Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 70-6.&

/6.

#&3

'&#36"3:

*44/

$/
51

Control Theory and Technology

http://link.springer.com/journal/11768

Model reference adaptive impedance control for

physical human-robot interaction
Bakur ALQAUDI 1,2,3† , Hamidreza MODARES 1,2 , Isura RANATUNGA 1,2 , Shaikh M. TOUSIF 1,2 ,
Frank L. LEWIS 1,2,4 , Dan O. POPA 1,2
1.The University of Texas at Arlington Research Institute (UTARI), Fort Worth, TX 76118, U.S.A.;
2.Department of Electrical Engineering, College of Engineering, The University of Texas at Arlington, Arlington, TX 76019, U.S.A.;
3.Electronics and Instrumentation Engineering Technology Department, Yanbu Industrial College,
P.O. Box 30436, 41912, Yanbu al-Sinaiyah, K.S.A.;
4.State Key Laboratory of Synthetical Automation for Process Industries, Northeastern University, Shenyang Liaoning 110819, China
Received 18 December 2015; revised 7 January 2016; accepted 7 January 2016

Abstract
This paper presents a novel enhanced human-robot interaction system based on model reference adaptive control. The
presented method delivers guaranteed stability and task performance and has two control loops. A robot-specific inner loop,
which is a neuroadaptive controller, learns the robot dynamics online and makes the robot respond like a prescribed impedance
model. This loop uses no task information, including no prescribed trajectory. A task-specific outer loop takes into account the
human operator dynamics and adapts the prescribed robot impedance model so that the combined human-robot system has
desirable characteristics for task performance. This design is based on model reference adaptive control, but of a nonstandard
form. The net result is a controller with both adaptive impedance characteristics and assistive inputs that augment the human
operator to provide improved task performance of the human-robot team. Simulations verify the performance of the proposed
controller in a repetitive point-to-point motion task. Actual experimental implementations on a PR2 robot further corroborate the
effectiveness of the approach.

Keywords: Human-robot interaction, model reference adaptive control, model reference neuroadaptive, impedance control

DOI 10.1007/s11768-016-5138-2

† Corresponding author.

E-mail: [email protected].
The work was supported by the National Science Foundation (No. IIS-1208623), the Office of Naval Research grant (No. N00014-13-1-0562),
the AFOSR (Air Force Office of Scientific Research) EOARD (European Office of Aerospace Research and Development) grant (No. 13-3055),
the U.S. Army Research Office grant (No. W911NF-11-D-0001).
© 2016 South China University of Technology, Academy of Mathematics and Systems Science, CAS, and Springer-Verlag Berlin Heidelberg
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 69

1 Introduction on task requirements. In [13], adaptive controllers based

on neural networks were designed in which the error
Physical human-robot interaction (HRI) and coopera- dynamics parameters were tuned to become closer to a
tion has become significantly more important in recent prescribed error dynamics model.
years and is now of a major focus in robotics and con-
Most existing adaptive NN-based controllers and
trol society. The empirical evidence suggests that physi-
adaptive impedance controllers focus on tracking er-
cally embodied interactions are preferred by human op-
ror dynamics, and/or make the tracking error dynamics
erators over virtual or remote teleconference interac-
have a prescribed impedance characteristic. Moreover,
tions [1]. Unlike ordinary industrial robotics where the
the control torques derived in most work has been done
environment is structured and known, in HRI systems,
in the literature depends on the prescribed impedance
the robots interact with humans who have very different
model parameters. The objective of trajectory following
skills and capabilities. Therefore, it is of paramount im-
with an error dynamics having prescribed impedance
portance for robots to adjust themselves to the level of
properties often restricts the applications of these ap-
the skills and capability of the human and compensate
proaches in human-robot interactive systems.Modern
for possible human mistakes due to fatigue, stress, etc.
human-robotic interactive systems must be capable of
Control of industrial robots has often focused on fol- performing a wide range of tasks. Applications in indus-
lowing a desired trajectory in a well-known and struc- try, military, aerospace and the gaming industry focus
tured environment. For robot manipulators with un- on semi-autonomous features of robotic systems in in-
known nonlinear dynamics, modeling inaccuracies, and teracting with humans. This requires that task-specific
disturbances, nonlinear adaptive robot controllers have controls include the effects of both the robot dynamics
often been designed based on computed torque con- and the human dynamics, and their interactions. In this
trol [2] and/or feedback linearization [3,4] to yield guar- setting, trajectory following design for robot torque con-
anteed trajectory following. Adaptive control using neu- trollers is not suitable and limits system performance to
ral networks (NNs) has been successfully employed for a narrow range of tasks. In HRI systems, any trajectory
control of uncertain robot systems in the literature. tracking objectives cannot be implemented solely by the
These mentioned adaptive control methods, however, inner robot control loop, because the human dynamics
do not consider the interaction between the robot and must be included in task trajectory following objectives.
the environment or the human. When the robot is in This paper is motivated by the human factor studies,
contact with an object or a human, it must be able to and as opposed to most existing results does not de-
control not only positions, but also forces. sign a robot torque controller for trajectory following.
Impedance control has been widely studied in The purpose of this paper is (1) to avoid the need for
robotics as a control technique to perform robotic con- the human to learn robot-specific models, so he can
tact tasks. The purpose of impedance control is to pro- focus on the task and (2) to adapt the robot perfor-
vide stable tracking during robot contact with the ex- mance to assist the human-robot system in performing
ternal environment [5–9] by regulating the mechanical the task.The contributions of this paper are as follows.
impedance response of a robot to a desired reaction ac- An inner-loop torque controller is first designed to make
cording to a given task. In trajectory following, the im- the robot behave like a prescribed impedance model
portant feature is the tracking error dynamics. Therefore, from the human force input to the robot motion coordi-
impedance control in these applications has focused on nates. This means the human does not need to learn an
making the tracking error dynamics behave like a pre- inverse dynamics model to compensate for robot non-
scribed impedance model [9–15]. Adaptive impedance linearities and is a completely different philosophy than
control can be used to guarantee stable contact with un- making a trajectory error dynamics follow a prescribed
known environments and specify the desirable response impedance model [11]. Then, a task-specific outer-loop
of the robot to an external force profile. This can po- controller is designed, taking into account the human
tentially be used to regulate the interactions between a transfer characteristics, to tune the robot impedance
robot and a human operator while dynamically perform- model to assist the human in effectively performing the
ing a task. Various considerations have been taken into task. The outer-loop task-specific controller is designed
account to tune the impedance parameters. In [11], an to make the combined transfer function of the human
adaptive impedance feedforward term was used based and the robot resemble a desirable performance model
70 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

