International Journal of Control, Automation and Systems 19(X) (2021) 1-14 ISSN:1598-6446 eISSN:2005-4092

http://dx.doi.org/10.1007/s12555-020-0427-4 http://www.springer.com/12555

A New Model-free Robust Adaptive Control of Cable-driven Robots

Yaoyao Wang*, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

Abstract: For high performance control of cable-driven robots, a new model-free robust adaptive control is pro-
posed using adaptive proportional-integral-derivative nonsingular fast terminal sliding mode (PID-NFTSM) in this
paper. The designed control scheme has three elements, which are time-delay estimation (TDE) and a new PID-
NFTSM manifold and a combined adaptive reaching law. The TDE element is applied to achieve the lumped system
dynamics and hence founds a practical model-free structure. Meanwhile, a new PID-NFTSM manifold is designed
to inherit the advantages of the NFTSM and high-order SM (HOSM) and integral SM (ISM) controls. Because of
this design, the PID-NFTSM manifold can effectively ensure the non-singularity and high comprehensive control
performance. Then, by combining the fast-TSM-type reaching law and a constant speed one with adaptive gain, we
design a new combined adaptive reaching law for performance reinforcement. Benefiting from above three parts,
our proposed control is model-free, highly precise and strongly robust. Stability is analyzed using Lyapunov theory.
The effectiveness of our control is proved by comparative simulation studies.

Keywords: Cable-driven robots, nonsingular fast terminal sliding mode (NFTSM), PID-NFTSM, robust adaptive
control, time-delay estimation (TDE).

1. INTRODUCTION through an output feedback method. To realize high per-

formance of a two-link flexible robotic manipulator, a
In recent decades, cable-driven robots have attracted full-state feedback control and an output feedback con-
much attention because of the distinct characters, such as trol were proposed using the n-dimensional discretized
small moving mass and good flexibility and high safety for model [10]. By using an artificial neural network (ANN)
working around human [1–3]. Benefiting from above char- to approximate the relation between actuators and soft
acteristics, cable-driven technique have been broadly uti- robot position, two sliding mode (SM) control schemes
lized in many practical systems, such as continuum robots were proposed and investigated [11]. Two synchronization
[4], parallel robot [5], and planar robotic system [6] etc. controllers were presented to guarantee the synchroniza-
However, the utilization of cable-driven technique also tion motion between cables and reduce the tracking errors
brings about extra control difficulties, such as large non- of the moving platform [12]. Afterwards, a novel adap-
linearities and strong dynamical couplings and low stiff- tive synchronization control (ASC) was proposed for the
ness and cable-pully network friction [7]. Hence, high per- cable-driven parallel robots (CDPRs) to further improve
formance control of cable-driven robots still remains an the control performance under uncertain kinematics and
open challenge to date. dynamics [13]. Satisfactory performance have been en-
sured using above control schemes, high control accuracy
To effectively handle above difficulties and ensure high
and fast response have been observed. But they still re-
control performance of robots with flexibilities, scholars
quire either detailed system dynamics or complex estima-
and engineers have proposed several methods [8–13]. In
tion algorithms, which is often not easy to realize under
[9], a novel control scheme was proposed for the flexi-
practical situations. A simple, efficient and accurate con-
ble robotic manipulator using the infinite-dimensional dy-
trol is what we want for practical applications.
namic model. The immeasurable signals were obtained

Manuscript received June 16, 2020; revised September 21, 2020 and October 31, 2020; accepted December 3, 2020. Recommended by
Associate Editor Maolin Jin under the direction of Editor Kyoung Kwan Ahn. This work was supported in part by the National Natural
Science Foundation of China (51705243),National Key R&D Program of China (2018YFB1309202), Natural Science Foundation of Jiangsu
Province (BK 20170789) and the Open Foundation of the State Key Laboratory of Fluid Power and Mechatronic Systems (GZKF-201915).

Yaoyao Wang, Lufang Liu, Bai Chen, and Hongtao Wu are with the College of Mechanical and Electrical Engineering, Nanjing University
of Aeronautics and Astronautics, Nanjing 210016, China (e-mails: [email protected], [email protected], {chenbye_cmee,
hongtaowu_cmee}@163.com). Yaoyao Wang is also with the State Key Laboratory of Fluid Power and Mechatronic Systems, Zhejiang
University, Hangzhou. 310027, China. Mingxing Yuan is the Institute of Robotics and Automatic Information System, Nankai University,
Tianjin 300071, China (e-mail: [email protected]). Qingxun Di is with North Automatic Control Technology Institute, Taiyuan 030006,
China (e-mail: [email protected]).
* Corresponding author.

c ICROS, KIEE and Springer 2021

2 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

The time-delay estimation (TDE) technique is a strong founds an attractive model-free) scheme. Then, a novel
tool to handle above expectations [14–21]. The core con- PID-NFTSM surface is designed to assure good control
cept of TDE is to utilize the intentionally time-delayed performance including high accuracy and fast dynamic
system signals to calculate the lumped system dynamics. response and strong robustness. Afterwards, an adaptive
By utilizing this novel mechanism, the TDE can provide a reaching law is proposed using PID-NFTSM variable to
fascinating model-free control scheme. Thanks to its obvi- strengthen the performance. The utilized adaptive strat-
ous attractions, TDE has been widely adopted for practical egy will effectively regulate the gain according to the
applications, such as manipulators [22–27], exoskeletons present performance and therefore suitable gain will be
[28,29], and humanoid robots [30–32], etc. On the other generated. Thanks to the TDE, our proposed control is
hand, the utilization of TDE will unavoidably lead to es- model-free and applicable for the real situations. Due to
timation error which is often known as the TDE error. By the designed PID-NFTSM manifold and adaptive reach-
analyzing the working mechanism of TDE, we can see that ing law, our control is highly accurate and fast convergent
when the system contains fast time-varying dynamics, like and strongly robust. The Lyapunov theory is applied to
frictions, the estimation with TDE will bring about rela- prove the stability of the system. Finally, four compara-
tive large error and may degrade the control performance tive simulations are conducted to verify the effectiveness
obviously [33]. To suppress the potential TDE error and and advantages of our proposed model-free robust adap-
ensure good control performance, scholars usually com- tive control.
bine robust control schemes with TDE, such as adaptive Conclusively, our contributions compared with the ex-
control, sliding mode (SM) control and others [34–41]. isting work [45] can be emphasized as following:
As a widely utilized robust control strategy, SM con- 1) Propose a new PID-NFTSM manifold, which elim-
trol and its variants have been widely utilized to sup- inates the singularity issue existing in [45] and can still
press the TDE error and obtained good control perfor- guarantee high control performance.
mance [22–29]. In [22], a novel nonsingular terminal SM 2) Design an adaptive reaching law, which can up-
(NTSM) control was designed using TDE technique for date the control gain accordingly and also solve the non-
the robot manipulators. High control accuracy and fast dy- decreasing and potential over-estimate issues in [45].
namical response were observed. Then, continuous NTSM 3) Develop a new model-free robust adaptive control
(CNTSM) control [42] and continuous nonsingular fast using above PID-NFTSM manifold, adaptive reaching law
terminal sliding mode (CNFTSM) control [43] were de- and TDE, which is fast convergent, highly accurate and
signed using TDE to further improve the control perfor- strongly robust.
mance given by [22]. Afterwards, TDE-based adaptive The rest is arranged as following: Section 2 gives the
super-twisting NFTSM (AST-NFTSM) control was also system dynamics and control objective, and Section 3
reported for control performance improvement [44]. To presents the main theoretical results. Finally, simulation
enhance performance with uncertainties, a novel adaptive results and conclusions are given in Section 4 and Section
fuzzy PID-NFTSM control was proposed for the robot 5.
manipulators recently [45]. By utilizing NFTSM surface
as an inner loop and PID surface as an outside one, the
reported PID-NFTSM manifold enjoys main merits of 2. SYSTEM DYNAMICS AND CONTROL
the NFTSM and integral SM (SM) and high-order SM OBJECTIVE
(HOSM) controls, such as high precision and fast response
and strong robustness. However, the recently reported The dynamics of n-DoFs (degree of freedoms) cable-
adaptive fuzzy PID-NFTSM control expresses three seri- driven robots are given as [2].
ous issues [45]. 1) The control scheme is naturally singu-
Iφ̈ + Dm φ̇ + τ joint = τ, (1)
lar). This problem is caused by the designed PID-NFTSM
surface and will lead to serious control signal fluctuation M (q) q̈ + C (q, q̇) q̇ + g (q) +f (q, q̇) +τd =τ joint , (2)


