IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 69, NO.

4, APRIL 2022 2151

A Novel Interaction Controller Design for Robotic

Manipulators With Arbitrary Convergence Time
Yana Yang , Changchun Hua , Senior Member, IEEE, and Junpeng Li

Abstract—In this brief, robot-environment interaction control force sensors is easy to impose a burden on system integra-
problem of n-link robotic manipulator system without force sen- tion. Therefore, sensorless based robot control has attracted
sors is investigated. At first, a novel sliding mode (SM) based more and more researchers’ attention. Reference [8] intro-
two-layer adaptive force observer is proposed to estimate the
robot-environment interaction force. In comparison with the most duced the early external force estimation methods and applied
existing force observers, the proposed force observer ensures these methods to robot control. In [9], the idea of distur-
arbitrary-time convergence of the observer errors and relaxes bance observer is also be used to design robotic force observer.
assumptions on the external force. Moreover, an admittance con- Furthermore, according to the generalized momentum, a force
troller is developed to ensure exact trajectory tracking within observer was proposed without involving joint acceleration for
arbitrary time. It is mathematically proved that both trajectory
tracking errors and force estimation errors will converge to zero robotic system [10].
at any setting time. Finally, numerical simulations and compar- Although both the controllers and the observers mentioned
isons show that the proposed force observer-based control scheme above have realized the stable operation of the robot, the rated
can provide superior interaction performance. convergence problem is indeed ignored. It is more desirable
Index Terms—Interaction control, nonlinear force observer, to guarantee the tracking error and force estimation error con-
robotic manipulator, arbitrary-time convergence. verging to zero within finite/fixed time. Yet, it is observed that
for the traditional finite-time approach, the system settling time
function depends on its initial conditions [11]. Additionally,
I. I NTRODUCTION the fixed-time controller design relies on the information of
UE TO the inevitable physical interaction with
D the environment or human-being, satisfactory
environment/human-being interactive behavior has grad-
global system behavior [12]. Therefore, new types of stability
called predefined-time stability [13] or called prescribed-time
stability [14] and arbitrary-time stability [15] have been intro-
ually become the main control design goal of robots [1]–[3]. duced, allowing the settling time to be directly user defined
A large number of researchers have made great efforts through one or several specific parameters. However, in [13]
to provide numerous interaction control methods. Among and [15], zero-error stabilization problem is investigated and
these methods, admittance control can produce a satisfactory in [14] the tracking errors converge to a bounded neighborhood
interaction behavior by making good use of the external of the origin. Besides, by employing the time scaling function-
interaction force so as to adjust its motion [4], [5] and based method [13], [15], the time-varying gains approach
has been widely used in robot active compliant control. In infinity as time tends to the desired convergence time and it
addition, due to its fast transient response and robustness is difficult to be extended to zero-error tracking control.
against uncertainties and disturbances, SM based admittance Considering these problems, arbitrary-time interaction con-
control has attracted considerable attention in the fields of trol problem is investigated without force sensors in this brief.
robotic control [6], [7]. The main difficulty is to realize arbitrary-time convergence
In order to establish the ideal admittance model, force of tracking error and force observer error simultaneously.
observer is often required to provide force sensing. Yet, the On the one hand, a novel arbitrary-time two-layer adap-
force sensors are always expensive and the installation of tive SM force observer is presented firstly to estimate the
external force exactly. Compared with some existing force
Manuscript received August 17, 2021; revised October 12, 2021 and observers, the assumption on external force is less strict,
October 29, 2021; accepted November 12, 2021. Date of publication
November 18, 2021; date of current version March 28, 2022. This work was because any bound on the external force is not consid-
supported in part by the National Natural Science Foundation of China under ered. This significantly includes a wider range of force. On
Grant 61933009 and Grant 62073276; in part by the Top talents of Hebei the other hand, a new SM based admittance tracking con-
provincial Education Department under Grant BJ2019047; in part by the Key
Projects of Science and Technology Plan of Colleges and Universities of Hebei troller is proposed. The designed control law is continuous
Provincial Department of Education under Grant ZD2019031; and in part by and can ensure arbitrary-time convergence of the trajec-
the Natural Science Foundation of Hebei Province under Grant F2020203062. tory tracking error. The application of SM can simplify the
This brief was recommended by Associate Editor J. Wu. (Corresponding
author: Yana Yang.) design process and enhance the robustness of the closed-loop
The authors are with the School of Electrical Engineering, University of system [16]. It is to be observed that the upper bound on
Yanshan, Qinhuangdao 066004, China (e-mail: [email protected]). system/observer settling time given in this brief is independent
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCSII.2021.3128958. of system initial conditions, any system parameter can be set as
Digital Object Identifier 10.1109/TCSII.2021.3128958 operator will.
1549-7747 
c 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on October 28,2023 at 12:00:37 UTC from IEEE Xplore. Restrictions apply.
2152 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 69, NO. 4, APRIL 2022

