Available via license: CC BY

Content may be subject to copyright.

Special Issue Article

Advances in Mechanical Engineering

2017, Vol. 9(2) 1–9

ÓThe Author(s) 2017

DOI: 10.1177/1687814016686665

journals.sagepub.com/home/ade

Single-input adaptive fuzzy sliding

mode control of the lower extremity

exoskeleton based on human–robot

interaction

Xinglai Jin

1

, Shiqiang Zhu

1

, Xiaocong Zhu

1

, Qingcheng Chen

2

and

Xuequn Zhang

1

Abstract

This article introduces a human–robot interaction controller toward the lower extremity exoskeleton whose aim is to

improve the tracking performance and drive the exoskeleton to shadow the wearer with less interaction force. To

acquire the motion intention of the wearer, two subsystems are designed: the first is to infer the wearer is in which

phase based on floor reaction force detected by a multi-sensor system installed in the sole, and the second is to infer

the motion velocity based on the multi-axis force sensor and admittance model. An improved single-input fuzzy sliding

mode controller is designed, and the adaptive switching controller is combined to promote the tracking performance

considering system uncertainties. Adaptation laws are designed based on the Lyapunov stability theorem. Therefore, the

stability of the single-input adaptive fuzzy sliding mode control can be guaranteed. Finally, the proposed methods are

applied to the lower extremity exoskeleton, especially in the swing phase. Its effectiveness is validated by comparative

experiments.

Keywords

Human–robot interaction, lower extremity exoskeleton, admittance model, floor reaction force, single-input adaptive

fuzzy sliding mode control

Date received: 16 October 2016; accepted: 2 December 2016

Academic Editor: Zheng Chen

Introduction

Lower extremity exoskeleton is a typical kind of inte-

grated human–robot system which is superior to any

autonomous robotic system in unstructured environ-

ments that demand significant adaptation.

1

It attempts

to combine the strength and endurance of modern

robotics with the intelligence and agility of a human

operator.

2,3

Obviously, it has many potential advan-

tages, such as allowing the user to carry more loads

and to traverse irregular terrain surfaces. So, the lower

extremity exoskeleton can help individuals employed in

particular recreational, occupational, and military

activities who often carry heavy loads; it can help fire

fighters and other emergency personnel to carry oxygen

tanks and other equipment; and it can assist impaired

man to walk.

4

In one word, the exoskeleton benefits

from the combination of the wearer intellect and

machine strength.

1

State Key Laboratory of Fluid Power and Mechatronic Systems, Zhejiang

University, Hangzhou, China

2

Shanghai Institute of Special Equipment Inspection and Technical

Research, Shanghai, China

Corresponding author:

Xinglai Jin, State Key Laboratory of Fluid Power and Mechatronic

Systems, Zhejiang University, Hangzhou 310027, China.

Email: 21125069@zju.edu.cn

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License

(http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without

further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/

open-access-at-sage).

To provide effective physical support according to

each wearer’s intention, it is necessary to strongly focus

on the control system. An important challenge in the

exoskeleton system is the ability to sense wearer’s

motion intention in time.

5

According to the measure-

ment of the human–robot interaction (HRI), exoskele-

ton system can be divided into cognitive human–robot

interaction (cHRI)-based system and physical human–

robot interaction (pHRI)-based system.

6

The cHRI-

based system measures the electric signals from the

central nervous system (CNS) to the musculoskeletal

system of the human and subsequently uses them as

input information for the robot controller, such as

HAL.

7

However, the pHRI-based system measures

change of force or position that result from the motions

of the human musculoskeletal system and uses them as

input information for the robot controller, such as

BLEEX and HEXAR.

8,9

The voluntary controller of

HAL is developed which uses the wearer’s bioelectrical

signals. These signals are based on the myoelectric sig-

nal detected on the skin surface of the supporting mus-

cles.

10,11

Kwon and Kim

5

proposed an upper limb

motion estimation method using surface electromyo-

graphy (SEMG) and joint angular velocities based on

artificial neural network. Although the EMG sensor is

generally used to measure muscular activity signals, it

is disadvantageous in which it must be directly attached

to the skin; it requires high sampling frequencies for

signal collection; and it is difficult to quantify the sig-

nals.

12,13

As a novel control method, sensitivity amplifi-

cation control (SAC) has been designed in BLEEX.