based on task requirements. Techniques from model ref- controller appears in Fig. 1 and is developed in Section 3.
erence adaptive control are modified to accommodate The objective in this loop is to design the controller
the fact that the tunable impedance model appears af- torque τ to make the error between the robot positon,
ter the plant, not before as in standard model-reference i.e., q, and the prescribed impedance model positon, i.e.,
adaptive control (MRAC). This task control loop incor- qm , go to zero. That is to design τ to make em = q−qm go
porates a human dynamics system identifier. Adaptive to zero. The input to both robot and impedance model
tuning algorithms are given for the robot impedance is the human torque τh . This is not the same as the bulk
model parameters and proofs of performance are for- of the work in robot impedance control [6] and neu-
mally presented. Novel extensions to MRAC are made ral network adaptive control [9–22] which is directed
in the design of both the robot-specific inner loop and towards making a robot follow a prescribed trajectory,
the task-specific outer loop controller design. and causing the trajectory error dynamics to follow a
This paper is organized as follows. Section 2 provides prescribed impedance model [11]. In our approach, no
an overview of the design philosophy in this paper. Sec- trajectory information and no information of the pre-
tion 3 designs a neural network adaptive torque con- scribed impedance model is needed for the inner loop
troller that makes a robot dynamics appear like a pre- design. This leaves the freedom to incorporate all task in-
scribed robot impedance model. This design is not based formation in an outer loop design. It will be seen that the
on trajectory following. In Section 4 an outer-loop con- robot torque input does not depend on the impedance
troller is designed using a novel MRAC structure that model parameters. This is in contrast to other adaptive
takes into account both the human dynamics model impedance control approaches which have a trajectory
and the prescribed robot impedance model to ensure following objective [11].
the effective performance of a task. Adaptive methods
are given for tuning the robot impedance model to as-
sist the human in the performance of the task. Section 5
gives simulation results and implementation results on
a PR2 robot are given in Section 6.

2 Structure of adaptive human-robot inter-

action Fig. 1 Inner-loop robot-specific model reference neuroadap-
tive control.
In this section, we preview the overall control ar-
chitecture developed in this paper. Two control loops An outer task-specific loop is next designed that con-
are designed. These control loops are motivated by hu- siders the human operator dynamics. All task perfor-
man factors studies [16–19] that show a human oper- mance details are relegated to this outer-loop design.
ator learns two components in performing tasks with a The task-specific outer loop design is shown in Fig. 2
robotic system. He learns a robot-specific inverse dy- and designed in Section 4. The objective is to tune the
namics model to compensate for the nonlinearities of robot impedance model, which is performed by design-
the robot. This appears to occur in the cerebellum, ing the control input u, as described later, to make the
where supervised learning is used to learn the envi- position of impedance model tracks the position of a
ronment [16]. Simultaneously, he learns a task-specific reference model, i.e., qr . It is a novel form of MRAC
feedback control component that is particular to the of a different sort than Fig. 1. The application of MRAC
successful performance of the task. Some recent work must be modified since the tunable parameter robot
in adaptive impedance control follows this approach of impedance model appears after the unknown human
robot-specific impedance control inner loop design fol- plant model, not before it as in standard MRAC design.
lowed by a task-specific outer loop design that includes Human-robot interactive systems can perform a variety
the human dynamics [20]. of quite general tasks. In this paper, we consider the task
In this paper, a robot-specific inner loop is first de- to be following a desired trajectory, as in point-to-point
signed to make the robot dynamics from the human op- motion control by a human operator [20, 21]. Then, the
erator input to the robot motion appear as a prescribed task reference input uc (t) in Fig. 2 is interpreted as the
robot impedance model. The robot-specific inner-loop desired task trajectory to be followed by the combined
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 71