Notation: In is an unit matrix with proper dimension n. ∀x = 0 and the derivative of V along the trajectories of the
sign(·) is a sign function. · represents the Euclidean 2-norm system (3) satisfies
of a vector. For the vector x = [x1 , . . . , xn ]T , xi , i = 1
1, 2, . . . , n denotes the ith term of vector x, |x|γ = xx1−γ , V̇ ≤ − exp(V ρ )V 1−ρ (4)
Ta ρ
∂|x|γ T γ
γ −1 , ∂x = γ |xT |γ −1 , γ
∂x = [In − (γ − 1) xxx
2 ]x ∂x Then, the origin is globally arbitrary-time stable for (3) and
is a positive constant. Instead of distinguishing between force T(x0 ) < Ta .
and torque, they are called force for convenience of further Proof: By solving the equation (4), it has V(t) ≤
1
presentation. (ln( −1 1
)) ρ . If Ta−1 t + exp(−V(0)ρ ) = 1, V(t) = 0
Ta t+exp(−V(0)ρ )
can be obtained. Thus, it can be derived that T(x0 ) ≤
1−exp(−V(0)ρ )
II. P RELIMINARIES AND P ROBLEM S TATEMENT . Since 0 ≤ 1 − exp(−V(0)ρ ) < 1, one has
Ta−1
Consider a general nonlinear n-link rigid robotic manipula- T(x0 ) < Ta . The proof is completed.
tor given as the following model: Remark 1: The variable exponent function has already been
used in backstepping control of robotic manipulator [19]. But
M(q)q̈ + C(q, q̇)q̇ + G(q) = τ + τext (1) to the best of the authors’ knowledge, it has never been used
where q ∈ n is the vector of generalized coordinates; for observer and interaction controller design. Additionally,
M(q), C(q, q̇) ∈ n×n are the system inertia matrix and cen- compared with [19], the application of SM strategy will sim-
tripetal/coriolis terms; G(q) ∈ n is the vector of gravitational plify the control design process and enhance its robustness for
torque; τ ∈ n is the designed control input torque; τext ∈ n disturbed systems.
denotes the robot-environment interaction force.
Some useful properties of (1) are presented as follows. III. M AIN R ESULTS
Property 1: M(q) is a symmetric positive-definite function. A. Adaptive Arbitrary-Time Force Observer Design
Property 2: Matrix Ṁ(q) = C(q, q̇) + C T (q, q̇). In this subsection, based on generalized momentum a novel
Next, an admittance model introduced to describe the rela- adaptive arbitrary-time force observer is constructed firstly.
tionship between the external force τext and the trajectory of In general cases, the robot’s generalized momentum can be
the robot (1) as described as  = M(q)q̇ [10], its time derivative is given by
˙ = Ṁ(q)q̇ + M(q)q̈. Then by applying the property 2 of
Md (q̈r − q̈d ) + Cd (q̇r − q̇d ) + Kd (qr − qd ) = −τext (2)
robot system, it has
where qr , qd ∈ n denote the virtual and desired joint posi- ˙ = Ṁ(q)q̇ − C(q, q̇)q̇ − G(q) + τ + τext