14

It

needs no direct measurements from the wearer or the

human–machine interface. However, the system perfor-

mance is proportional to the precision of the exoskele-

ton dynamic model which is difficult to depict.

15

Measuring the human–machine interaction force is an

intuitionistic and widely used method. Yu et al.

16

have

designed a model-free proportional–integral–derivative

(PID)-type admittance control for an upper limb exos-

keleton to infer the desired trajectory from the human–

machine interaction force. To generate the wearer’s

intension from the interaction, Lee et al.

6

have adopted

virtual impedance control to model the human–

machine interaction. Lee et al.

17

also adopted a PI con-

troller to simplify the system controller.

As far as we know, no current technologies can

directly measure and extract the human intentions.

However, we can infer human intentions from their

appearance or motion. In this article, we will design an

intention estimator based on floor reaction force (FRF)

and force sensor which detects the HRI force.

11,18

Impedance control and admittance control are widely

used in modeling the human–machine interaction rela-

tionship.

16,19

However, the former cannot guarantee

zero contact force which may cause wearer discomfort.

So, the latter is adopted to infer the wearer’s motion

intention. Sliding mode control has been used in many

kinds of electromechanical systems, such as linear

motor,

20,21

hydraulic actuator,

22,23

active suspension

systems,

24,25

and teleoperation systems.

26,27

Finally, to

design a controller which is independent of mathemati-

cal model and is easy to accomplish, a direct adaptive

fuzzy sliding mode controller based on single input is

proposed and a new nonlinear integral sliding surface is

introduced to deal with the system chattering. Because

the swing phase occupies the 40% of one cycle, effec-

tiveness of the proposed algorithms is verified by

experiments in swing phase.

A brief overview of the article is given as follows: in

section ‘‘Architecture,’’ the lower extremity exoskeleton

is introduced which is developed to enhance the wearer

strength. In section ‘‘Acquisition of human intent,’’ the

pHRI system is established which is essential to acquire

human intent. The relationship between the walking

phase and FRF is verified by experiment. In section

‘‘Human–machine interaction model,’’ the admittance

is recommended to model HRI and the parameters can

be designed based on human impedance properties. In

section ‘‘Controller design,’’ a novel direct adaptive

fuzzy sliding mode controller based on single input is

proposed. To reduce the complexity of the fuzzy rule

and decrease the input variables of the fuzzy controller,

traditional sliding mode surface s(t) and its derivative

_

s(t) are replaced by sliding mode switching surface s(t).

The adaptive regulation law is designed to enhance the

robustness of the system by the Lyapunov theory. In

section ‘‘Experiment,’’ comparative experiments are

performed to verify the effectiveness of the proposed

controller. Section ‘‘Conclusion’’ concludes this article.

Architecture

The lower extremity exoskeleton is shown in Figure 1,

and it comprised two powered anthropomorphic legs, a

pair of shoes, and a spine which is designed to carry

heavy load. The structure of the exoskeleton belongs to

pseudo-anthropomorphic architecture to ensure maxi-

mum safety and minimum collisions with the environ-

ment and operator which means the structure of

exoskeleton is designed by reference to human legs, but

does not include every degree of freedom (DOF) of

human legs. Overall, the exoskeleton has seven distinct

DOFs per leg: 3 DOFs at the hip, 1 DOF at the knee,

and 3 DOFs at the ankle. Since the human leg and exos-

keleton leg kinematics are merely similar, the wearer and

exoskeleton are rigidly connected at the feet, shanks,

and upper body. It is unnecessary to actuate all DOFs

because most of the power needed for normal walking is

in the sagittal plane according to some researches. So,

only the freedoms of hip and knee in sagittal plane are

actuated by hydraulic. The ankle joint is passive, and

the springs are placed to help the foot reset.

2Advances in Mechanical Engineering

Acquisition of human intent

An entire walking gait cycle can be divided into the

swing phase and the standing phase. Researches have

shown that the leg undergoes large motions but it needs

relatively small torques which only supports its own

weight in the swing phase; the leg executes a small

motion but supports the entire torso and payload in

the standing phase. Thus, most of the lower extremity

exoskeletons have designed hybrid control, such as

BLEEX and HAL.