man-robot system. The outer-loop design has two com- trajectory tracking.
ponents. An assistive input is generated that helps the The robot dynamics equation is adapted from [2]
human in task performance and the prescribed robot
impedance model in Fig. 1 is adapted to enhance the M(q)q̈ + V(q, q̇)q̇ + F(q̇) + G(q) + τd = τ + τh , (1)
human in task performance. This design must take into
account the unknown human dynamics as well as the where q ∈ Rn are the robot positions, M(q) is the inertia
desired overall dynamics of the human-robot system, matrix, V(q, q̇) is the Coriolis/centripetal forces, G(q) is
which depends on the task. the gravity vector, and F(q̇) is the friction term. The dis-
turbance is τd ∈ Rn and the human operator input is τh .
Control torque τ is to be designed to fulfil the control
objective outlined above and detailed below.
Equation (1) can be considered as being either in joint
space or Cartesian operational space. If it is in the joint
space, the inputs τd , τh are torques. If it is in Cartesian
Fig. 2 Outer-loop task-specific MRAC for adaptive human- space, the inputs τd , τh are forces. Forces f and torques
robot interaction. τ are related by τ = JT f where J is the robot Jacobian
matrix. The Cartesian inertia, Coriolis/centripetal forces,
friction and gravity terms are likewise determined from
their joint space counterparts by using the Jacobian ma-
3 Inner-loop control design
trix, according to standard techniques [2].
In this section, the inner-loop torque controller for the Select the prescribed robot impedance model whose
robot manipulator shown in Fig. 1 is derived to make the dynamics are to be followed by the robot as
robot dynamics from human operator input to robot mo-
tion appear like a prescribed robot impedance model. Mm q̈m + Dm q̇m + Km qm = τh , (2)
A neural network approximator is used to compensate
for the unknown nonlinear robot dynamics. We call this where qm (t) is the model trajectory, Mm is the desired
approach neuroadaptive control. The detailed result of mass matrix, Dm is the desired damping matrix, and Km
this design is shown in Fig. 3. No task trajectory infor- is the desired spring constant matrix. The impedance pa-
mation is needed in this design, so that this work is rameters Mm , Dm , and Km will be designed in Section 4
different from most existing work in robot control and in an outer task-specific loop that takes into account
neural network control [23]. both the human operator dynamics and the task objec-
tives.
Robot-loop control design objective Design a
robot torque controller that makes the robot dynamics
(1) from the human input τh to the manipulator motion
q(t) behave like the prescribed impedance model (2).
To this end, define the model-following error

em = q − q m , (3)

and the sliding mode error

Fig. 3 Model reference neuroadaptive controller. r = ėm + Λem , (4)

where Λ is a symmetric, positive definite design param-

3.1 Robot impedance model and model-following
eter matrix. Since (4) is a stable system, considering r(t)
error dynamics
as input and em (t) as output, the control torque τ in (1)
In this section, we formulate a novel control objective is now designed to guarantee that r(t) is bounded. This
for an inner-loop robot controller that does not involve guarantees bounded model-following error em (t).
72 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

Using (1), (3), and (4) the dynamics of the sliding positive scalar ZB according to
mode error are given by ⎡ ⎤
⎢⎢ W 0 ⎥⎥
ZF = ⎢⎢⎢⎣ ⎥⎥  Z
⎥ (9)
 0 V ⎦
B
M(q)(q̈m − (ṙ − Λėm )) + V(q, q̇)(q̇m − (r − Λem )) F
+F(q̇) + G(q) + τd = τ + τh , (5)
with  · F the Frobenius norm. Define matrix Ẑ com-
mensurately with the definition of Z.
which yields the sliding error dynamics
To make the model-following error defined in (3) sta-
ble and consequently make the robot dynamics (1) be-
M(q)ṙ = −V(q, q̇)r + f (x) + τd − τ − τh , (6)
have like the prescribed impedance model, the control
torque is designed as
where
τ = Ŵ T σ(V T x) + Kv r − v − τh , (10)
f (x) = M(q)(q̈m + Λėm )
+V(q, q̇)(q̇m + Λem ) + F(q̇) + G(q) (7) where Kv r is a proportional-plus-derivative loop with
Kv = KvT a gain matrix, and
is a nonlinear function of robot parameters which is as- fˆ(x) = Ŵ T σ(V T x) (11)
sumed unknown. It is important to note that f (x) does
not depend on the impedance model parameters Mm , is the NN approximation for the unknown function f (x),
Dm , and Km in (2). This is in contrast to impedance and
control robot controllers that have a trajectory following
objective [11] where a tracking error is used instead of v(t) = −Kz (ẐF + ZB )r (12)
the model-following error (3).
with Kz > 0 a scalar gain is a robustifying signal that
3.2 Neuroadaptive model-following controller compensates for unmodeled and unstructured distur-
bances.
In this section, a control structure is given which It is shown in Theorem 1 how to tune the NN weights
uses a neural network (NN) to approximate the un- V̂ and Ŵ such that the control torque in (10) makes the
known function f (x) in (7) and guarantees the stability model-following error (3) bounded and consequently
of the model-following error (3). Therefore, the robot the robot dynamics (1) from human input τh to the out-
dynamics (1) with human input τh appears as the pre- put q(t) behaves like the prescribed robot impedance
scribed impedance model (2). We call this a neuroad- model (2).
aptive model-following controller. The use of NN in Remark 1 The structure of the robot controller de-
robot control is a standard approach used by many prior signed here is given in Fig. 3. It is important to note
works [23]. In contrast to almost all thesestandard ap- that this controller guarantees model-following behav-
proaches, there is no trajectory-following objective here, ior of the robot dynamics (1) given the prescribed robot
so that a desired reference trajectory is not needed by impedance model (2), based on the model-following
the neuroadaptive controller. error (3). There is no objective for tracking a desired
To provide an approximation for the unknown func- trajectory. This is in contrast to almost all existing work
tion f (x) in (7), a neural network (NN) is introduced. in robot control [23]. Second, the impedance model pa-
According to the NN approximation property [24–30] rameters Mm , Dm , and Km do not appear in the control
the nonlinear function in (7) can be approximated by law (10) or in the function f (x) in (7), so that the NN
does not need to identify the already-known impedance
f (x) = W T σ(V T x) + ε, (8) model parameters. This is reflected in Fig. 3, where the
prescribed impedance model (2) does not appear. This
where W and V are unknown ideal NN weights and is contrast to the work on adaptive impedance control
σ( · ) is a vector of activation functions. The NN input based on a trajectory tracking error dynamics [11]. As
vector is x = [eTm ėTm qTm q̇Tm q̈Tm ]T . It is known that the a result, the approach given here cleanly decouples the
NN approximation error ε is bounded on a compact set. robot-specific control design given here from the task-
Assume the ideal weights are bounded by a constant specific control design which is given in the next section.
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 73