tions, respectively; Md , Cd and Kd are matrices of desired
= C T (q, q̇)q̇ − G(q) + τ + τext (5)
system inertia, damping and stiffness, respectively, which are
usually to be chosen as diagonal and positive definite. In Let us define x1 = q, x2 = , then one has

reality, it is more desirable that the admittance parameters ẋ1 = M(x1 )−1 x2
can be adjusted according to time-varying environment and (6)
ẋ2 = C T (x1 , ẋ1 )ẋ1 − G(x1 ) + τ + τext
task by applying linear quadratic regulator, adaptive dynamic
programming, reinforcement learning and so on [5]. Introduce the following auxiliary error variable evir = ξ −x2
Assumption 1: Both qd and qr are bounded and with
differentiable. ξ̇ = C T (x1 , ẋ1 )ẋ1 − G(x1 ) + τ − 1 (evir , To1 ) + vF (7)
Assumption 2: The first and second derivatives of τext sat-
where 1 (evir , To1 ) = To11ρo1 exp(evir ρo1 ) evirevirρo1 , 0 < ρo1 <
isfy τ̇ext  < τ̄1 and τ̈ext  < τ̄2 , where τ̄1 is assumed to be
1 and To1 > 0.
unknown and only the order of τ̄2 is known.
Theorem 1: Considering the system (6), if the vF is
For a nonautonomous nonlinear system
designed as
x = f (t, x; Ta ), x(0) = x0 (3) v̇F = −αo sign(so ) − 2 (so , To2 ), vF (0) = 0 (8)
where x ∈ is the system state; Ta ∈
n p
is the constant where αo ∈ + , with so = ėvir + 1 (evir , To1 ) and
designed parameters; f : + × n → n represents a nonlinear 2 (so , To2 ) = To21ρo2 exp(so ρo2 ) sosoρo2 , 0 < ρo2 < 1 and
function satisfying f (t, 0; T) = 0, i.e., x = 0 is an equilibrium To2 > 0. Then the estimation of the force τext is obtained as
point of (3). τ̂ext = vF within To2 and the auxiliary error variable evir will
To save space, the interested reader can consult [17], [18] also converge to zero in arbitrary time To = To1 + To2 .
for details on finite-time stability and fixed-time stability. Proof: Choose Lyapunov function candidate as Vobs = so .
Definition 1: The origin of (3) is arbitrary-time stable if it Differentiating evir with respect to time, it has ėvir = ξ̇ − ẋ2 =
is fixed-time stable, and ∃Ta > 0, which is a known designed −1 (evir , To1 ) + vF − τext . According to the definition of so ,
parameter and can be evaluated in advance. It is possible to one has so = vF − τext . Applying v̇F , the derivative of so is
adjust Ta arbitrarily under the assumption that such choices deduced as ṡo = v̇F − τ̇ext = −αo sign(so ) − 2 (so , To2 ) − τ̇ext .
are permitted for the system physical condition. The ẋ2 can be obtained by applying the predefined-time robust
Lemma 1: Assume that there exist a real-valued continu- exact differentiator given in [20].
ously differentiable function V : n → + ∪ {0}, and real Differentiating Vobs with respect to time and applying the
numbers Ta > 0, and 0 < ρ < 1 such that V(0) = 0, V(x) > 0, Assumption 2, it has
Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on October 28,2023 at 12:00:37 UTC from IEEE Xplore. Restrictions apply.
YANG et al.: NOVEL INTERACTION CONTROLLER DESIGN FOR ROBOTIC MANIPULATORS 2153