28,29

We have uncoupled the acquisi-

tion system of human intent into two subsystems. The

first subsystem is designed to detect the walking phase,

and the second subsystem is designed to detect the HRI

force. Conventional researches on human transient

walking have made clear that a FRF shift in a sole to

one leg is a prior motion to a walk.

7

So, if we can sense

the FRF shift induced by the wearer’s intention, we

can infer the leg is in which phase. According to some

researches, the foot pressures on the front and rear

parts are larger than other parts. To detect the FRF in

a sole, we have designed the exoskeleton foot and its

sensor system as shown in Figure 2.

The FlexiForce A401 force-sensing resistor (FSR)

developed by Tekscan is selected because of its compact

size and wide measurement range. To ensure the output

voltage has linear relation with the pressure, an analog

circuit has been designed and the wearer can change the

maximum measurable force. Obviously, the output vol-

tage of the FSR can reflect FRF and the output voltage

variation can reflect the pressure change. Considering

the complexity of the application environment of the

exoskeleton, Ribbon Switch developed by AbleNet is

selected as footswitch which is a moisture-resistant

switch that is activated by moving the 4 30.3-in or

10.2 35-cm activation surface in either direction with

4 oz or 110 g of force. The controller records the foots-

witch’s status, and the current and previous values of

the FSR can infer the four possible states of each leg

including stance, swing, heel-strike, and toe-off. The

rules to infer the leg is in which phase have been formu-

lated and are listed as follows. The leg is considered to

be in heel-strike mode when foot contact condition is

satisfied

fh p =0\fh c.0\ft c =0\SWh=0ð1Þ

where fh p and fh c are the previous and current FRFs

of the foot heel side, respectively; ft c is the current

FRF of the foot toe side; and SWhis the state of the

footswitch of the foot heel side. It means before the

Figure 1. Lower extremity exoskeleton.

Figure 2. Exoskeleton foot and its sensor system:

(a) force-sensing resistor and (b) footswitch.

Jin et al. 3

footswitch of the foot heel side is on, the FRF of the

foot heel side will have a sudden change when the leg is

in the heel-strike mode. The leg is considered to be in

stance mode when foot contact condition is satisfied

fh c.a\SWh=1\SWt=1\ft c .bð2Þ

where SWtis the state of the footswitch of the foot toe

side. In addition, aand bare the thresholds to detect

whether the foot heel side and toe side are on the floor.

These values can be set by the wearer. The leg is consid-

ered to be in toe-off mode when foot contact condition

is satisfied

fh c =0\SWh=0\SWt=1\ft c\gð3Þ

where ft p is the previous FRF of the foot toe side, and

gis the threshold of the FRF and it is used to judge

whether the foot is going to leave the floor. The leg is

considered to be in swing mode when foot contact con-

dition is satisfied

fh c =0\SWh=0\ft c =0\SWt=0ð4Þ

As a sole prototype, we installed one FSR and two

footswitches. Figure 3 shows the experiment results. We

have identified the four walking phases even if the FRF

of the foot front part is omitted.

In Figure 3, we have defined that number 0 repre-

sents swing mode, number 1 represents heel-strike

mode, number 2 represents stance mode, and number 3

represents toe-off mode. Based on the first subsystem,

we have inferred the walking phase of the human–

machine system. To activate the exoskeleton to shadow

the wearer with minimum interaction force in swing

phase, the second subsystem needs to detect the interac-

tion force.

Human–machine interaction model

As a pHRI-based exoskeleton, force sensor is a com-

monly used method to detect the HRI force. Figure 4

illustrates that the interaction forces are generated from

the connected interface where two 6-axis force sensors

are installed and the pHRI is modeled as an admittance

relationship.

Many studies have modeled the HRI as an impe-

dance model because the impedance control is an excel-

lent controller in hybrid control of position and

force.

30,31

Theoretically, the input of the impedance is

position and its output is force. Admittance control is

always seen as the inverse of impedance control while

its input is force and output is velocity or position

which is more fit to this system. So, we apply the

admittance relation in exoskeleton system as follows

_

ud(s)= Mas+Ba+Da

s

f(s)ð5Þ

where f(s) represents the HRI force detected by the

force sensor; Ma,Ba, and Daare the parameters of the

admittance model.

16

Each joint of the lower limb can be modeled as a

second-order biomechanical system in the musculoske-

letal system.

