sin(ωt)/δ 1−α dωt
Aπ 0
function is as Eq.10. An extreme case is α = 1 and δ = 0,  π/2
4
then f al(e, α, δ) turns to be completely linear. + Aα sinα (ωt)sin(ωt)dωt
 Aπ τ1
 τ1 (16)
|e|α sgn(e), |e| > δ 4
f al(e, α, δ) = (10) = 1−α sin2 (ωt)dωt
e/δ 1−α , |e|  δ πδ 0
 π/2
4
Then the control variable can be regarded as Eq.(11), + sin1+α (ωt)dωt
Jn z2 (t) is the equivalence disturbance. πA1−α τ1
Note that N (A) depends on the input amplitude only.
u(t) = ur (t) − Jn z2 (t) (11) According to the state equation Eq.(8), the corresponding
block diagram of ADRC in the forms of transfer function and
IV. A NALYSIS ON S IMILARITY
describing function is given as Fig.4.
In this section, the similarity analysis between ADRC and
DOB is investigated. Section IV-A describes the equivalence
transformation of ADCR in frequency domain. Section IV-B ƴ ƴ

analyzes the similarity and compares the frequency character-

istics of ADRC and DOB.

A. Equivalence Analysis ƴ

To compare the parts of disturbance rejection in ADRC and ƴ

DOB, the two disturbance estimators should be changed into

a similar structure. But the nonlinear function f al(e, α, δ) can ƴ
ƴ

not be expressed in frequency domain directly. $'5&(62

Describing function analysis [22] is an useful linearization
method to analyze nonlinear control system in frequency
Fig. 4. Block diagram of the transformed ADRC
domain. It is a kind of approximation and promotion applied
to nonlinear system research, so that classic frequency domain
methods in linear control theory can be extended into nonlinear Via equivalence transformation, the ESO of ADRC is able
control area. to be transformed into Fig.5 in a simplified form.
When the input is a sinusoid function as Asin(ωt), The
describing function of the nonlinear function f al(e, α, δ)) can
be described as N (A). ƴ ƴ

B1 + jA1
N (A) ≈ (12) ƴ
A
Where A1 and B1 are the first order harmonic components of
the output. ƴ

 2π
1
A1 = f al(e, α, δ))cos(ωt)dωt
π 0 (13)
$'5&(62

 2π
1 Fig. 5. Equivalence structure of ADRC
B1 = f al(e, α, δ))sin(ωt)dωt (14)
π 0

Because that f al(e, α, δ) is an odd function, A1 is 0 and Where


 TABLE I
F ILTER PARAMETERS COMPARISON OF DOB AND ADRC

1 1
Filters s2 +a1 s+a2 s3 +a1 s2 +a2 s+a3
[a1 , a2 ] [a1 , a2 , a3 ]
2 2 3
DOB filter [2gdis , gdis ] [3gdis , 3gdis , gdis ]
ADRC fliter [β01 , β02 N (A)] [β01 , β02 , β03 N (A)]
ADRC fliter
2 2 3
after parameter tuning [2gdis , gdis N (A)] [3gdis , 3gdis , gdis N (A)]

⎧ Specific numeral comparisons on the frequency characteris-

⎪ H(s) = P (s) + M
⎪
⎪
tics of DOB and ADRC filters are analyzed. Fig.6 shows the
⎨ P (s) = F (s)/s comparison of equivalence ADRC filter with various δ. Noted
(17)
⎪
⎪ F (s) = β02 Jn N (A)
that the bode plots coincides with others when δ is small.
⎪
⎩ Since that δ corresponds the sensor noise, which is usually far
M = Jn β01 .
less than 1, δ is ignored in the remaining frequency analysis
Considering that the structure of ADRC in Fig.5 is similar for simplicity,.
to the structure of DOB in Fig.2, an equivalence filter of
Bode Diagram
ADRC can be deduced in the frequency domain. Compared 0

with Fig.2, the ADRC equivalent low-pass filter QADRC (s) −10

is able to be described as −20

Magnitude (dB)
−30

Gn (s) −40

QADRC (s) = P (s) . (17) −50

1 + Gn (s)H(s) −60
−70
B. Frequency Characteristics Analysis
−80

When Gn (s) = 1/Jn s, the corresponding ADRC low-pass

0
δ=0
filter is −45 δ=0.01
δ=0.1
Phase (deg)

β02 N (A) δ=0.5

QADRC2 (s) = . (18) −90
δ=1
s2 + β01 s + β02 N (A) δ=2
−135

QADRC2 is a second-order filter for a first-order control plant.