∂so 
V̇obs = ṡo
∂so
∂so 
≤ −αo + τ̄1 − 2 (so , To2 )
∂so
1
=− exp(V ρ2 )V 1−ρ2 (9)
To2 ρ2
which yields the arbitrary-time convergence of the variable so
according to the Lemma 1. Directly, it obtains that vF = τ̂ext
within To2 . Furthermore, based on the definition of so , one
gets ėvir = −1 (evir , To1 ). Therefore, the error variable evir
will also converge to zero in arbitrary time To according to Fig. 1. Block diagram of the force observer-based arbitrary-time interaction
the Lemma 1. The proof is completed. control scheme.
Remark 2: In most cases, it is difficult to choose the αo
to guarantee αo ≥ τ̄1 , when τ̄1 is unknown. Considering this
problem, the two layers adaptive law is designed as where e2 (sec , Tec2 ) = Tec21ρec2 exp(sec ρec2 ) secsecρec2 , 0 <
ρec2 < 1 and Tec2 > 0. The block diagram of the proposed
α̇o (t) = −βo (t)sign(δo (t)) (10) control law is shown in Fig. 1.
βo (t) = βod + ro (t) (11) Theorem 2: Consider robot system (1) with the

γo δo (t) if δo (t) > δod admittance model (2) and the controller (15), then e1
ṙo (t) = (12) and e2 will converge to zero within arbitrary time as
0 otherwise
1  Tec = Tec1 + Tec2 .
δo (t) = αo (t) − ūeq−d  − o (13) Proof: Consider the Lyapunov candidate Vcon = sec .
ao
· 1 Differentiating Vcon with respect to time along the trajectory
ūeq−d = [−αo sign(so ) − 2 (so , To2 ) − ūeq−d ] (14) of sec , it has
ιo
If the constants 0 < ao < 1, δod , γo , o and φo > ∂sec  de1 (e1 , Tec1 )
V̇con = (q̈ − q̈r + ) (16)
sup(1,  dtd (ūeq−d (t))/τ̄2 ) are chosen to make the following ∂sec dt
inequality 14 o2 > δod 2 + 1 ( φo τ̄2 )2 holds, then the condition
γo ao Substituting the system model (1) and controller (15) into
αo (t) ≥ τ̄1 can be ensured in finite time. It results a persis- ρec2 1−ρ
(16), it has V̇con = − Tec21ρec2 exp(Vcon )Vcon ec2 . According to
tent sliding motion. Furthermore, the gains αo (t) and βo (t) the Lemma 1, it is easy to obtain that Vcon will converge to
are bounded. By following [21], the two-layer adaptive law in zero in arbitrary time Tec2 . Based on the definition of Vcon ,
(10)-(14) can be employed to achieve these results. the arbitrary-time convergence of sec also can be ensured.
Remark 3: Compared with the existing force observers, our Moreover, with the designed e1 (e1 , Tec1 ), it has that e1 and
force observers has three distinguished advantages: 1) the mea- e2 will converge to zero within Tec = Tec1 + Tec2 directly
surement of joint acceleration q̈ and the inverse of the inertia for case 1. For case 2 |e1 i | ≤ φi , e1 and e2 will approach
matrix M(q) are not required; 2) the assumptions on external zero along the general sliding manifold. Because of the small
force has an upper bound or the first derivative of external enough φi , the convergence time can be neglected. The proof
force is known are removed; 3) force estimation accuracy and is completed.
speed are improved and the estimation time can be adjusted Remark 4: Compared with the existing finite/fixed-time con-
based on the actual demand, arbitrarily. trollers, the arbitrary-time controller (15) proposed in our
paper is simple and elegant. Additionally, compared with
B. Arbitrary-Time SM Controller Design the predefined/prescribed time and arbitrary time control
Define a new SM variable sec = e2 + e1 (e1 , Tec1 ) with approaches, the developed SM control scheme can provide the
position and velocity tracking errors e1 = q − qr , e2 = q̇ − q̇r capabilities of zero-error arbitrary-time tracking, disturbance
and rejection, and singularity avoidance.
⎧
e 
⎪
⎪
⎪
1
exp(|e1 i |ρec1 ) e 1 ρi ec1
⎨ Tec1 ρec1 | 1 i| IV. S IMULATION R ESULTS
e1 i = if sec i = 0 or sec i = 0, |e1 i | > φi In this section, the simulation will be conducted on a two-
⎪
⎪ k1i e1 i + k2i sign(e1 i )e1 2i
⎪
⎩ link rigid robotic manipulator. The dynamic model of the robot
if sec i = 0, |e1 i | ≤ φi and specific system parameters can be described and presented
where φi is a small positive constant, 0 < ρec1 < 1, Tec1 > 0, as [11]. q = [ q1 q2 ]T is the joint position vector of the first
ρ
exp(φi ec1 ) and the second joint, respectively. The desired positions are
ζi = 1
Tec1 ρec1 φ ρec1 , k1i = ζi (2 − ρec1 )φi−1 , k2i = ζi (ρec1 −
i q1d (t) = 0.5 sin(0.35π t)(rad), q2d (t) = 0.5 sin(0.2π t)(rad).
1)φi−2 . The initial states are chosen as q1 (0) = 3/4π(rad),
Then the SM controller is designed as q2 (0) = 1/4π(rad), q̇1 (0) = q̇2 (0) = 0(rad/s). The τext is
defined as
τ = C(x1 , ẋ1 )x2 + G(x1 ) − τ̂ext + M(x1 )(q̈r 
de1 (e1 , Tec1 ) ke (q − qe ), q ≥ qe
− − e2 (sec , Tec2 )) (15) τext = ,
dt 0, q < qe
Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on October 28,2023 at 12:00:37 UTC from IEEE Xplore. Restrictions apply.
2154 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 69, NO. 4, APRIL 2022