32

Lee has defined the human impedance

model as follows

H(s)= 1

Mhs2+Bhs+Kh

ð6Þ

where Mh,Bh, and Khare the virtual impedance

parameters.

In order to find the value of parameters, experiment

can be designed to measure and analyze the motions of

a human leg using the joint angle and forward kine-

matics. Relationship has been deduced as follows

33

Ma=100MhBh

Kh

,Ba=10Bh,Da=4Kh

3ð7Þ

To improve the operation convenience, we have

designed input interface which permits the wearer to

Figure 3. (a) State of the footswitch installed in toe and heel

part, (b) voltage of the FSR installed in toe part, and (c) state of

the corresponding leg.

Figure 4. pHRI model using the admittance relationship.

4Advances in Mechanical Engineering

adjust the admittance parameters according to wearing

feeling.

Thus far, the block diagram to control the swing leg

of the lower extremity exoskeleton can be designed as

shown in Figure 5.

Controller design

Considering the model error and uncertainties of the

hydraulic exoskeleton system, the dynamics model of

the swing leg can be written as

(M+DM)€

u+(C+DC)_

u+(G+DG)=t+dð8Þ

where M2R232is the symmetric positive-definite iner-

tia matrix; C2R232represents the centripetal Coriolis

matrix; G2R2stands for the gravitational vector;

t2R2is the joint torques; DM,DC, and DGrepresent

the model error; u2R2is the position of the joints; and

drepresents the unknown disturbance.

Because of the difficulties in deriving an exact math-

ematical model and the complexity of traditional fuzzy

controller, we have combined the single-input fuzzy

logic controller and adaptive fuzzy sliding mode con-

troller.

34,35

An assumption should be pointed out that

the desired trajectory of every joint of the lower extre-

mity exoskeleton has been inferred. The tracking error

can be defined as follows

e=uudð9Þ

where udis the desired trajectory.

In traditional sliding mode control, the first step is to

select a sliding surface. We have used an improved slid-

ing surface as follows

s=ce +_

e+kSat(se)ð10Þ

where c=diag½c1 cn,ci.0, and k=diag

½k1 kn,ki.0, are the design parameters which

satisfy Hurwitz.

Obiviously, atan(x) is a nonlinear saturation func-

tion, so another class of quasi-potential function is

introduced in equation (11) and its derivative is written

in equation (12), where smeans an adjustable

parameter

Sat(sx)=(sx)atan sxðÞ

1

2ln 1+(sx)2

ð11Þ

sat(sx) = atan(sx)ð12Þ

The number of input variables directly decides the

complexity of a convertional fuzzy system. As the

dimension and complexity of a system increase, the size

of the rule base increases exponentially. By defining the

sliding surface as the input variable of fuzzy rules, the

fuzzy system is simplified effectively and is easier to

implement. The fuzzy rules are given in the following

form

36

Rule 1 : IF sis Fi

s, THEN uis aið13Þ

where ai,i=1,2,... is the singleton control actions

and Fi

sis the label of fuzzy set. The defuzzification of

the control output is accomplished by the method of

center of gravity

tfz =P

m

i=1

aimi

P

m

i=1

mi

ð14Þ

where miis the membership grade of the ith rule.

From the sliding surface equations (10)–(12) and the

dynamic model equation (8), we can infer

_

s=c_

e+€

e+katan(se)ð15Þ

Tradition sliding mode control law can be designed

as

t=M€

ud+c_

e+katan(se)

+C_

u+G^

fð16Þ

Since the system dynamics and the disturbance are

unknown, the traditional control law is difficult to

implement. So, the adaptive fuzzy sliding mode control

(AFSMC) is proposed. The control law in equation

(14) has been improved by choosing aias an adjustable

parameter. Equation (14) can be written as

$GPLWWD

QFH

PRGHO

,QYHUVH

MDFRELDQ

PDWUL[

*HQHUDWH

MRLQW

DQJOHV

&RQWU

ROOHU 5RERW

:HDUHU

θ

&

d

θ

τ

v,

θ θ

&

0

d

f=

hr

f

−

+

++

−

−

Figure 5. Block diagram of the controller for human–robot interaction exoskeleton.

Jin et al. 5

tfz(s,a)=aTzð17Þ

where a=½a1,a2,...,amTand z=½z1,z2,...,zmTis

a regressive vector and is defined as zi=miPm

i=1mi.