This is in keeping with human factor studies [17] which Parks [32]. In [33–38], general approaches to MRAC de-
indicate that the human learns two control components sign and its applications were developed. Seminal work
in task performance, one to compensate for nonlinear was done by [37], and others.
robot dynamics and one to assure task performance. The objective in this section is to design the human-
robot interaction task-specific controller in Fig. 2 that
4 Outer-loop model reference adaptive takes into account the human dynamics, which are un-
HRI controller known, and the task objectives. The detailed result is
in Fig. 4. It will be seen that this task-loop controller
In this paper, the task-specific outer loop controller is performs two functions. It adapts the parameters of the
designed using extensions of model-reference adaptive robot impedance model (2) so that the task performance
control. The pioneering research work for model ref- of the human-robot system is improved, and also pro-
erence adaptive control (MRAC) was carried on during vides assistive inputs that enhance the human’s task
the 1960s by H. P. Whitaker, P. V. Osburn and A. Keze. performance. No robot-specific information is needed
Initial work in MRAC depended on gradient descent al- in the task loop design presented in this section. This
gorithms, including the MIT rule [31]. More rigorous decoupling of control objectives goes along with human
Lyapunov designs for MRAC were proposed by P. C. factors studies in [17].

Fig. 4 Overall system of model reference adaptive control.

4.1 Model reference adaptive control (MRAC) for- human-robot interactive task performance, the human
mulation of adaptive HRI adapts itself to compensate for robot dynamics nonlin-
earities and also learns task-specific controls. However,
The problem of adapting the robot impedance model
after learning, it has been observed that the expert op-
in Fig. 2 to assist the human in performing a task is
erator exhibits the transfer characteristics of a simple
now formulated as a nonstandard MRAC problem. The
linear model with a time delay. For many tasks, this hu-
challenge to be overcome is that the tunable parameter
man operator model is a first-order linear system of the
compensator is the prescribed robot impedance model
form [21].
in Figs. 2 and 4, which occurs after the unknown plant
(the human dynamics), not before, as in standard MRAC. τh = B(sI + A)−1 uc . (13)
This problem is overcome by adding a system identifier
for the human dynamics. In this section, it is assumed that the human transfer
The prescribed robot impedance model (2) has matrices A and B are unknown. It is observed in human
qm (t) ∈ Rn , with n the number of degrees of freedom task studies there is a reaction time delay τ that is in-
of the robot. It is assumed here that the robot dynam- dependent of the particular operator once the task has
ics (1) are in Cartesian task space, so that for n = 6 been learned, and is almost constant at 0.4 s [20, 21].
degrees of freedom, the vector q(t) has three position Therefore, it can be compensated for, so that, without
components and three angular rotation components [2]. loss of generality, the delay τ can be taken as zero in
Regarding the human transfer characteristic, it is known (13) by shifting the measured time signals.
from human factors neurocognitive studies [17] that, in Regarding the task reference model in Fig. 2, it is fur-
74 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