Fig. 2. (a) Force observer response. (b) Force estimation errors. Fig. 4. (a) Values of e1 with ρec1 = ρec2 = 17 . (b) Values of e1 with
(c) Trajectories of evir . (d) Adjustment of system trajectories. ρec1 = ρec2 = 15 . (c) Values of e1 with ρec1 = ρec2 = 13 . (d) Values of e1
with ρec1 = ρec2 = 12 .

Fig. 3. (a) Values of sec with Tec1 = Tec2 = 1.5s. (b) Values of sec with
Tec1 = Tec2 = 0.5s. (c) Values of e1 with Tec1 = Tec2 = 1.5s. (d)Values of
e1 with Tec1 = Tec2 = 0.5s.
Fig. 5. (a) Values of tracking error with two functions. (b) Values of control
input with two functions. (c) Values of sec with our controller and the con-
where qe = [0; 0] denotes the contact joint position between troller proposed in [13]. (d) Values of e1 with our controller and the controller
proposed in [13].
the robotic manipulator and the object and ke = 10 is the
scaling coefficient. The parameters applied in the brief are set
as Md = diag[1, 1], Cd = diag[20, 20], Kd = diag[100, 100];
Tec1 = Tec2 = 1.5s, ρec1 = ρec2 = 13 , To1 = To2 = 1s, ρo1 = ρec1 = ρec2 = 17 , 15 , 13 , 12 are presented as Fig. 4. The results
ρo2 = 13 , φi = 0.001, βod = 0.1, ao = 0.999, δod = 0.09, depict that the smaller the parameters ρec1 , ρec2 are, the faster
γo = 470, o = 0.65, ιo = 0.001. the convergence rate is.
The estimation of force and the trajectory adjustment are Comparing with the predefined-time tracking controller [13]
presented in Fig. 2. vF1 and vF2 represent the estimations with linear SM sec = e2 +λe1 and e2 (sec , Tec2 ) is redesigned
of τext1 and τext2 with τext = [τext1 ; τext2 ]. The variable as e2 (sec , Tec2 ) = ηTπec2 (sig(sec )1−η/2 + sig(sec )1+η/2 ) with
so = [so1 ; so2 ] enables representing the force estimation errors sig(sec )γ = [|sec1 |γ sign(sec1 ); |sec1 |γ sign(sec2 )], λ = 5, η =
vF − τext . According to Figure 2, it can be stated that the 1/3, Tc = 1.5s. The tracking comparison results between
π
proposed force observer provides arbitrary-time accurate force ė1 = − T1c ρ exp(e1 ρ ) ee11ρ and ė2 = − ηT c
(sig(e2 )1−η/2 +
estimation. Additionally, due to the effect of external force, sig(e2 ) 1+η/2 ) with ideal trajectory xd = 0.5 sin(0.4t) are
for the joint 1, the desired trajectory (blue solid line) will shown in Fig. 5 (a) and (b). It can be concluded that the same
be modified (green dotted line) to be compliant to the exter- convergence rate can be obtained under the same control force
nal force. The joint tracking performance after the trajectory by setting Tc = 0.5, ρ = 1/7, η = 1/2. However, linear SM
modification is illustrated in Fig. 3 with Tec1 = Tec2 = 1.5s is adopted in [13] to avoid the singularity problem. As shown
and Tec1 = Tec2 = 0.5s. It is evident that, in contact motion, in Fig. 5 (c) and (d), when the SM surface reaches zero, our
the trajectory tracking performance can be ensured in arbi- proposed nonlinear SM surface can provide faster convergence
trary time. The joint tracking error comparison results with speed.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on October 28,2023 at 12:00:37 UTC from IEEE Xplore. Restrictions apply.
YANG et al.: NOVEL INTERACTION CONTROLLER DESIGN FOR ROBOTIC MANIPULATORS 2155