There exists an optimal fuzzy control system to

approximate tbased on the fuzzy approximation the-

orem. It is listed as

t=t

fz(s,a)+e=aTz+eð18Þ

where eis the approximation error and is assumed to

satisfy e

jj

\E. Now, we use a fuzzy control system to

approximate t

fz(s,a). Thus, we obtain

^tfz(s,^a)=t

fz(s,a)=^aTzð19Þ

where ^ais the estimator of a.

Considering the approximation error, switching con-

trol is introduced to compensate it. Finally, the control

law is listed in equation (20) and the block diagram of

AFSMC is depicted in Figure 6

t=^tfz(s,^a)+tsc (s)ð20Þ

From equation (18), we can get

~tfz =^tfz t=^tfz t

fz eð21Þ

To infer the control equation, we define ~a=^aa.

Then, equation (21) can be simplified as

~tfz =~aTzeð22Þ

Combining equations (15) and (16), we can obtain

t=M€

ud_

s+€

e

+C_

u+G^

f=tM_

sð23Þ

Substituting equation (20) into equation (23), a new

equation can be written as

_

s=M1(^tfz +tsc t)=M1(~aTz+tsc e)ð24Þ

Thus, the Lyapunov function can be designed as

V(s,~a)= 1

2s2+M1

2l~aT~að25Þ

where lis a positive constant.

Differentiating equation (25) and using equation

(24), we obtain

_

V(s,~a)=s_

s+M1

l~aT_

~a

=sM1(~aTz+tsc e)+ M1

l~aT_

~a

=M1~aTsz+

_

~a

l

+sM1(tsc e)

ð26Þ

To achieve _

V(s,~a)0, so that the system is stable,

the adaptive law and the switching controller are

designed as

_

~a=_

^a=lszð27Þ

tsc =Esgn(s)ð28Þ

where E=max(e

jj

)+d,d.0, and sgn( ) is a sign

function.

Then, equation (26) can be written as

_

V(s,~a)= M1Es

jj

M1es

M1Es

jj

+M1e

jj

s

jj

=M1(Eejj)sjj

0

ð29Þ

By Barbalat’s lemma, we can conclude that s!0as

t!‘.

In equation (28), it is difficult to measure the approx-

imation error E. If a too large Ehas been determined, a

large chattering may take place. To decrease the chat-

ting problem, we adopt fuzzy switching method

37

to

approximate eand the approximation of Eis written as

^

E=^

bTfð30Þ

Figure 6. Illustrative diagram of the single-input adaptive fuzzy sliding mode control.

6Advances in Mechanical Engineering

where fis the fuzzy vector and ^

badjusts with adaptive

law.

To obtain the adaptive law, we define that bis the

best value and ~

b=b^

bis the estimated error. A

Lyapunov function is defined as

V2(s,~a,~

b)= 1

2s2+M1

2l1

~aT~a+M1

2l2

~

bT~

bð31Þ

where l2.0. Differentiating equation (31) and using

equation (27)

_

V2(s,~a,~

b)=sM1(^

Esgn(s)e)+ M1

l2

~

bT_

~

b

=M1^

Es

jj

es+1

l2f(^

EE)_

~

b

ð32Þ

For achieving _

V20, the adaptive law is designed as

_

^

b=l2fsð33Þ

Equation (32) can be written as

_

V2(s,~a,~

b)=M1^

Es

jj

es+^

EE

s

jj

M1(e

jjs

jj

Es

jj

)

=M1(Ee

jj

)s

jj

0

ð34Þ

By Barbalat’s lemma, we can conclude that s!0as

t!‘. By applying this estimation law, the AFSMC

system with a simplified fuzzy switching can be guaran-

teed to be stable.

Experiment

To verify the effectiveness of the sensor system in the

sole and the proposed algorithm to estimate human

intentions based on admittance and single-input adap-

tive fuzzy sliding mode control (S-AFSMC), we per-

form experiment on the swing leg of the lower extremity

exoskeleton as shown in Figure 1. The lower extremity

exoskeleton is a typical HRI system. So, the wearer will

first move his leg as the desired trajectory in the swing

phase. The controller needs to infer the wearer motion

intention and moves the swing leg to shadow the

motion in time.