ther observed in human-robot interactive task learning Based on the above and referring to Figs. 2 and 4,
studies that the human operator adapts to make the consider the dynamics for the human, nominal robot
overall transfer characteristic of the human-robot sys- impedance model, and task reference model, given re-
tem appear as a simple linear first-order system with spectively by
high bandwidth. This is known as the crossover model.
Specifically [21], the skilled operator in a man-machine τ̇h = −Aτh + Buc , (16)
system adapts his own dynamics to make the total sys- q̇m = −An qm + Bn u, (17)
tem transfer characteristic of the human-plus-robot re-
q̇r = −Am qr + Bm uc . (18)
main unchanged over wide variations in the robot dy-
namics. The total man-robot transfer characteristic is
therefore prescribed here as the task reference model Here, the prescribed task trajectory is uc (t) and an MRAC
control law is to be designed for the control input u(t)
qr = Bm (sI + Am )−1 uc (14) in (17).
4.2 Adaptive impedance control and human-
with prescribed matrices Am and Bm . These parameters
assistive inputs using Lyapunov design
are selected based on the specific task.
The class of tasks depicted in Fig. 2 includes trajec- Given this setup, the basic concept of model refer-
tory-following tasks where the human operates the ence adaptive control (MRAC) [33–37] can be used in
robot to follow a prescribed trajectory. This includes this section to confront the design of the task loop of
point-to-point motion tasks in force fields as studied Fig. 2. The dynamics for the human, robot impedance
in [20, 21]. This class of tasks can be considered as hav- model, and task reference model, given respectively by
ing a model-following objective based on the overall (16)–(18).
task reference model (14), and given the unknown hu- Unfortunately, applying MRAC to this problem is
man dynamics (13) and the robot response detailed by complicated by the fact that in standard MRAC, the tun-
the robot impedance model (2). able controller appears before the unknown plant dy-
Remark 2 It is noted that if the task is trajectory fol- namics and provides its control input so that the plant
lowing by the man-machine system, the parameters of has the transfer characteristics of the reference model.
the task reference model (14) should be selected so that By contrast, in adaptive impedance control for human-
Am = Bm . This has low-frequency gain of 1, so that the robot interaction (Fig. 2), the tunable impedance model
trajectory is followed with zero steady-state error. Ma- occurs after the unknown human dynamics. This causes
trix Am should be selected based on desired transient some complications and requires the introduction of a
response characteristics of the man-machine system. system identifier for the human dynamics. The overall
This choice of task reference model does not restrict the setup for adaptive HRI using MRAC approach is given
objective to following constant trajectories. If the trajec- in Fig. 4. The approach given here provides a formal
tory is time varying, suitable choice of the time-constant model-following stability proof using Lyapunov tech-
matrix A−1m will still result in good trajectory following.
niques, and formalizes the human dynamics identifier
approach used in [21].
In Section 3, it was assumed that the prescribed robot
Task-loop control design objective Design an
impedance model (2) is of second order. However, the
MRAC for control input u(t) so that the combined
model does not appear in the control design given in
human-robot transfer function is equal to the prescribed
Theorem 1. Only the model motion trajectory qm , q̇m ,
task reference model (18). See Fig. 2.
and q̈m is needed in the design of the robot-specific con-
troller there. Therefore, in this section we take a nominal It will be seen that the MRAC for u(t) has two
prescribed robot impedance model as components. One component tunes the parameters
of the robot impedance model (17). Then, the robot
qm = Bn (sI + An )−1 u, (15) impedance model (17) provides the model reference
trajectory qm , q̇m , and q̈m used in the inner-loop torque
where An and Bn are initial nominal matrices. It is shown controller of Fig. 1 and Theorem 1, through the slid-
in the following how the overall prescribed impedance ing mode error (4) and the NN input vector x =
model will be changed and tuned by MRAC design to [eTm ėTm qTm q̇Tm q̈Tm ]T . The second component of the
assist the human to perform a task. MRAC provides assistive inputs that augment the oper-
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 75

ator’s output τh (t) to enhance his task performance. See V̂˙ = Gx(σ̂T Wr)T − κGrV̂, (26)
comments at the end of Theorem 1.
The human transfer function (16) is unknown. There- where F and G are symmetric positive definite matri-
fore, a system identifier is introduced as ces and κ > 0 is a small design parameter. Tune the
outer-loop control parameters in (21) according to
τ̂˙ h = −Âτh + B̂uc (19)
θ̇1 = γ1 B−1
n Pm euc ,
T

for the human response. Define the human response es- θ̇2 = γ2 B−1
n Pm eqm ,
T
timation error τ̃h = τh − τ̂h . Then, the estimation error
θ̇3 = γ3 Pm BTn eτ̂Th + γ1 ÂB−1
n Pm euc
T
dynamics becomes
−γ4 θ1 Ph τ̃Th τ̂Th , (27)
τ̃˙ h = τ̇h − τ̂˙ h = −Aτh + Buc + Âτh − B̂uc θ̇4 = γ1 (B̂ + B−1
n )Pm euc
T

= −Aτ̃h + Ãτ̂h − B̃uc , (20)