−180
According to Eq.(6), a second-order low-pass fiter of DOB 10
0
10
1
10
2 3
10
4
10

is described as Eq.(19). Frequency (rad/s)

1 Fig. 6. Comparison of ADRC filters Q2ADRC with various δ

Q20DOB (s) =
τ s2
+ 2τ s + 1
2 (19)
gdis
= 2 2 In Fig.7, QADRC2 (s) and Q20DOB (s), QADRC3 (s) and
s + 2gdis s + gdis
Q30DOB (s), Q10DOB are compared in frequency domain.
where gdis = 1/τ . When δ = 0, A = 1, the bode plots of QADRC2 (s) and
For a second-order motor-driven system as Gn (s) = Jn1s2 , Q20DOB (s), QADRC3 (s) and Q30DOB (s) almost overlap,
the equivalence filter of ADRC in frequency domain can be respectively. It means that the filters of ADRC and DOC with
derived as Eq.(20). same order are equivalent to some extent.
The bode plot in Fig.7 as Line Q10DOB or Line Q20DOB
β03 N (A) represents that the DOB filter owns a constant bandwidth
QADRC3 (s) = . (20) in frequency domain. However, since the describing function
s3 + β01 s2+ β02 s + β03 N (A)
changes with the input amplitude A, a ADRC compensator
The third-order low-pass filter of DOB is described as with constant parameters β01 , β02 and α is a crowd of plots
3
gdis
Q30DOB (s) = s3 +3gdis s2 +3g 2 3 . changed with various A in bode diagram. A corresponds
dis s+gdis
According to the bandwidth-parameterization based con- to the value of error e in practical control system. Fig.8
troller tuning method in [23], the parameters of ADRC filter shows the frequency characteristics comparison of ADRC with
can be tuned. The comparison on filter parameters of DOB various A. It means that the frequency characteristics of the
and ADRC is shown in Table I. It shows that QADRC2 (s) compensator changed with control error e in Eq.(9).
and Q20DOB (s), QADRC3 (s) and Q30DOB (s) own similar As it is known, in low-frequency stage, the closer that
structure, respectively. Without considering the nonlinear part control margin is near to 0 dB, the better performance that sys-
of ADRC, the ADRC filter is equivalent to a Qn0 filter of tem rejects equivalence disturbance. While in high-frequency
DOB, where n denotes the filter order. stage, system is more sensitive to measurement noise when


 Bode Diagram TABLE II
0
E XPERIMENTAL PARAMETERS
−20

Parameter Value Description

Magnitude (dB)

−40

−60
Jn 0.004347826 [k] Nominal mass
Jp 0.005 [k] Real mass
−80
gdis 100 [rad/s] Filter cutoff frequency of DOB
−100 Kcp 1.74 [kg/s2 ] Proportional gain
−120 Kcd 0.174 [kg/s] Derivative gain
0
β01 200 Parameter of ADRC
−45 β02 10000 Parameter of ADRC
Q10DOB
−90 α 0.5 Parameter of ADRC
Phase (deg)

Q20DOB
−135
Q30DOB
−180
QADRC2 δ=0,A=1
−225 Q3ADRC δ=0,A=1 V. S IMULATION
−270
10
0
10
1
10
2 3
10
4
10 A. Simulation Set-up
Frequency (rad/s)
Simulations are implemented to compare the disturbance
Fig. 7. Comparison of equivalence ADRC filter and DOB filters in frequency evaluation performance of ADRC and DOB filters. The Sim-
domain ulation parameters are shown in Table II.
A position feedback controller C(s) as Eq.(21) is applied
in control system.