at .Then, the control performance may be degraded. 2) The KD φ̇ −q̇ +KP (φ −q) =τ joint , (3)
presented adaptive strategy keeps strictly non-negative,
which means the gain cannot be reduced overall. This de- where Dm and I are the damping and inertia matrices of
sign may result in over-estimate issue and even lead to in- system motors, φ and q stand for the position vectors of
stability. 3) The designed control still utilizes partial sys- the motors and joints. τ and τ joint represent torque vec-
tem dynamics, which can be further enhanced. tors given by the motors and joint compliance. M(q) and
To effectively solve above-mentioned issues and en- C (q, q̇) are inertia and Coriolis/centrifugal matrices, g(q)
sure satisfactory control performance for the cable-driven and f (q, q̇) are gravitational and friction vectors. τ d repre-
robots, we design a new model-free robust adaptive con- sents comprehensive uncertainties including time-varying
trol using TDE and PID-NFTSM. The proposed control disturbance and unknown dynamics. KD and KP represent
applies TDE to obtain system dynamics and therefore the joint damping and stiffness matrices.
 A New Model-free Robust Adaptive Control of Cable-driven Robots 3

To apply the TDE technique, substituing (1) into (2) and Combining (4), (10) and (11), we have
introducing a constant matrix M̄, we have Z
sPID =k p s+kI s+kD [ë+η1 ė+η2 ϖ̇ (e)]
M̄q̈ + N = τ, (4) Z
=k p s+kI s
where the M̄ is introduced for the utilization of TDE
+kD q̈d −M̄−1 (τ − N) +η1 ė+η2 ϖ̇ (e) . (12)
 
scheme, and N is utilized to describe the lumped system
dynamics and is mathematically defined as
To guarantee fast and robust dynamical performance dur-
ing reaching phase for sPID , we propose the following
 
N = M (q) −M̄ q̈ + C (q, q̇) q̇ + g (q)
combined adaptive reaching law as
+f (q, q̇) +Iφ̈ +Dm φ̇ +τd . (5)
ṡPID = −k1 sPID −k2 sig(sPID )β −k̂3 sgn (sPID ) , (13)
Taking (5) to analyze, we can see that the element N is
quite complicated and it mainly consists of three parts. where k1,2 =diag(k1,2i ), k1,2i > 0 are constant gains and
The first part is the rigid link dynamics described by (2); 0<β <1; meanwhile, the is updated by the following de-
the second one is motor dynamics by (1); and the third one signed adaptive law
is the flexible dynamics and unknown disturbance given

εi |sPID,i | sgn k3i,mid − k̂3i ,

by (3). What we try to accomplish is to design a proper 


control τ for the cable-driven robots and therefore it can  if k̂3i = k3i,max or k̂3i = k3i,min ,

˙
kˆ3i = (14)
provide highly accurate tracking of the given trajectory qd sgn(|sPID,i |−Θi )
in a simple matter. εi |s̃PID,i |

 sgn (|sPID,i | − Θi ) ,


if k3i,min < k̂3i < k3i,max ,
3. PROPOSED ROBUST ADAPTIVE CONTROL where k3i,max > 0 and k3i,min > 0 are the designed maxi-
AND DISCUSSIONS mum and minimum values, and k3i,mid =(k3i,max +k3i,min )/2.
3.1. Proposed robust adaptive control ε i and Θi are positive gains,while |s̃PID,i | = |sPID,i | + ϕi
under |sPID,i | < Θi and |s̃PID,i | = |sPID,i | under |sPID,i | ≥
Define the control error as e=qd -q, and the NFTSM
Θi The ϕ i is used to suppress the potential singularity issue
manifold is given as [ [16].
when sPID,i approaches 0. Combining (6) and (10)-(13),
s = ė+η1 e+η2 ϖ (e) , (6) we presents the robust adaptive control as
(
λ1i ei + λ2i sig(ei )ρ2i , s̄i 6= 0, |ei | < ∆i , τ = M̄u + N̂, (15)
ϖ(ei )i = (7)
sig(ei )ρ1i , s̄i = 0 ∨ (s̄i 6= 0, |ei | ≥ ∆i ) ,
 Z
u = q̈d + η1 ė + η2 ϖ̇ + kD −1 k p s + kI s
where s̄i = ėi + η1i ei + η2i sig(ei )ρ1i , sig(x)y = |x|y sgn(x), Z h i
and i=1 ∼ n. The η 1 , η 2 , ρ 1 , ρ 2 , ∆ are positive parame- + k1 sPID + k2 sig(sPID )β + k̂3 sgn (sPID ) ,
ters satisfying 0.5<ρ 1i <1, 1<ρ 2i <2.Condition (8) must be
(16)
fullfilled to ensure the continuity of ϖ (e) and ϖ̇ (e) .

ρ2i − ρ1i ρ1i −1 ρ1i − 1 ρ1i −ρ2i where N̂ stands for the estimation of N given in (5).
λ1i = ∆i , λ2i = ∆i . (8) Remark 1: The proposed PID-NFTSM surface (10) has
ρ2i − 1 ρ2i − 1
three parts: the proportional part k p s used to remain thead-
The first-order derivative of ϖ (e), i.e., ϖ̇ (e), is vantages of NFTSM, say fast R
convergence and high ac-
  curacy; the integral part kI s applied to obtain the su-
 λ1i + λ2i ρ2i |ei |ρ2i −1 ėi , s̄i 6= 0, |ei | < ∆i , periorities of ISM, such as strong robustness and high
ϖ̇(ei )i = steady precision; and the derivative part kD ṡ utilized to
ρ |e |ρ1i −1 ė , s̄ = 0 ∨ (s̄ 6= 0, |e | ≥ ∆ ) .
1i i i i i i i purse the merits of HOSM, say chattering elimination.
(9) Benefiting from above three parts, our proposed PID-
NFTSM manifold enjoys the main attractions of NFTSM,
Then, a new PID-NFTSM manifold is proposed as ISM and HOSM successfully. Moreover, thanks to the
Z proposed adaptive reaching law(13), the performance has
sPID =k p s+kI s+kD ṡ, (10) been effectively enhanced. Thus, our control is fast con-
vergent, highly accurate and strongly robust due to the
where k p , kI , kD are constant gains. The first-order deriva- PID-NFTSM manifold (6), (10) and adaptive reaching law
tive of s (6) can be given as (13).
Remark 2: Note from (13) that there is a signum
ṡ = ë+η1 ė+η2 ϖ̇ (e) . (11) term k̂3 sgn (sPID ) in the combined adaptive reaching law,
4 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