with Pm > 0 and Ph > 0, and the parameters in the
where the identifier parameter errors are à =  − A, human system identifier (19) according to
and B̃ = B̂ − B. Now, consider the control law
Â˙ = −γ4 Ph τ̃h τ̂Th , B̂˙ = γ5 Ph τ̃Th uTc . (28)
u = −θ1 τ̂˙ h − θ2 qr − θ3 τ̂h − θ4 uc , (21)
Then, the inner-loop model-following error em (t), the
where θ1 , θ2 , θ3 and θ4 are tunable matrices of appro- outer-loop model following error e(t) and the human re-
priate dimension. Then, the overall system is illustrated sponse estimation error are bounded, so that the prod-
in Fig. 4. To derive tuning laws for the parameters θ1 , uct of the human dynamics and robot dynamics follows
θ2 , θ3 , θ4 , Â and B̂ such that the control objective is the task reference model (18).
achieved, define the model-following output error as
Proof Define a Lyapunov function as
e = qm − qr . (22)
1 1
L = rT M(q)r+tr(W̃ T F−1 W̃)+tr(Ṽ T F−1 Ṽ)+ eT Pm e
Then 2 2
1
+ tr([Bn θ2 +An −Am ] [Bn θ2 +An −Am ])
T
ė = q̇m − q̇r 2γ2
= −An qm + Bn u + Am qr − Bm uc . (23) 1
+ tr([Bm +Bn θ1 B̂+Bn θ4 ]T [Bm +Bn θ1 B̂+Bn θ4 ])
2γ1
Substituting the control law (21) into this equation and 1 1
manipulating yields the model-following error dynamics + tr([θ1 Â−θ3 ]T [θ1 Â−θ3 ])+ τ̃Th Ph τ̃Th
2γ3 2
as 1 1
+ tr(ÃT Ã)+ tr(B̃T B̃), (29)
2γ4 2γ5
ė = − Am e − [Bn θ2 + An − Am ]qm
− [Bm + Bn θ1 B̂ + Bn θ4 ]uc + Bn [θ1 Â − θ3 ]τ̂h . (24) where the weight estimation errors are W̃ = W − Ŵ,
Ṽ = V − V̂. Differentiating this Lyapunov function and
The next result provides tuning laws for the control using (24) yields
parameters in (21), the human dynamics identifier (19)
and the neural network weights for the inner-loop con- 1 ˙ )+tr(Ṽ T F−1 Ṽ˙ )
troller in (10) that make the overall human-robot system L̇ =rT M(q)ṙ+ rT Ṁ(q)r+tr(W̃ T F−1 W̃
2
behave like prescribed reference model (14). 1
+ eT {− (ATm Pm +Pm Am )e−Pm [Bn θ2 +An −Am ]qm
Theorem 1 Consider the prescribed impedance 2
model (2), and the robot dynamics (1) with control − Pm [Bm +Bn θ1 B̂+Bn θ4 ]uc +Pm Bn [θ1 Â−θ3 ]τ̂h }
input (10) for the inner-loop controller. Consider the 1
+ tr([Bn θ2 + An − Am ]T Bn θ̇2 )
unknown human dynamics (16), the robot impedance γ2
model (17), and the outer-loop control input (21). Tune 1
+ tr([Bm +Bn θ1 B̂ + Bn θ4 ]T (B̂Bn θ̇1 + Bn θ̇4 ))
the NN weights in the inner-loop controller (10) as γ1
1
˙ = Fσ̂rT − Fσ̂ V T xrT − κFrŴ, + tr([θ1 Â − θ3 ]T [Aθ̇1 + θ1 Â˙ − θ̇3 ])
Ŵ (25) γ3
76 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

1 1 tr(ÃT Ph τ̃h τ̃Th + B̃T Ph uc τ̃Th ) and using the tuning rules for
+ τ̃Th Ph τ̃˙ h + tr(ÃT Ã˙ ) + tr(B̃T B̃˙ ). (30)
γ4 γ5 the inner–and outer–loop controllers gives

Since −Am is Hurwitz, there exists a Qm > 0 such that 1 1

L̇ = −rT Kv r − eT Qm e − τ̃h Qh τ̃Th
2 2
ATm Pm + Pm Am = Qm . +krtr(W̃ T (W − W̃))
+krtr(Ṽ T (V − Ṽ)) + rT (w + v)
The robot manipulator dynamics (1) is assumed to be
unknown and therefore the function f in (7) is unknown = −rT Kv r + krtr(Z̃T (Z − Z̃)) + rT (w + v). (35)
and approximated online by (11). Then, the closed-loop
Using the following equality in (35),
filtered error dynamics (6) becomes
tr(Z̃T (Z − Z̃)) =< Z̃, Z > −Z̃2F  Z̃F ZF − Z̃2F .
M(q)ṙ = −V(q, q̇)r + Ŵ T φ(V T z) + τd − τ − τh + f˜,
(31) One has

where f˜(x) = f (x) − fˆ(x) is the estimation error. Substi- L̇  −rT Kv r − eT Am e − τ̃Th aτ̃h + kr · Z̃F (ZB − Z̃F )
tuting τ form (10) in (31) gives −KZ (ẐF + ZB )r2 + r · w
 Kvmin r2 + kr · Z̃F (ZB − Z̃F )
M(q)ṙ = −V(q, q̇)r − Kv r + τd + f˜(x) + v(t). (32)
−KZ (ẐF + ZB )r2
On the other hand, since e {Pm [Bn θ2 + Bn − Bm ]qm −
T
+r[C0 + C1 Z̃F + C2 r · Z̃F ]
Pm [Bm + Bn θ1 B̂ + Bn θ4 ]uc + Pm Bn [θ1 Â − θ3 ]τ̂h } is scalar,  −r{Kvmin r − k · Z̃F (ZB − Z̃F )
one has −C0 − C1 Z̃F }, (36)
eT {Pm [Bn θ2 + An − Am ]qm − Pm [Bm + Bn θ1 B̂ where Kvmin is minimum singular value of Kv in the last
+ Bn θ4 ]uc + Pm Bn [θ1 Â − θ3 ]τ̂h } inequality. L̇ is negative as long as the term in braces is
= tr([Bn θ2 + An − Am ]T Pm eqTm positive. Defining C3 = ZB + C1 /k and completing the
− [Bm + Bn θ1 B̂ + Bn θ4 ]T Pm euTc square yields
+ [θ1 Â − θ3 ]T Pm BTn eτ̂Th ). (33) Kvmin r − k · Z̃F (ZB − Z̃F ) − C0 − C1 Z̃F
= k(Z̃F − C3 /2)2 + Kvmin r − C0 − kC23 /4,
Using (25), (26), (32) and (33) into (30) gives