C(s) = Kcp + Kcd s. (21)

B. Simulation Results
Disturbance rejection ability and dynamic performance were
performed under DOB and ADRC.
The expected position command rin is shown as Eq.(22).
⎧
⎨ 1.5sin(2π(t − 4)),
⎪ 1.5s ≤ t < 4s
rin (t) = − 2, 4s ≤ t < 6.5s (22)
⎪
⎩
0, else
The additional disturbance dadd is as dadd (t) = 0.1ẋ1 when
|ẋ1 | ≥ 0.2 and as Eq.(23) when |ẋ1 | < 0.2, where Ft (t) =
Jn ẍ1 denotes the driving force.
⎧
⎨ 0.5,
⎪ Ft (t) > 0.5
Fig. 8. Comparison of equivalence ADRC filter and DOB filters in frequency
dadd (t) = Ft (t), −0.5 ≤ Ft (t) ≤ 0.5 (23)
domain ⎪
⎩
− 0.5, Ft (t) < −0.5
Fig.9 demonstrates the dynamic responses when the motor-
driven system with a QDOB10 ,QDOB20 or QADRC compen-
control margin is larger. Although a higher order filter may sator experiences state switch and additional disturbance.
offer improved disturbance restraint ability, phase lag of it Comparisons on the ability of disturbance rejection of DOB
is larger than a lower-order one and the calculation of the and ADRC are shown in Fig.10.
controller increases. For the typical motor-driven system as Simulation results shows the similarity on disturbance re-
Eq.(4), its practical low-pass filter of DOB is usually chosen jection of ADRC and DOB. QADRC and QDOB20 are second-
as Q10DOB (s) = s+g gdis
, while the corresponding filter of order compensators, they owns better disturbance restraint
dis
ADRC is a second-order filter. ability. However, the calculation time of them increases com-
pared with QDOB10 . Meanwhile, the nonlinearity of QADRC
However, ADRC filter possesses the nonlinear merit that increased the rate of system convergence.
its gain is big for small error so that control error converges
fast, while its gain is small for big error so that the DA VI. C ONCLUSION
converter will not be saturated for practical applications. For This paper has considered the similarity analysis of DOB
this reason, a new disturbance compensator can be designed and ADRC for motor-driven systems. A detailed comparison
under an unified framework of ADRC and DOB, combining and similarity analysis in frequency domain were provided. It
the similarities and relative advantages of both sides. is shown that ADRC is a special form of DOB with nonlinear


 1.5
[2] D. Lei, T. Wang, D.Cao, and J. Fei, “Adaptive Dynamic Surface Control
1 of MEMS Gyroscope Sensor Using Fuzzy Compensator,” IEEE Access,
vol. 4, pp.4148-4154, 2016.
Expected output and real control outputs