which, however, will not generate chattering issue. By ob- 3.3. Comparisons with recently reported method
serving our proposed control (16), we can see that the Our model-free robust adaptive control is inspired by
signum term is arranged behind an integral function and the one given in [45], which will be deduced afterwards.
thus the chattering issue has been successfully eliminated. First, the NFTSM surface is reported in [45] as
Remark 3: Although our proposed approach can pro-
vide high control performance, it also has obvious limita- s1 = e + µ1 sig(e)ϑ1 + µ2 sig(ė)ϑ2 , (20)
tions. First, its structure is more complex than the widely
used PID control. Second, it also has more parameters to where µ1,2 = diag(µ1,2i ), µ1,2i > 0, 1 < ϑ2i < 2, and ϑ1i >
tune than PID control. Hence, our proposed approach is ϑ2i . Note the term e[ϑ ] used in [45] is equivalent to our
more suitable for the applications which require high con- sig(e)ϑ (20).
trol performance and can deal with our limiations; other- Then, the PID-NFTSM surface was presented as
wise, the PID control may be more suitable due to its sim- Z
ple structure and relative satisfactory control performance. s1PID = k1p s1 + k1I s1 + k1D ṡ1 , (21)
Remark 4: By observing our proposed control (15)-
(16), we can see that the control gain M̄ is the most sen- where k1P , k1I , k1D are constant gains.
sitive parameter than all others. This is reasonable since Afterwards, the adaptive reaching law was selected as
M̄ is a coefficient to the overall control term u (16). Thus,

ṡ1PID = − k̂4 + a sgn (s1PID ) , (22)
to further improve the control performance, we will try
to update M̄ with properly designed adaptive algorithm in k̂˙ = |s
4 1PID |/c, (23)
the future studies.
where a is a constant postive gain and c stands for the
adaptation gain.
3.2. Application of TDE scheme
With (20)-(23), the PID-NFTSM control was given as
As shown in (15), if we can efficiently obtain N̂, then
the objective can be easily fulfilled. However, as explained τ1 = M̄u1 + N̂1 , (24)
above, the term N is quite complicated and can be very dif-  
ficult and time-consuming to obtain using traditional tech- u1 = q̈d + µ2 −1 ϑ2 −1 1 + µ1 ϑ1 |e|ϑ1 −1 sig(ė)2−ϑ2
nique, such as parameter identification. To solve this prob-
 Z
−1 −1 1−ϑ2
lem, we utilize TDE in this work to obtain N̂ in a simple + µ2 J2 |ė| k1p s1 + k1I s1
but highly effective way as Z


+ k̂4 + a sgn (s1PID ) , (25)
N̂t ∼
= Nt−L , (17)
where N̂1 stands for the estimation of N using TDE.
where L represents the delayed time which is often se- Remark 5: To make the comparisons fair, we reduced
lected as one or several sampling periods. the PID-NFTSM control from [45] using the same way
Combining (4) and (17), then we have as ours. Above control (24)-(25) is almost the same as
the original one from [45] despite the different forms they
N̂t ∼
= τt−L − M̄q̈t−L . (18) have. The main difference is that the reduced control (24)-
(25) is totally model-free while the original one still re-
Taking (18) to analyze, we can observe that the core con- quires system dynamics. For example, the equivalent con-
cept of TDE is to use the intentionally time-delayed sys- trol ueq , given in (14) of [45], contains the known function
tem dynamics to estimate the current system dynamics. f (x1 , x2 ), i.e., partial system dynamics element.
Therefore, an attractive model-free scheme is founded. To Then, comparisons between our control (15)-(16) with
be specific, no system dynamics has been utilized in (18) the one from [45] (24)-(25) are highlighted as
and only two terms are required, i.e., τt−L and q̈t−L . The 1) Model-free and continuous: note from (19) and
τt−L can be obtained using τ, and q̈t−L is usually achieved (16) that our proposed robust control requires no system
by numerical differentiation [43]. To effectively supress dynamics and leads to an attractive model-free scheme.
the noise effect caused by above numerical differentiation, Thanks to this benefit, our control is suitable for practi-
small M̄ or low-pass filter are very helpful. cal applications and quite easy to use. For comparisons, as
Substituting (18) into (15), our model-free robust adap- explained in Remark 5, the original PID-NFTSM control
tive control can be given as from [45] still requires system dynamics which may lead
to extra difficulties for practical applications. In the mean-
τ = M̄u + τt−L − M̄q̈t−L , (19) time, both our control and (24)-(25) are naturally continu-
ous since the sign element hides behind the integral oper-
where u is the same as (16). ator as shown in (16) and (25).
 A New Model-free Robust Adaptive Control of Cable-driven Robots 5