1 which is guaranteed positive as long as either r >

L̇ = −rT Kv r + rT (Ṁ(q) − 2V(q, q̇))r C0 + kC23 /4 C0 C3
2
2 C3
˙ + σ̂rT − σ V̂ T xrT )) or Z̃ > + + . This completes
+tr(W̃ T (F−1 W̃ Kvmin 2 k 4
1 the proof. 
+tr(Ṽ T (G−1 Ṽ˙ + xrT Ŵ T σ̂ )) − eT Qm e
2 This result provides a method for tuning the robot
−tr([Bn θ2 + An − Am ]T Pm eqTm ) − tr([Bm + Bn θ1 B̂ impedance model (17) to provide a desired model ref-
+Bn θ4 ]T Pm euTc ) + tr([θ1 Â − θ3 ]T Pm BTn eτ̂Th ) erence output qr (t) such that the human (16) plus robot
1 impedance model follows the prescribed task reference
+ tr([Bn θ2 + An − Am ]T Bn θ̇2 )
γ2 model (18). This output is sent as qm (t) to the inner
1 robot control loop in Fig. 3 to compute the inner-loop
+ tr([Bm + Bn θ1 B̂ − Bn θ̇4 ]T (B̂Bn θ̇1 + Bn θ̇4 ))
γ1 model following error (3). The human input in Fig. 3 and
1 in (10) is τh (t). These relationships are shown in Figs. 2
+ tr([θ1 Â − θ3 ]T [Âθ̇1 + θ1 Â˙ − θ̇3 ])
γ3 and 4.
1 It is interesting to examine the operation of the con-
+τ̃Th [− (AT Ph + Ph A)τ̃h + Ph Ãτ̂h − Ph B̃uc ]
2 trol input (21). After convergence of the human system
1 1 identifier, one has τ̂h (t) = τh (t). Then, the closed-loop
+ tr(ÃT Ã˙ ) + tr(B̃T B̃˙ ). (34)
γ4 γ5 robot impedance model is
Noting that since −A is stable there exists a Qh > 0 Bn Bn
such that AT Ph + Ph A = Qh , using τ̃Th [Ph Ãτ̂h − Ph B̃uc ] = qr = u= (−θ1 τ̂˙ h − θ2 qr − θ3 τ̂h − θ4 uc )
s + An s + An
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 77

−Bn system identifier model (19) is designed to adaptively

= [(θ1 s + θ3 )τh + θ4 uc ]
s + (An + Bn θ2 ) identify the human in the loop.
−Bn
≡ τ̄h , (37) The performance of the outer task loop MRAC in Fig. 4
s + (An + Bn θ2 ) is shown in Figs. 5–7. A square wave is selected for the
task reference input uc (t). This is interpreted as a point-
where τ̄h can be considered as a modified human force
to-point motion task where the human-robot system is
defined, according to (13), by
required to cycle from one point to another point repet-
B(θ1 s + θ3 ) itively. Fig. 5 shows the output of the robot impedance
τ̄h = [ + θ4 ]uc . (38) model qm (t) and the output qr (t) of the task reference
s+A
model. It is seen that qm (t) closely follows qr (t), with
Therefore, control parameter θ2 modifies the robot performance improving after several cycles. This shows
impedance model time constant, whereas control pa- the adaptive improvement of the controller as the robot
rameters θ1 , θ3 , and θ4 provide a proportional-plus- impedance model is tuned and the assistive inputs to the
derivative controller that augments the human force sig- human are learned. See discussion after Theorem 1. The
nal τh . This can be viewed as an assistive term that aids effectiveness of the human system identifier is revealed
the human so that task performance is improved. In in Fig. 6, which shows the output of the human trans-
fact, it is observed in [21] that the expert human op- fer function τh (t) and the output of the human identifier
erator, after learning to accomplish a task, incorporates τ̂h (t), which follows τh (t) more closely with each motion
a PD controller that seems to come from a task model cycle. The convergence of the human identifier parame-
learned in the cerebellum [17]. ters to the actual human model parameters is shown in
Fig. 7.

5 Simulation
In this section, the results from simulating the pro-
posed controllers on a 2-link robotic arm in MATLAB are
presented. The 2-link robot arm is a revolute-revolute
planar arm described in [2, Example 3.2-2]. First are
shown the simulation results for the outer-loop con-
troller in Fig. 2 and Fig. 4 that adapts the parameters of
the prescribed robot impedance model. Next are shown Fig. 5 Robot impedance model qm (t) output and prescribed
task reference output qr (t).
the simulation results for the inner-loop model reference
neuroadaptive controller in Figs. 1 and 3.

5.1 Outer-loop simulation

This simulation is for the outer task loop shown in
Figs. 2 and 4. In this simulation the prescribed robot
impedance model (2) is chosen to have qm (t) ∈ R2 ,
with n = 2 the number of degrees of freedom of the
robot. It is assumed here that the robot dynamics (1) are Fig. 6 Human output τh (t) and human identifier output τ̂h (t).
in Cartesian task space. The nominal robot impedance
model (17) is chosen for each degree of freedom as
q̇m = −3qm + 3u. These nominal time constant and gain
parameters are modified through the action of the adap-
tive control (21). See the discussion after Theorem 1. The
task reference model (18) is taken as q̇r = −12qr + 12uc ,
where uc is the desired trajectory to be reached in a
point-to-point motion task. The unknown human dy-
namics model (16) is chosen as τ̇h = −1τh + 0.5uc . The Fig. 7 Parameter convergence of adaptive human identifier
human dynamics model is unknown, and the human model.
78 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