V. C ONCLUSION [8] A. Alcocer, A. Robertsson, A. Valera, and R. Johansson, “Force estima-

tion and control in robot manipulators,” IFAC Proc. Vol., vol. 36, no. 17,
The force observer based arbitrary-time interaction control pp. 55–60, 2003.
law has been derived for the nonlinear robotic manipula- [9] K. S. Eom, I. H. Suh, W. K. Chung, and S.-R. Oh, “Disturbance observer
tor in this brief. An adaptive SM based arbitrary-time force based force control of robot manipulator without force sensor,” in Proc.
IEEE Int. Conf. Robot. Autom., 1998, pp. 3012–3017.
observer is proposed which enables estimating the robot- [10] G. Peng, C. Yang, W. He, and C. L. P. Chen, “Force sensorless admit-
environment contact force within preset time. Moreover, in tance control with neural learning for robots with actuator saturation,”
order to enhance the robot’s ability to learn and adapt IEEE Trans. Ind. Electron., vol. 67, no. 4, pp. 3138–3148, Apr. 2020.
[11] J. Zhai and G. Xu, “A novel non-singular terminal sliding mode tra-
to the external environment and ensure position, velocity jectory tracking control for robotic manipulators,” IEEE Trans. Circuits
tracking, the arbitrary-time admittance control method is for- Syst., II, Exp. Briefs, vol. 68, no. 1, pp. 391–395, Jan. 2021.
mulated. Finally, simulation results adopting the proposed [12] G.-H. Xu, F. Qi, Q. Lai, and H. H.-C. Iu, “Fixed time synchronization
control for bilateral teleoperation mobile manipulator with nonholo-
interaction controller confirm the theoretical findings and com- nomic constraint and time delay,” IEEE Trans. Circuits Syst., II, Exp.
parison results depict the superior performances. In the future, Briefs, vol. 67, no. 12, pp. 3452–3456, Dec. 2020.
the system uncertainties and external disturbances will be [13] A. Munoz-Vazquez, J. Sánchez-Torres, E. Jiménez-Rodriguez, and
A. G. Loukianov, “Predefined-time robust stabilization of robotic
considered. manipulators,” IEEE/ASME Trans. Mechatronics, vol. 24, no. 3,
pp. 1033–1040, Jun. 2019.
R EFERENCES [14] Z. Wang, B. Liang, Y. Sun, and T. Zhang, “Adaptive fault-tolerant
prescribed-time control for teleoperation systems with position error
[1] J. E. Colgate and N. Hogan, “Robust control of dynamically interacting constraints,” IEEE Trans. Ind. Inf., vol. 16, no. 7, pp. 4889–4899,
systems,” Int. J. Control, vol. 48, no. 1, pp. 65–88, 1988. Jul. 2020.
[2] M. T. Mason, “Compliance and force control for computer controlled [15] A. K. Pal, S. Kamal, S. K. Nagar, B. Bandyopadhyay, and L. Fridman,