2) Singularity issue: by observing (9), we can clearly where k̄1,2 and β̄ are positive gains fulfilling 0 < β̄ < 1 ,
see that the singularity issue has been successfully elim- and the sliding variable s is defined as(6). Then, the com-
inated. When s̄i = 0 ∨ (s̄i 6= 0, |ei | ≥ ∆i ) holds, we have parisons (mainly differences) between our control (15)-
ϖ̇(ei )i = ρ1i |ei |ρ1i −1 ėi . Considering −0.5 < ρ1i − 1 < 0, (16) with (27) from [16] can be given as following:
there may be a singularity issue at ei= 0, i.e., under s̄i = 1) Sliding mode surface: note from (6)-(10) that our
0.When s̄i = 0, ei= 0, we have ei = ėi = 0 which means control utilizes a novel PID-NFTSM surface which can
ϖ̇(ei )i = ρ1i |ei |ρ1i = 0. Obviously, there is no singularity enjoy the advantages of NFTSM, ISM and HOSM simul-
under this case. When s̄i 6= 0, |ei | < ∆i holds, the power taneously. Meanwhile, the (27) utilizes only a NFTSM
index is positive considering 0 < ρ2i − 1 < 1 which means surface which may lead to relative weaker control perfor-
no singularity will occur either. Hence, our NFTSM (6) mance than ours.
and PID-NFTSM (10) surfaces are nonsingular. For com- 2) Adaptive algorithm: by observing (13)-(14), we can
parisons, the NFTSM surface (20) and its deduced control see that a novel adaptive algorithm has been utilized in
(24)-(25) given in [45] still has the singularity issue. Tak- our combined adaptive reaching law. For comparison, tra-
ing (25) to analyze, we can see that it contains an element ditional fast-TSM-type reaching law with constant gains
|ė|1−ϑ2 . Since 1 < ϑ2i < 2, thus the power index is neg- was adopted in (27) from [16].
ative and will lead to singularity issue. To conclude, our Remark 7: Comparing with the methods from [15] and
proposed method is singularity-free while the one given [45], our newly proposed control has the following dis-
in [45] still has singularity problem. tinctive contributions and differences:
3) Adaptive algorithm: to enhance the performance, 1) Propose a novel PID-NFTSM surface, which has suc-
both our proposed control and the exiting one [45] use cessfully solve the singularity issue in [45] and can also
adaptive algorithms to update the control gain. Our de- provide better control performance than the NFTSM sur-
signed adaptive algorithm (14) is naturally bounded, and face given in [16];
it can also timely and accurately regulate the control gain 2) Design a novel TDE-based adaptive PID-NFTSM
using the current control performance. To be specific, our control using a newly proposed adaptive reaching law,
adaptive algorithm will increase the control gain when the which is model-free and effectively eliminates the non-
control performance is relative poor, i.e. |sPID,i | ≥ Θi ; then, decreasing and potential over-estimate problems in [45].
the control performance will be enhanced using this large Meanwhile, our control ensures better comprehensive
gain leading to |sPID,i | < Θi . Afterwards, our adaptive al- control performance than the one from [16] thanks to its
gorithm will reduce the control gain accordingly. Thanks PID-NFTSM surface and adaptive reaching law.
to this novel design, our adaptive algorithm can provide
proper control gain accordingly and successfully avoid 3.4. Guidelines for gain selections
the over-estimated issue. For comparisons, as shown in To make our proposed control easy to utilize in prac-
(23), the adaptive algorithm utilized in [45] is strictly non- tical applications, guidelines for gain selections are given
negative which will lead to over-estimated gain and even in this subsection. Note that the delay time L is usually
instability. To handle this so-called “parameter drift prob- chosen as several sampling periods. Then, the following
lem”, the following revised version was utilized in [45] guidelines will be used to tune the parameters which is
( similar with the ones given in [2].
˙k̂ = 0, if |s1PID | < Ξ, 1) Set kp =kI = ε =k1 =k2 =k3i,max =k3i,min = ∆ = 0
4 (26)
|s1PID | /c, if |s1PID | ≥ Ξ, and ρ1 = ρ2 = 1n , note condition (8) is invalid and λ 1 =
λ 2 = 0 is used under this case. Then, tune η 1 , η 2 based
where Ξ is a small positive constant. on the results from [16]; tune M̄ by increasing itself from
By analyzing (26), we can see that (26) still remains small one untill the control performanec tend to degrade;
non-negative feature which means the gain keeps non- 2) Keep kp =kI = ε =k1 =k2 = k3i,max =k3i,min = 0, set
decreasing. It will still generate over-estimated gains after ∆ as a relative large one. Tune ∆ by decreasing it from
a long run. In the meantime, the selection of Ξ can also small one while checking the control performance; tune
be quite tricky under practical applications.Above claims ρ 1 and ρ 2 by deceasing themselves from 1n and 2n while
will be demonstrated by comparative simulations. checking the errors and vibrations;
Remark 6: Although the inner NFTSM surface (6) is 3) Maintain ε =k1 =k2 =k3i,max =k3i,min = 0, tune
taken from [16], our designed TDE-based adaptive PID- kp =kI and kD considering the meaning of PID control;
NFTSM is distinctively different from the one given in 4) Maintain ε =k3i,max =k3i,min = 0, tune k1 , k2 by in-
[16] which can be written as creasing them from 0 to large ones while checking the
  control performance;
τ =M̄ q̈d + η1 ė + η2 ϖ̇ + k̄1 s + k̄2 sig(s)β̄ 5) Maintain ε = 0, tune k3i,max by increaing it from 0 to
large one untill the control tend to vibrate; tune k3,min by
+ τt−L − M̄q̈t−L , (27) decreasing it from k3i,max until the system fails to track
6 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

the reference trajectory; tune ε by increasing it from 0 to 0, i.e., k̂˙ 3i = −εi |sPID,i |, will be analyzed in the following
large one while checking the control performance; two cases:
6) Repeat above procedures until the control perfor- ˙
Case 1: k̂3i = k3i,max . Then, we have k̂3i =−εi |sPID,i |,
mance is satisfactory. k̃3,i = 0 and V̇i ≤ − k1,i sPID,i 2 + k2,i |sPID,i |β +1 . Thus, the
3.5. Stability proof system trajectory will consistently converge towards equi-
librium point until sPID,i =0 is arrived.
Lemma 1 [16]: When the NFTSM surface (6)-(8) is
Case 2: |sPID,i | < Θi and k3i,min < k̂3i < k3i,max . Then,
bounded as |si | < Φi , the tracking error e and ė will con- ˙k̂ = −(ε |s −1
verge to the following field as: 3i i PID,i | + ϕi ) will hold . Afterwards, the (32)
will turn into
|ei | ≤ Φe = max {Φe1 , Φe2 } ,  
n
−1 o V̇i ≤ − sPID,i k1,i sPID,i + k2,i sig(sPID,i )β
Φe1 = min ∆i , Φ η1i + η2i ∆i ρ1i −1 ,
n  n + k̃3,i (εi |sPID,i | + ϕi )−1

1/ρ1ioo
Φe2 =min max ∆i , Φη1i−1 , max ∆i , Φη2i−1 ,
 
≤ − sPID,i k1,i sPID,i + k2,i sig(sPID,i )β + ϕ̄i , (33)
ρ1i
|ėi | ≤ Φė = Φ + η1i Φe + η2i Φe . (28)
where ϕ̄i = k̃3,i ϕi −1 is utilized for simplicity.
For the proof of Lemma 1, please refer to the Theorem 1 For (33), it can be given into the following two forms:
and its proof from [16]. h i
V̇i ≤ − k1,i − ϕ̄i sPID,i −2 sPID,i 2 + k2,i |sPID,i |β +1


Then, the stability of closed-loop control system is
proved in two steps. First, the boundedness of the PID- h i
NFTSM variable sPID is proved; then, the boundedness of ≤ − k4,i sPID,i 2 + k2,i |sPID,i |β +1 , (34)
tracking error e and will be proved using Lemma 1. h   i
Step 1: boundedness proof of sPID . V̇i ≤ − k1,i sPID,i 2 + k2,i −ϕ̄i |sPID,i |−β −1 |sPID,i |β +1
Combining the control (15)-(16) and PID-NFTSM sur- h i
≤ − k1,i sPID,i 2 + k5,i |sPID,i |β +1 , (35)
face (6), (10), we have

where k4,i = k1,i − ϕ̄i sPID,i −2 , k5,i = k2,i − ϕ̄i |sPID,i |−β −1
Zh i
sPID =kD ψ − k1 sPID +k2 sig(sPID )β +k̂3 sgn (sPID )
Z When k4 , i> 0 and k5 , i> 0 hold, the system trajectory
=k p s + kI s + kD [q̈d − M̄−1 (τ − N) + η1 ė will consistently converge towards equilibrium point un-
til the condition k4 , i> 0 and k5 , i> 0 fail. Therefore, the
+ η2 ϖ̇(e)], (29) system trajectory will converge to the following filed as:

where ψ = −M̄−1 N̂ − N represents the bounded TDE |sPID,i | ≤ Ωi = min {Ω1i , Ω2i , Θi } ,

error, and the ψ̇ is also bounded by |ψ̇i | ≤ Ei . The
1/2

Ω1i = k̃3,i ϕi −1 k1,i −1 ,
bounded analysis can be found in [17, 22]. To make the
1/(β +1)
Ω2i = k̃3,i ϕi −1 k2,i −1


analysis clear, we will analyze the ith-DOF for simplicity. . (36)
Define a Lyapunov candidate as
Based on above analysis, we can clearly observe that
Vi = sPID,i 2 2 + k̃3,i
2
 