As shown in Figure 5, the human motion intention

is inferred based on the force sensor, so the HRI force

is measured by the six-axis pressure sensor. The con-

troller proposed in this article aims to minimize the

force vector and finally reduced the force vector to

zero. Thus we can justify the system’s performance by

the amplitude of the interaction force vector. We com-

pared the proposed controller with the conventional

fuzzy sliding mode control (FSMC). The parameters in

equations (15) and (20) are given as:

c=5,k=15,s=1:5,l=30, and l2=0:5:

As a human–machine system, our aim is to let the

swing leg shadow the wearer. So the tracking error and

interaction force are chosen to justify the controller.

Figure 7 shows the results of the experiment.

Through Figure 7, it is obvious that the new control-

ler has less interaction force and better tracking perfor-

mance. The switching control law can compensate the

difference between the fuzzy controller and the desired

controller. The adaptive law of the switching gain can

decrease the chattering problem. The interaction force

is obtained by force sensor installed between wearer

and exoskeleton and the tracking error is calculated by

admittance model and joint encoders. To evaluate the

performance of the controller, we have adopted two

evaluating indicator: the first is the root-mean-square

(RMS) residuals and the second is the maximum abso-

lute value which are defined as

Figure 7. (a) Interaction force along the X axis in S-AFSMC and FSMC; (b) Interaction force along the Y axis in INSM and CSM

controller; (c) Tracking error of the hip joint in S-AFSMC and FSMC; (d) Tracking error of the knee joint in S-AFSMC and FSMC.

Jin et al. 7

eRMS =ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

1

NX

N

i=1

e2(i)

v

u

u

t

emax =max

0iNe(i)jjfg

ð35Þ

The evaluation results of the controller performance

are illustrated in Table 1. Obviously, the RMS value

and the maximum value of the two joints gained by

S-AFSMC are smaller than the ones gained by FSMC.

So the wearer needs less power to drive the swing leg

and feels more comfortable by using the S-AFSMC.

Conclusion

In this article, a sensor system to detect the centor of

force is designed in a sole. The estimation algorithm

based on FRF which can reflect wearer’s motion status

is investigated through the walking experiment. This

method can divide walking pace into four modes: heel-

strike mode, stance mode, toe-off mode and swing

mode. Admittance control is adopted to model the

HRI which is normally described by impedance model.

An input interface is designed to improve the wearer’s

feeling which permits the wearer to adjust the admit-

tance parameters. Further, considering the uncertain-

ties of the HRI system and the hydraulic system model,

an S-AFSMC is proposed with a novel nonlinear inte-

gral sliding surface. The whole controller contains

AFSMC and fuzzy switching control. Both adaptation

laws are designed based on Lyapunov stability theo-

rem. Therefore, the stability of the S-AFSMC can be

guaranteed. Finally, the proposed methods in this arti-

cle are verified on the lower extremity exoskeleton,

especially in the swing phase. Experiments prove the

effectiveness and reliability of the proposed controller

compared to traditional fuzzy sliding mode controller.

The wearer feels more comfortable to move the swing

leg. These methods can be applied in other HRI

system.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with

respect to the research, authorship, and/or publication of this

article.

Funding

The author(s) disclosed receipt of the following financial sup-

port for the research, authorship, and/or publication of this

article: This research work was supported by the Science

Fund for Creative Research Groups of the National Natural

Science Foundation of China (no. 51521064) and Zhejiang

Provincial Natural Science Foundation of China (no.

LY13E050001), and Hangzhou Civic Significant

Technological Innovation Project of China (no.

20132111A04) and SANLIAN (ShangHai) Group (no.

H20131864).

References

1. Chu A, Kazerooni H and Zoss A. On the biomimetic

design of the Berkeley lower extremity exoskeleton. In:

Proceedings of the IEEE international conference on

robotics and automation (eds A Chu, H Kazerooni and

A Zoss), Barcelona, 18–22 April 2005, pp.4345–4352.

New York: IEEE.

2. Kim S and Kazerooni H. High speed ring-based distribu-

ted networked control system for real-time multivariable

applications. In: Proceedings of the ASME 2004 interna-

tional mechanical engineering congress and exposition,

Anaheim, CA, 13–19 November 2004, pp.1123–1130.

New York: ASME.