manipulators,” IEEE Trans. Syst., Man, Cybern., vol. SMC-11, no. 6, “Design of controllers with arbitrary convergence time,” Automatica,
pp. 418–432, Jun. 1981. vol. 112, Feb. 2020, Art. no. 108710.
[3] S. P. Buerger and N. Hogan, “Complementary stability and loop shaping [16] B. Jiang and C.-C. Gao, “Decentralized adaptive sliding mode control
for improved human–robot interaction,” IEEE Trans. Robot., vol. 23, of large-scale semi-Markovian jump interconnected systems with dead-
no. 2, pp. 232–244, Apr. 2007. zone input,” IEEE Trans. Autom. Control, early access, Mar. 11, 2021,
[4] A. Q. L. Keemink, H. van der Kooij, and A. H. A. Stienen, “Admittance doi: 10.1109/TAC.2021.3065658.
control for physical human–robot interaction,” Int. J. Robot. Res., [17] S. P. Bhat and D. S. Bernstein, “Finite-time stability of continu-
vol. 37, no. 11, pp. 1421–1444, 2018. ous autonomous systems,” SIAM J. Control Optim., vol. 38, no. 3,
[5] I. Ranatunga, F. L. Lewis, D. O. Popa, and S. M. Tousif, “Adaptive pp. 751–766, 2000.
admittance control for human–robot interaction using model refer- [18] A. Polyakov, “Nonlinear feedback design for fixed-time stabilization
ence design and adaptive inverse filtering,” IEEE Trans. Control Syst. of linear control systems,” IEEE Trans. Autom. Control, vol. 57, no. 8,
Technol., vol. 25, no. 1, pp. 278–285, Jan. 2017. pp. 2106–2110, Aug. 2012.
[6] G. Peng, C. L. P. Chen, and C. Yang, “Neural networks enhanced optimal [19] E. Jiménez-Rodríguez, J. Sánchez-Torres, and A. G. Loukianov,
admittance control of robot-environment interaction using reinforcement “Predefined-time backstepping control for tracking a class of mechanical
learning,” IEEE Trans. Neural Netw. Learn. Syst., early access, Mar. 2, systems,” IFAC Papers Online, vol. 50, no. 1, pp. 1680–1685, 2017.
2021, doi: 10.1109/TNNLS.2021.3057958. [20] R. Seeber, H. Haimovich, M. Horn, L. M. Fridman, and H. De Battista,
[7] G. Peng, C. L. P. Chen, W. He, and C. Yang, “Neural-learning- “Robust exact differentiators with predefined convergence time,”
based force sensorless admittance control for robots with input dead- Automatica, vol. 134, Dec. 2021, Art. no. 109858.
zone,” IEEE Trans. Ind. Electron., vol. 68, no. 6, pp. 5184–5196, [21] C. Edwards and Y. B. Shtessel, “Adaptive continuous higher order sliding
Jun. 2021. mode control,” Automatica, vol. 65, pp. 183–190, Mar. 2016.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on October 28,2023 at 12:00:37 UTC from IEEE Xplore. Restrictions apply.