2, (30) the designed PID-NFTSM variable sPID,i will be bounded
under our proposed control (15)-(16).
where k̃3,i = k3 max,i − k̂3,i ≥ 0. Step 2: boundedness proof of e and ė.
Differentiating (30) along time and taking |ψ̇i | ≤ Ei into Considering (10) and the bounded input bounded out-
consideration , we have put (BIBO) stabilty theory, the NFTSM variable si will be
  bounded as |si | ≤ Φi using proper gains kP , kI , kD when
V̇i = − sPID,i k1,i sPID,i + k2,i sig(sPID,i )β sPID is bounded. Then, according to Lemma 1, the e and ė
will be bounded as (28).
− k̂3,i |sPID,i | + kD,i ψ̇i sPID,i − k̃3,i k̂˙ 3,i . (31) Finally, the stability analysis is completed.
Since |ψ̇i | ≤ Ei , choose k3i,min ≥ kD,i Ei . Then, (31)becomes
4. SIMULATION DEMONSTRATIONS
 
V̇i ≤ −sPID,i k1,i sPID,i + k2,i sig(sPID,i )β − k̃3,i k̂˙ 3,i . 4.1. Setup
(32) Four comparative simulations are conducted to ver-
ify the validity and superiorities of our proposed control.
Due to k̃3,i ≥ 0, the (32) will be limited by a stricter First, the system is commanded to track a smooth desired
condition k̂˙ 3,i < 0 than k̂˙ 3,i ≥ 0. Thus, the condition k̂˙ 3,i < trajectory; then, a triangular wave is utilized as desired
 A New Model-free Robust Adaptive Control of Cable-driven Robots 7

trajectory; afterwards, a square wave is sent to the sys- PID-NFTSM manifold (6), (10) and adaptive reaching law
tem; finally, three different disturbances are utilized in the (13) compared with the existing twos. Above results also
simulations to verify the robustness of our control. The verified the superiorities of PID-NFTSM surface over the
rigid part of the dynamics (2) is obtained from [22]; mean- traditional NFTSM surface under exactly the same param-
while, the other parameters are set as I = [0.5, 0; 0, 0.2] eters. By analyzing Fig. 1(c), we can see that the control
Kg·m2 ,Dm = [0.1, 0; 0, 0.02] N·m·s/rad, KD = [50, 0; 0, torques for all three control schemes are relative smooth
40] N·m·s/rad, KP = [2, 0; 0, 1]×104 N·m/rad. For the and no noticeable chatters have been observed.
unknown disturbance τd , it is selected as τd = sin(πt/2) Considering Fig. 1(e), we can see that obvious estima-
N·m. Three controllers are simulated which are our pro- tion errors have appeared in the initial and peak phases,
posed control (15)-(16), M. Van’s control (24)-(25) from which are primarily invoked by the frictions. These large
[45] and Y. Wang’s control from [16]. estimation errors, named TDE errors, will greatly influ-
The following parameters are used for our pro- ence the control performance as indicated in Fig. 1(b)
posed control during all three simulation cases as and Fig. 1(d). To suppress TDE errors, adaptive reach-
η1 = η2 =diag (1,1), ρ1 =diag (0.9, 0.9), ρ2 =diag ing laws (13) and (22) have been utilized. The corre-
(1.05, 1.05), ∆ = 0.01×diag(1,1),kP =diag(80,80), sponding adaptive gains are presented in Fig. 1(f). By
kI =diag(6,6), kD =diag(1,1), k1 =k2 =diag(1,1), comparing Fig. 1(e) with Fig. 1(f), we can observe that
β = 0.6, k3,max =diag(20,20), k3,min =diag (1,1), our adaptive algorithm increases k̂3 when the control per-
ε =diag(0.5,0.3), Θ =diag(0.5, 0.5), ϕ = 0.01×diag (1,1), formance is relative good and reduce it otherwise. This
M̄ =diag(0.15,0.15), L= 1 ms, and the initial value of k̂3 adaptive mechanism can ensure both high control perfor-
is set to k3,min . To make the comparisons fair and compa- mance and small/bounded adaptive gain. For comparisons,
rable, we select the same or equivalent parameters for as indicated in Fig. 1(f), the adaptive algorithm given in
M. Van’s control (24)-(25) from [45], and ϑ1 =diag [45] has generated consistently increasing gain, which is
(1.2,1.2), ϑ2 = 1/ρ1 , µ1 = µ2 =diag(1,1), k1P =kP , very dangerous for long-time-running practical applica-
k1I =kI , k1D =kD , c=diag(0.2, 0.15), a=diag(0.5, 0.5), tions. Therefore, our adaptive reaching law can assure bet-
and the initial value of k̂4 are also set to k3,min . In the ter comprehensive control performance than the existing
meantime, the parameters for Y. Wang’s control are se- ones.
lected exactly the same as ours. For M. Van’s control (24), By analyzing Figs. 1(g)-(h), we can see that the singu-
define larity issue has appeared under Van’s control (24)-(25) in
the simulations. As indicated in the A-D parts of Fig. 1(h),
û =µ2 −1 ϑ2 −1 |ė|1−ϑ2 when ė fluctuates around zero, the |ė|1−ϑ2 and û will fluc-
 
tuate even more violently, which may bring about perfor-
Z Z

× k1p s1 + k1I s1 + k̂4 + a sgn (s1PID ) . mance degradation and even dangerous situations. Fur-
(37) thermore, when ė crosses ė = 0 , the control element
û will change hugely. For comparisons, our control is
Since Van’s control (24)-(25) from [45] has obvious singu- singularity-free thanks to our PID-NFTSM manifold (6),
larity issue, the initial value of ė cannot be zero, otherwise (10). Hence, better comprehensive control performance
the simulations cannot be excuted. Thus, we set qt=0 =- can be ensured using our proposed method.
diag(0.5, 0.5) deg,q̈t=0 = diag(5, 5) deg/s, and the initial 2) Case 2: in this case, a triangular wave is sent to
values for φ are set to zero. the system. And the obtained results are given in Fig. 2.
Note from Figs. 2(a)-(d) that satisfactory tracking of tri-
4.2. Results and discussions angular wave trajectory is assured by three controls.This
1) Case 1: a smooth desired trajectory is sent to the once again verifies the effectiveness of TDE, our PID-
system as shown in Fig. 1(a). Note qd,t=0 = q̇d,t=0 = 0, NFTSM manifold and adaptive reaching law, the existing
thus we have et=0 =diag(0.5,0.5) deg and ėt=0 =diag(5, 5) twos from [45], and the NFTSM surface from [16]. Ob-
deg/s. Then, the obtained results are presented in Fig. 1. viously, higher precision and faster response are ensured
As shown in Figs. 1(a)-(d), the system has successfully using our method, which strongly demonstrates the advan-
tracked the desired smooth trajectory with high control ac- tages of our method over existing ones. Taking Figs. 2(e)-
curacy under all three control schemes. This strongly ver- (f) to analyze, we can observe the advantages of our adap-
ifies the effectiveness of TDE scheme, our PID-NFTSM tive algorithm than the existing one from [45]. Ours can
manifold (6), (10) and adaptive reaching law (13), ex- regulate the control gains and generate proper large ones
isting PID-NFTSM manifold (20), (21) and its adaptive accordingly; in the meantime, the adaptive algorithm from
reaching law (22) and NFTSM surface from [16]. Still, our [45] has led to obvious over-estimated gains due to its
proposed algorithm can provide lower tracking error and non-decreasing structure. In the meantime, singularity is-
faster dynamical response as shown in Figs. 1(b)-(d) than sue is observed from M. Van’s control under this case as
the other twos. These results prove the superiorities of our Figs. 2(g)-(h). Taking A-D parts of Fig. 2(h) to analyze,
8 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