0.5 [3] Z.Q. Gao. “On the Centrality of Disturbance Rejection in Automatic
Control,” ISA Transactions, vol 53, pp. 850-857, 2014.
0 [4] S. K. Kim, C. R. Park, T. W. Yoon, and Y. I. Lee, “Disturbance-observer-
based Model Predictive Control for Output Voltage Regulation of Three-
−0.5 phase Inverter for Uninterruptible-power-supply Applications,” European
Journal of Control, vol. 23, pp.71-83, 2015.
−1 [5] M. Chen, S. S. Ge, “Direct Adaptive Neural Control for a Class of
Uncertain Nonaffine Nonlinear Systems Based on Disturbance Observer,”
−1.5 IEEE Transactions on Cybernetics, vol. 43, no. 4, pp.1213-1225, 2013.
Expected output [6] F. Li, L. Wu, P. Shi, and C.C. Lim “State Estimation and Sliding
−2 Real control outputs QDOB10 Mode Control for Semi-Markovian Jump Systems with Mismatched
Real control outputs QADRC Uncertainties,” Automatica, vol. 51, pp.385-393, 2015.
−2.5
Real control outputs QDOB20 [7] Z. G. Sun, N. C. Cheung, S. W. Zhao, and W. C. Gan, “Adaptive
RBFNNs/integral Sliding Mode Control for Aquadrotor aircraft,” Neu-
−3
0 2 4 6 8 10 rocomputing, vol. 216, pp.126-134, 2016.
Time (s) [8] Z. L. Zhao, B. Z. Guo, “On active disturbance rejection control for
nonlinear systems using time-varying gain,” European Journal of Control,
Fig. 9. Dynamic responses of practical system with different compensators vol. 23, pp.62-70, 2015.
[9] B.Z. Guo, H.C. Zhou, “The Active Disturbance Rejection Control to Sta-
bilization for Multi-Dimensional Wave Equation With Boundary Control
Matched Disturbance,” IEEE Transactions on Automatic Control, vol.60,
no.1, pp.143-157, 2015.
[10] Z. Chen, B. Yao, Q. Wang, “μ-Synthesis-Based Adaptive Robust Control
of Linear Motor Driven Stages With High-Frequency Dynamics: A
Case Study,” IEEE/ASME Transactions on Mechatronics, vol. 20, no.3,
pp.1482-1490, 2015.
[11] R. Babaghasabha, M. A. Khosravi, H. D. Taghirad, “Adaptive robust
control of fully-constrained cable driven parallel Robots,” Mechatronics,
vol. 25, pp.27-36, 2015.
[12] G. A. Ramos, J. C. Romero, H. C. Enriquez, “Spatial Observer-
based Repetitive Controller: An Active Disturbance Rejection Approach,”
Control Engineering Practice, vol. 42, pp.1-11, 2015.
[13] K. Ohnishi, M. Shibata, T. Murakami,“Motion Control for Advanced
Mechatronics,” IEEE/ASME Transactions on Mechatronics,vol. 1, no. 1,
pp.56-67, 1996.
[14] D.P. Tian, D. Yashiro, K. Ohnishi, “Improving Transparency of Bilateral
Control System by Sliding Mode Assist Disturbance observer,” IEEE
Transactions on Electrical and Electronic Engineering, vol.8 ,no. 3, pp.
Fig. 10. Dynamic responses of practical system with different compensators 277-283, 2013.
[15] C. Liu, J. H. Wu, J. Liu, Z. H. Xiong, “High Acceleration Motion Con-
trol based on a Time-domain Identification Method and the Disturbance
Observer,” Mechatronics, vol. 24, pp.672-678, 2014.
[16] E. Sariyildiz, K. Ohnishi, “Stability and Robustness of Disturbance-
characteristics in nature. According to the similarity analysis of Observer-Based Motion Control Systems,” IEEE Transactions on Indus-
this paper, sensitivity function and complementary sensitivity trial Electronics, vol. 62, no.1, pp. 414-422, 2015.
[17] J.Q. Han,“From PID to Active Disturbance Rejection Control,” IEEE
function, which are usually considered in DOB design, could Transactions on Industrial Electronics,vol. 56, no. 3, pp. 900-906 ,2009.
be applied in the design of ADRC for kinds of complex [18] J. Li, X. Qi, Y. Xia, F. Pu, K. Chang, “Frequency Domain Stability
systems. In the meantime, DOB can be redesigned based on Analysis of Nonlinear Active Disturbance Rejection Control System,”
ISA Transactions, vol. 56, pp. 188-195, 2014.
the advantage of nonlinear anti-saturation characteristics of [19] D. Wu, K. Chen,“Frequency-Domain Analysis of Nonlinear Active
ADRC for better adaptability. Future research will focus on Disturbance Rejection Control via the Describing Function Method,”
these research directions. IEEE Transactions on Insustrial Electronics,, vol. 60(9), pp. 3096-3914,
2013.
[20] W.C. Xue, Y. Huang, “Comparison of the DOB Based Control, A Special
ACKNOWLEDGMENT Kind of PID Control and ADRC,” American Control Conference, pp.
4373-4379, San Francisco, USA, June 2011.
This work was supported in part by the National Science [21] J. Su, W.H. Chen, J. Yang, “On Relationship between Time-Domain and
Foundation of China under Grant 61673365 and 51705496, in Frequency-Domain Disturbance Observers and Its Applications,” Journal
part by CIOMP Knowledge Innovation Program under Grant of Dynamic Systems, Measurement, and Control, vol. 138, no. 9, pp.
091013, 2016.
Y4CX1SS145 and Y4CX1SS146, and in part by the Youth [22] H.K.Khalil, “Nonlinear System,” 3rd ed. Englewood Cliffs, NJ, USA:
Innovation Promotion Association of Chinese Academy of Prentice-Hall, 2002.
Sciences under Grant 2017257. [23] Z.Q.Gao, “Scaling and bandwidth-parameterization based controller
tuning,” American Control Conference, pp. 4989-4996, Denver, Colo,
USA, June 2003.
R EFERENCES
[1] H.L. Xing, J.H. Jeon, K.C. Park, I.K. Oh, “Active Disturbance Rejection
Control for Precise Position Tracking of Ionic Polymer-Metal Composite
Actuators,” IEEE/ASME Transactions on Mechatronics, vol. 18, no. 1,
pp.86-95, 2013.