3. Kim S, Anwar G and Kazerooni H. High-speed commu-

nication network for controls with the application on the

exoskeleton. In: Proceedings of the American control con-

ference, Boston, MA, 30 June–2 July 2004, pp.355–360.

New York: IEEE.

4. Walsh CJ, Endo K and Herr H. A quasi-passive leg exos-

keleton for load-carrying augmentation. Int J Human

Robot 2007; 4: 487–506.

5. Kwon S and Kim J. Real-time upper limb motion esti-

mation from surface electromyography and joint angular

velocities using an artificial neural network for human–

machine cooperation. IEEE T Inf Technol B 2011; 15:

522–530.

6. Lee H, Lee B, Kim W, et al. Human–robot cooperative

control based on pHRI (physical human–robot interac-

tion) of exoskeleton robot for a human upper extremity.

Int J Precis Eng Man 2012; 13: 985–992.

7. Suzuki K, Mito G, Kawamoto H, et al. Intention-based

walking support for paraplegia patients with Robot Suit

HAL. Adv Robotics 2007; 21: 1441–1469.

8. Lim D, Kim W, Lee H, et al. Development of a lower

extremity exoskeleton robot with a quasi-

anthropomorphic design approach for load carriage. In:

Table 1. Performance evaluation.

e

RMS

of joint (°)e

max

of joint (°)Fx

RMS

(N) Fx

max

(N) Fy

RMS

(N) Fy

max

(N)

Hip Knee Hip Knee

S-AFSMC 0.0056 0.1993 0.0104 0.325 7.9912 24.366 3.5343 8.9757

FSMC 0.018 0.561 0.0364 0.728 21.919 59.387 7.1646 16.185

S-AFSMC: single-input adaptive fuzzy sliding mode control; FSMC: fuzzy sliding mode control.

8Advances in Mechanical Engineering

Proceedings of the 2015 IEEE/RSJ international confer-

ence on intelligent robots and systems (IROS), Hamburg,

28 September–2 October 2015, pp.5345–5350. New York:

IEEE.

9. Toyama S and Yamamoto G. Development of wearable-

agri-robot: mechanism for agricultural work. In: Pro-

ceedings of the 2009 IEEE/RSJ international conference

on intelligent robots and systems, St. Louis, MO, 10–15

October 2009, pp.5801–5806. New York: IEEE.

10. Kawamoto H, Lee S, Kanbe S, et al. Power assist method

for HAL-3 using EMG-based feedback controller. In:

Proceedings of the 2003 IEEE international conference on

systems, man and cybernetics, Washington, DC, 8 Octo-

ber 2003, pp.1648–1653. New York: IEEE.

11. Kawamoto H, Taal S, Niniss H, et al. Voluntary motion

support control of Robot Suit HAL triggered by bioelec-

trical signal for hemiplegia. In: Proceedings of the 2010

annual international conference of the IEEE engineering in

medicine and biology society (EMBC), Buenos Aires,

Argentina, 31 August–1 September 2010, pp.462–466.

Piscataway, NJ: IEEE.

12. Lee H, Kim W, Han J, et al. The technical trend of the

exoskeleton robot system for human power assistance.

Int J Precis Eng Man 2012; 13: 1491–1497.

13. Rosen J and Perry JC. Upper limb powered exoskeleton.

Int J Human Robot 2007; 4: 529–548.

14. Kazerooni H and Steger R. The Berkeley lower extremity

exoskeleton. J Dyn Sys Meas Contr 2006; 128: 14–25.

15. Kazerooni H, Racine J-L, Huang L, et al. On the control

of the Berkeley lower extremity exoskeleton (BLEEX).

In: Proceedings of the 2005 IEEE international conference

on robotics and automation (ICRA 2005), Barcelona, 18–22

April 2005, pp.4353–4360. New York: IEEE.

16. Yu W, Rosen J and Li X. PID admittance control for an

upper limb exoskeleton. In: Proceedings of the 2011 Amer-

ican control conference, San Francisco, CA, 29 June–1

July 2011, pp.1124–1129. New York: IEEE.

17. Lee H-D, Lee B-K, Kim W-S, et al. Human–robot coop-

eration control based on a dynamic model of an upper

limb exoskeleton for human power amplification. Mecha-

tronics 2014; 24: 168–176.