Fig. 1. Control performance using our method (black solid), Van’s method (blue dashed) and Wang’s method (red dashdot)
under case one: (a)-(d) are trajectory tracking performance,

tracking error, control torque and enlarged peak phase
of tracking error; (e) is the estimation error N − N̂ ; (f) is the adaptive gain k̂3 ; (g) tracking error ė(black solid),
û(red dashdot) and |ė|1−ϑ2 (blue dashdot, right y-axis) using Van’s control; (h) is partial enlarged of (g).
 A New Model-free Robust Adaptive Control of Cable-driven Robots 9

Fig. 2. Control performance using our method (black solid), Van’s method (blue dashed) and Wang’s method (red dashdot)
under case two: (a)-(d) are trajectory tracking performance,

tracking error, control torque and enlarged peak phase
of tracking error; (e) is the estimation error N − N̂ ; (f) is the adaptive gain k̂3 ; (g) tracking error ė(black solid),
û(red dashdot) and |ė|1−ϑ2 (blue dashdot, right y-axis) using Van’s control; (h) is partial enlarged of (g).
10 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

we can see that the |ė|1−ϑ2 fluctuates violently when ė ducted to prove the superiorities of our model-free robust
fluctuates around zero. This may lead to obvious control adaptive control.
performance degradation. For comparisons, our control is To further enhance the control performance, we will
nonsingular because of PID-NFTSM manifold(20), (21). conduct the following studies in the future: 1) designing
3) Case 3: a square wave is utilized as desired trajectory new PID-NFTSM surface for better comprehensive per-
and the results are presented in Fig. 3. formance, such as fast response and high accuracy; 2) uti-
By analyzing Fig. 3, we can see that similar results lizing adaptive algorithm to regulate the gain M̄.
have been obtained as case one and two. The effectiveness
and advantages of our PID-NFTSM manifold and adaptive COMPLIANCE WITH ETHICS GUIDELINES
reaching law have been further demonstrated.
4) Case 4: to verify the robustness of our control, we The authors declare that they have no conflict of interest
simulate three different subcases using our proposed con- or financial conflicts to disclose.
trol, i.e. samll τ d1 =sin (πt/2) and middle τ d2 =5sin (πt/2)
and largeτ d3 =10sin (πt/2). The results are given in Fig. 4.
REFERENCES
As indicated in Figs. 4(a)-(f), the control performance
remain almost unchanged when the disturbance increases [1] W. T. Townsend, The Effect of Transmission Design on
from τ d1 =sin (πt/2) toτ d3 =10sin (πt/2), which effectively Force-controlled Manipulator Performance, Ph.D. disser-
prove the robustness of our control. tation, Cambridge Artificial Intelligence Laboratory, Mas-
Overall, our method together with the other twos can all sachusetts Institute Technology, Cambridge, MA, 1988.
esnure good control performance under above three cases, [2] Y. Wang, F. Yan, J. Chen, F. Ju, and B. Chen, “A new adap-
which obviously demonstrate the effectiveness of the TDE tive time-delay control scheme for cable-driven manipula-
scheme, our PID-NFTSM manifold (6), (10) and adap- tors,” IEEE Transactions on Industrial Informatics, vol. 15,
tive reaching law (13), existing twos (20)-(22) [45] and no. 6, pp.3469-3481, 2019.
NFTSM manifold from [22]. Still, our control has obvi-
[3] Y. Wang, L. Liu, D. Wang, F. Ju, and B. Chen, “Time-delay
ous advantages over the existing twos from [16] and [45] control using a novel nonlinear adaptive law for accurate
in the following aspects: trajectory tracking of cable-driven robots,” IEEE Transac-
1) our proposed PID-NFTSM surface has successfully tions on Industrial Informatics, vol. 18, no. 8, pp. 5234-
eliminated the singularity issue shown in Van’s control 5243, 2020.
from [45] and assured higher control performance than the [4] G. He, Y. Fan, T. Su, L. Zhao, and Q. Zhao, “Variable
traditional NFTSM manifold from [16]; impedance control of cable actuated continuum manipu-
2) our adaptive algorithm (14) can update the gain ac- lators,” International Journal of Control, Automation and
cordingly and has avoided the over-estimation issue in the Systems, vol. 18, no. 7, pp. 1839-1852, 2020.
non-decreasing adaptive algorithm from [45];
[5] H. Choi, J. Piao, E. Kim, J. Jung, E. Choi, J. Park, and C.S.
3) our combined adaptive reaching law enjoys both fast- Kim, “Intuitive bilateral teleoperation of a cable-driven
TSM-type reaching law and constant speed one with adap- parallel robot controlled by a cable-driven parallel robot,”
tive gain, and thus it provides better performance than the International Journal of Control, Automation and Systems,
ones from [16] and [45]. vol. 18, no. 7, pp. 1792-1805, 2020.
To conclude, our TDE-based adaptive PID-NFTSM [6] H. J. Asl and J. Yoon, “Stable assist-as-needed controller
control is model-free, nonsingular and high performance. design for a planar cable-driven robotic system,” Interna-
tional Journal of Control, Automation and Systems, vol.
5. CONCLUSION 15,no. 6, pp. 2871-2882, 2017.
[7] M. Miyasaka, M. Haghighipanah, Y. Li, J. Matheson,
A new model-free robust adaptive control is proposed A. Lewis, and B. Hannaford, “Modeling cable-driven
using PID-NFTSM scheme for cable-driven robots. The robot with hysteresis and cable–pulley network friction,”
presented control utilizes TDE for system dynamics es- IEEE/ASME Transactions on Mechatronics, vol. 25, no. 2,
timation and founds a model-free structure. Then, a new pp. 1095-1104, 2020.
PID-NFTSM manifold is proposed to take advan-tages of [8] M. Yuan, Z. Chen, B. Yao, and X. Liu, “Fast and accu-
NFTSM and ISM and HOSM, simultaneously. Due to this rate motion tracking of a linear motor system under kine-
novel structure, high performance can be guaranteed. Af- matic and dynamic constraints: an integrated planning and
terwards, a new adaptive reaching law is utilized based on control approach,” IEEE Transactions on Control Systems
our PID-NFTSM variable to further enhance the perfor- Technology, vol. 99, pp. 1-8, 2019.
mance. Benefiting from TDE and our PID-NFTSM mani- [9] W. He, X. Y. He, M. F. Zou, and H. Y. Li, “PDE model-
fold and adaptive reaching law, our control is model-free, based boundary control design for a flexible robotic manip-
highly accurate and strongly robust. Stability is proved us- ulator with input backlash,” IEEE Transactions on Control
ing Lyapunov theory. Comparative simulations are con- Systems Technology, vol. 27, no. 2, pp. 790-797, 2019.
 A New Model-free Robust Adaptive Control of Cable-driven Robots 11

Fig. 3. Control performance using our method (black solid), Van’s method (blue dashed) and Wang’s method (red dashdot)
under case two: (a)-(d) are trajectory tracking performance,

tracking error, control torque and enlarged peak phase
of tracking error; (e) is the estimation error N − N̂ ; (f) is the adaptive gain k̂3 ; (g) tracking error ė(black solid),
û(red dashdot) and |ė|1−ϑ2 (blue dashdot, right y-axis) using Van’s control; (h) is partial enlarged of (g).
12 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

Fig. 4. Control performance using our method (black solid), Van’s method (blue dashed) and Wang’s method (red dashdot)
under case two: (a)-(d) are trajectory tracking performance,

tracking error, control torque and enlarged peak phase
of tracking error; (e) is the estimation error N − N̂ ; (f) is the adaptive gain k̂3 .

[10] H. J. Gao, W. He, C. Zhou, and C. Y. Sun, “Neural network matics and dynamics,” IEEE Transactions on Industrial
control of a two-link flexible robotic manipulator using as- Electronics, 2020. DOI: 10.1109/TIE.2020.3013776
sumed mode method,” IEEE Transactions on Industrial In- [14] T. C. Hsia, T. A. Lasky, and Z. Guo, “Robust independent
formatics, vol. 15, no. 2, pp. 755-765, 2019. joint controller design for industrial robot manipulators,”
[11] G. Zheng, Y. Zhou, and M. Ju, “Robust control of a silicone IEEE Transactions on Industrial Electronics, vol. 38, no.
soft robot using neural networks,” ISA Transactions, vol. 1, pp. 21–25, 1991.
100, pp. 38-45, 2020. [15] Y. Wang, S. Li, D. Wang, F. Ju, B. Chen, and H. Wu, “Adap-
[12] W. Shang, B. Y. Zhang, B. Zhang, F. Zhang, and S. Cong, tive time-delay control for cable-driven manipulators with
“Synchronization control in the cable space for cable- enhanced nonsingular fast terminal sliding mode,” IEEE
driven parallel robots,” IEEE Transactions on Industrial Transactions on Industrial Electronics, vol. 68, no. 3, pp.
Electronics, vol. 66, no. 6, pp. 4544-4554, 2019. 2356–2367, 1991.
[13] H. Ji, W. Shang, and S. Cong, “Adaptive synchronization [16] Y. Wang, K. Zhu, B. Chen, and M. Jin, “Model-free con-
control of cable-driven parallel robots with uncertain kine- tinuous nonsingular fast terminal sliding mode control for
 A New Model-free Robust Adaptive Control of Cable-driven Robots 13

cable-driven manipulators,” ISA Transactions, vol. 98, pp. [29] S. Han, H. Wang, Y. Tian, and N. Christov, “Time-delay es-
483-495, 2020. timation based computed torque control with robust adap-
tive RBF neural network compensator for a rehabilita-
[17] Y. Kali, M. Saad, K. Benjelloun, and C. Khairallah, “Super-
tion exoskeleton,” ISA Transactions, vol. 97, pp. 171-181,
twisting algorithm with time delay estimation for uncertain
2020.
robot manipulators,” Nonlinear Dynamics, vol. 93, no. 2,
pp. 557-569, 2018. [30] M. Jin, J. Lee, and N. G. Tsagarakis, “Model-free robust
adaptive control of humanoid robots with flexible joints,”
[18] M. Jin, S. H. Kang, P. H. Chang, and J. Lee, “Robust con- IEEE Transactions on Industrial Electronics, vol. 64, no.
trol of robot manipulators using inclusive and enhanced 2, pp. 1706-1715, 2017.
time delay control,” IEEE/ASME Transactions on Mecha-
tronics, vol. 22, no. 5, pp. 2141-2152, 2017. [31] M. Pi, Y. Kang C. Xu, G. Li, and Z. Li, “Adaptive time-
delay balance control of biped robots,” IEEE Transactions
[19] J. Baek, S. Cho, and S. Han, “Practical time-delay control on Industrial Electronics, vol. 67, no. 4, pp. 2934-2944,
with adaptive gains for trajectory tracking of robot manip- 2020.
ulators,” IEEE Transactions on Industrial Electronics, vol.
65, no. 7, pp. 5682-5692, 2018. [32] S. Roy, J. Lee, and S. Baldi, “A new adaptive-robust de-
sign for time delay control under state-dependent stability
[20] J. Baek, W. Kwon, B. Kim, and S. Han, “A widely-adaptive condition,” IEEE Transactions on Control Systems Tech-
time-delay control and its application to robot manipula- nology, vol. 99, pp. 1-8, 2020.
tors,” IEEE Transactions on Industrial Electronics, vol. 66,
no. 7, pp. 5332-5342, 2019. [33] Y. Wang, B. Li, F. Yan, and B. Chen, “Practical adaptive
fractional-order nonsingular terminal sliding mode control
[21] Y. Wang, F. Yan, K. W. Zhu, B. Chen, and H. T. Wu, “A for a cable-driven manipulator,” International Journal of
new practical robust control of cable-driven manipulators Robust and Nonlinear Control, vol. 29, no. 5, pp. 1396-
using time-delay estimation,” International Journal of Ro- 1417, 2019.
bust and Nonlinear Control, vol. 29, no.11, pp. 3405-3425,
[34] H. Bae, M. Jin, J. Suh, J. Y. Lee, P. H. Chang, and D. Ahn,
2019.
“Control of robot manipulators using time-delay estimation
[22] M. Jin, J. Lee, P. H. Chang, and C. Choi, “Practical nonsin- and fuzzy logic systems,” Journal of Electrical Engineer-
gular terminal sliding-mode control of robot manipulators ing and Technology, vol. 12, no. 3, pp. 1271-1279, 2017.
for high-accuracy tracking control,” IEEE Transactions on
[35] B. Park, P. T. Pham, and K. S. Hong, “Model reference
Industrial Electronics, vol. 56, no.9, pp. 3593-3601, 2009.
robust adaptive control of control element drive mechanism
[23] Y. Wang, L. Gu, Y. Xu, and X. Cao, “Practical tracking in a nuclear power plant,” International Journal of Control,
control of robot manipulators with continuous fractional- Automation, and Systems, vol. 18, no. 7, pp. 1651-1661,
order nonsingular terminal sliding mode,” IEEE Transac- 2020.
tions on Industrial Electronics, vol. 63, no. 10, pp. 6194-
[36] C. Sun, G. Gong, and H. Yang, “Sliding mode control with
6204, 2016.
adaptive fuzzy immune feedback reaching law,” Interna-
[24] M. Jin, S. H. Kang, and P. H. Chang, “Robust compliant tional Journal of Control, Automation, and Systems, vol.
motion control of robot with nonlinear friction using time 18, no. 2, pp. 363-373, 2020.
delay estimation,” IEEE Transactions on Industrial Elec- [37] F. Han, and Y. Jia, “Sliding mode boundary control for a
tronics, vol. 55, no. 1, pp. 258–269, 2008. planar two-link rigid-flexible manipulator with input dis-
[25] J. Lee, P. H. Chang, and M. Jin, “An adaptive gain dy- turbances,” International Journal of Control, Automation,
namics for time delay control improves accuracy and ro- and Systems, vol. 18, no. 2, pp. 351-362, 2020.
bustness to significant payload changes for robots,” IEEE [38] S. Yu, X. Yu, B. Shirinzadeh, and Z. Man, “Continuous
Transactions on Industrial Electronics, vol. 67, no. 4, pp. finite-time control for robotic manipulators with terminal
3076-3085, 2020. sliding mode,” Automatica, vol. 41, no. 11, pp. 1957–1964,
[26] Y. Wang, L. Liu, J. Chen, F. Ju, B. Chen, and H. Wu, “Prac- 2005.
tical robust control of cable-driven robots with feedfor- [39] K. Eliker and W. Zhang, “Finite-time adaptive integral
ward compensation,” Advances in Engineering Software, backstepping fast terminal sliding mode control applica-
vol. 145, pp. 102801, 2020. tion on quadrotor UAV,” International Journal of Control,
[27] S. Ahmed, H. Wang, and Y. Tian, “Adaptive high-order Automation, and Systems, vol. 18, no.2, pp. 415-430, 2020.
terminal sliding mode control based on time delay esti- [40] M. Rahmani, H. Komijani, and M. H. Rahman, “New slid-
mation for the robotic manipulators with backlash hys- ing mode control of 2-DOF robot manipulator based on ex-
teresis,” IEEE Transactions on Systems, Man, Cybernetics: tended grey wolf optimizer,” International Journal of Con-
Systems, vol. 51, no. 2, pp. 1128-1137, 2021. trol, Automation, and Systems, vol. 18, no. 6, pp. 1572-
1580, 2020.
[28] S. Ahmed, H. Wang, and Y. Tian, “Model-free control us-
ing time delay estimation and fractional-order nonsingu- [41] L. Qiao and W. Zhang, “Trajectory tracking control of
lar fast terminal sliding mode for uncertain lower-limb ex- AUVs via adaptive fast nonsingular integral terminal slid-
oskeleton,” Journal of Vibration and Control, vol. 24, no. ing mode control,” IEEE Transactions on Industrial Infor-
22, pp. 5273-5290, 2018. matics, vol. 16, no. 2, pp. 1248-1258, 2020.
14 Yaoyao Wang, Lufang Liu, Mingxing Yuan, Qingxun Di, Bai Chen, and Hongtao Wu