18. Kim JH, Shim M, Ahn DH, et al. Desi gn of a knee

exoskeleton using foot pressure and knee torque sen-

sors. IntJAdvRobotSyst2015; 12: 112. DOI:

10.5772/60682.

19. Jung S, Hsia TC and Bonitz RG. Force tracking impe-

dance control for robot manipulators with an unknown

environment: theory, simulation, and experiment. Int J

Robot Res 2001; 20: 765–774.

20. Chen Z, Yao B and Wang Q. Accurate motion control of

linear motors with adaptive robust compensation of non-

linear electromagnetic field effect. IEEE/ASME T Mecha-

tron 2013; 18: 1122–1129.

21. Chen Z, Yao B and Wang Q. m-synthesis-based adaptive

robust control of linear motor driven stages with high-

frequency dynamics: a case study. IEEE/ASME T

Mechatron 2015; 20: 1482–1490.

22. Yao J, Jiao Z, Ma D, et al. High-accuracy tracking con-

trol of hydraulic rotary actuators with modeling uncer-

tainties. IEEE/ASME T Mechatron 2014; 19: 633–641.

23. Yao J, Jiao Z and Ma D. Extended-state-observer-based

output feedback nonlinear robust control of hydraulic

systems with backstepping. IEEE T Ind Electron 2014;

61: 6285–6293.

24. Sun W, Gao H and Kaynak O. Finite frequency HNcon-

trol for vehicle active suspension systems. IEEE T Control

Syst Tech 2011; 19: 416–422.

25. Sun W, Gao H and Kaynak O. Vibration isolation for active

suspensions with performance constraints and actuator

saturation. IEEE/ASME T Mechatron 2015; 20: 675–683.

26. Chen Z, Pan YJ and Gu J. Integrated adaptive robust

control for multilateral teleoperation systems under arbitrary

time delays. Int J Robust Nonlin Contr 2015; 26: 2708–2728.

27. Yuan ZC, Yao B and Zhu X. Time optimal contouring

control of industrial biaxial gantry: a high-efficient analy-

tical solution of trajectory planning. IEEE/ASME

TMECH 2016. DOI: 10.1109/2591518.

28. Kim H, Seo C, Shin YJ, et al. Locomotion control strat-

egy of hydraulic lower extremity exoskeleton robot. In:

Proceedings of the 2015 IEEE international conference on

advanced intelligent mechatronics, AIM, Busan, South

Korea, 7–11 July 2015, pp.577–582. New York: IEEE.

29. Aguilar-Sierra H, Yu W, Salazar S, et al. Design and

control of hybrid actuation lower limb exoskeleton. Adv

Mech Eng 2015; 7: 1–13.

30. Ku N, Kwon J-H, Cho Y-o, et al. Dynamic simulation

and experimental study of impedance control for robotic

orthosis to assist the overhead operations in the ship-

building process. In: Proceedings of the SICE annual con-

ference 2010, Taipei, Taiwan, 18–21 August 2010,

pp.131–136. New York: IEEE.

31. Chen S, Chen Z, Yao B, et al. Adaptive robust cascade force

control of 1-DOF hydraulic exoskeleton for human perfor-

mance augmentation. IEEE/ASME TMECH 2016. DOI:

10.1109/ 2614987.

32. Pledgie S, Barner KE, Agrawal SK, et al. Tremor sup-

pression through impedance control. IEEE T Rehab Eng

2000; 8: 53–59.

33. Huang HP, Jeng JC and Luo KY. Auto-tune system

using single-run relay feedback test and model-based con-

troller design. J Process Contr 2005; 15: 713–727.

34. Wai RJ, Lin CM and Hsu CF. Adaptive fuzzy sliding-

mode control for electrical servo drive. Fuzzy Sets Syst

2004; 143: 295–310.

35. Qin L, Liu FC and Liang LH. The application of adaptive

backstepping sliding mode for hybrid humanoid robot

arm trajectory tracking control. Adv Mech Eng 2014.

DOI: 10.1155/307985.

36. Wang J, Rad AB and Chan P. Indirect adaptive fuzzy

sliding mode control: part I: fuzzy switching. Fuzzy Sets

Syst 2001; 122: 21–30.

37. Jiao X and Jiang J. Design of adaptive switching control

for hypersonic aircraft. Adv Mech Eng 2015; 7(10). DOI:

1687814015610465.

Jin et al. 9