[42] M. Jin, J. Lee, and K. K. Ahn, “Continuous nonsingular Qingxun Di received his B.S. degree in
terminal sliding-mode control of shape memory alloy actu- mechanical engineering from Yanshan
ators using time delay estimation,” IEEE/ASME Transac- University, Tsitsihar, China in 2014, and
tions on Mechatronics, vol. 20, no. 2, pp. 899-909, 2015. an M.S. degree in mechanical engineer-
[43] Y. Wang, B. Chen, and H. Wu, “Joint space tracking control ing from Zhejiang University, Hangzhou,
of underwater vehicle-manipulator systems with continu- China in 2017. He is currently a researcher
ous fast nonsingular terminal sliding mode,” Proceedings in the North Automatic Control Technol-
of the Institution of Mechanical Engineers, Part M: Jour- ogy Institute, Taiyuan, China. His current
nal of Engineering for the Maritime Environment, vol. 232, research interests include servo systems
no. 4, pp. 448-458, 2018. and robust control algorithms development.

[44] Y. Wang, J. Chen, F. Yan, K. Zhu, and B. Chen, “Adaptive

super-twisting fractional-order nonsingular terminal slid- Bai Chen received his B.S. and Ph.D.
ing mode control of cable-driven manipulators,” ISA Trans- degrees in mechanical engineering from
actions, vol. 86, pp. 163-180, 2019. Zhejiang University, Hangzhou, China, in
2000 and 2005, respectively. He is cur-
[45] M. Van, X. P. Do, and M. Mavrovouniotis, “Self-tuning
rently a Professor and Ph.D. candidate su-
fuzzy PID-nonsingular fast terminal sliding mode control
pervisor in the College of Mechanical and
for robust fault tolerant control of robot manipulators,” ISA
Electrical Engineering, Nanjing University
Transactions, vol. 96, pp. 60-86, 2020.
of Aeronautics and Astronautics, Nanjing,
China. His research interests include de-
sign and control of surgical robots, cable-driven robots and
Yaoyao wang received his B.S. degree sperm-like swimming micro robots.
in mechanical engineering from Southeast
University, Nanjing, China in 2011 and
a Ph.D. degree in mechanical engineer- Hongtao Wu received his B.S. degree
ing from Zhejiang University, Hangzhou, in mechanical engineering from Northeast
China in 2016. He is currently an associate Heavy Machinery Institute (the name was
professor in the College of Mechanical and changed to Yanshan University in 1997),
Electrical Engineering, Nanjing University Tsitsihar, China in 1982, his M.S., and
of Aeronautics and Astronautics, Nanjing, Ph.D. degrees in mechanical engineering
China. His current research interests include robust control and from Tianjin University, Tianjin, China, in
adaptive control of mechatronic systems, design and control of 1985 and 1992, respectively. He has been
cable-driven systems, such as cable-driven underwater manipu- a Professor in Nanjing University of Aero-
lators and cable-driven flying manipulators. nautics and Astronautics (NUAA) since 1998. He has presided
over numbers of the national research project of China. He
teaches courses of “Introduction to robotics” and “Computa-
Lufang Liu received his B.S. degree tional multibody system dynamics” in NUAA. His research in-
from Zhejiang University of Technology, terests include multibody system dynamics, robotics, parallel
Hangzhou, China in 2019. He is currently mechanisms and punching machines.
a Master’s degree candidate at the College
of Mechanical and Electrical Engineer-
Publisher’s Note Springer Nature remains neutral with regard
ing, Nanjing University of Aeronautics to jurisdictional claims in published maps and institutional affil-
and Astronautics, Nanjing, China. His cur-
iations.
rent research interests include cable-driven
manipulators and aerial manipulators.

Mingxing Yuan received his B.S. degree

in aircraft manufacturing engineering from
the Nanjing University of Aeronautics and
Astronautics, Nanjing, China, in 2013, and
a Ph.D. degree in mechatronic control en-
gineering from Zhejiang University, Zhe-
jiang, China, in 2019. From November
2016 to November 2017, he was a Vis-
iting Scholar in mechanical engineering
with Purdue University, West Lafayette, IN, USA. He is cur-
rently a Lecturer with the Institute of Robotics and Automatic
Information System and the Tianjin Key Laboratory of Intelli-
gent Robotics, Nankai University, Tianjin. His research interests
include motion planning, nonlinear adaptive robust control, and
telerobotics.