5.2 Inner-loop simulation with 10 inputs, including a constant bias input, 20 hid-
den layer neurons and 2 outputs. The sigmoid function
This simulation is for the inner robot control loop of 1
Figs. 1 and 3. The outer-loop design just described gen- σ(x) = was used for the activation functions.
1 + e−x
erates the human operator signal τh (t) and the robot The weights Ŵ and V̂ of the network were randomly
impedance model trajectory qm (t). Two parallel outer initialized.
loops were used, one for each joint of the 2-link robot The simulation results for both links are shown in
arm simulated here.
Fig. 8, where q1d (t), q2d (t) denote the 2 components of
The robot dynamics (1) used for this simulation was
the task trajectory uc (t). It is observed that, after a short
the 2-link revolute-revolute planar robot arm described
transient learning period of a few cycles of the square
in [2, Example 3.2-2]. The arm parameters are selected
wave task trajectory, the motion qm (t) generated by the
as m1 = 0.8 kg, m2 = 2.3 kg, l1 = 1 m, l2 = 1 m and
g = 9.8 m/s2 . The controller parameters used in Theo- robot impedance model and the robot motion q(t) are
rem 1 were Kv = I2 , Λ = 5I2 , F = 100I2 , G = 20I2 , identical. This verifies the performance of the model
κ = 0.07, Kz = 5, and ZB = 100, where I2 is the 2 × 2 reference neuroadaptive controller in making the robot
identity matrix. A two-layer neural network was used arm behave like the robot impedance model.

Fig. 8 Inner-loop simulation. (a) Joint angles for Joint 1. (b) Joint angles for Joint 2. (c) Joint velocity for Joint 1. (d) Joint velocity
for Joint 2.

Overall performance of the proposed controller It perimental layout and Fig. 10 shows the PR2 robot. The
can be seen from the simulation results that the two con- controller was implemented in real-time using the real-
trollers, inner robot loop and outer task loop, achieve time controller manager framework of the PR2 in ROS
the objectives of the design. The outer loop assists the Groovy. The real-time loop on the PR2 runs at 1000 Hz
human in achieving the task by providing two assistive and communicates with the sensors and actuators on
components and tuning the robot impedance model. an EtherCAT network. Human force is measured using
The robot specific inner-loop controller compensates an ATI Mini40 FT Sensor attached between the gripper
for the robot nonlinearities and makes the robot behave and forearm of the PR2.
like this robot impedance model. The experiment involves the seven degree-of-freedom
arm of the PR2 robot in a point-to-point motion (PTP)
task. PTP manipulation is an increasingly popular task,
6 Experimental case study
both in the game industry and in industrial applications.
In this section, a case study of a practical experi- In this experiment a human applies a force on the
ment to evaluate the controllers of the Human-Robot right arm of the PR2 to follow the PTP motion trajec-
interaction system is presented. The experiments were tory, as shown in Fig. 9. The experiment is setup with
conducted at the University of Texas at Arlington Re- a human operator and the PR2 arm across from each
search Institute on a PR2 robot. Fig. 9 shows the ex- other as seen in Fig. 10. The human operator was then
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 79

asked to hold the gripper of the PR2 to perform PTP mo- is shown (with a dotted line) followed by the robot
tion between point A and point B along the y axis. The impedance model output (with a dash-dot line) and
human is assumed to be working in open-loop without the real robot output (with a solid line). It is seen that
considering the visual feedback of the current location the inner-loop neuroadaptive controller makes the robot
and the target location of the gripper. The desired target (with a dashed line) follow the robot admittance model
location to be reached is switched every 5 seconds. output (with a dash-dot line), and the outer-loop MRAC
makes the human-robot interactive team follow the pre-
scribed task model. This is accomplished after a short
transient learning time where the adaptation mechanism
tunes the whole system in the first 6 seconds. There is
a small time delay of 0.4 s due to the human reaction
time.

Fig. 9 Experiment layout.

Fig. 11 Simulation of the human-robot interactive system.

7 Conclusions
This paper presents a novel method of enhancing
human-robot interaction based on model reference
adaptive control. The method presented delivers guaran-
teed stability and task performance and has two control
loops. A robot-specific inner loop is a model reference
neuroadaptive controller that learns the robot dynamics
online and makes the robot responds like a prescribed
impedance model. This loop uses no task information,
including no prescribed trajectory. A task-specific outer
loop takes into account the human operator dynamics
and adapts the prescribed robot impedance model so
that the combined human-robot system has desirable
Fig. 10 PR2 robot at UTARI. characteristics for task performance. This design is also
The controller parameters used were Kv = 5I6 , Λ = based on model reference adaptive control, but of a non-
20I6 , F = 100I6 , G = 200I6 , κ = 0.3, Kz = 0.001, and standard form. The net result is a controller with both
ZB = 100, where I6 is the 6 × 6 identity matrix. A two- adaptive impedance characteristics and assistive inputs
layer neural network was used with 35 inputs, including that augment the human operator to provide improved
task performance of the human-robot team. Simulations
the bias input, 10 hidden layer neurons, and 7 outputs.
1 verify the performance of the proposed controller in a
The sigmoid function σ(x) = was used for the repetitive point-to-point motion task. Actual experimen-
1 + e−x
activation functions. The weights Ŵ and V̂ of the net- tal implementations on a PR2 robot further corroborate
work were randomly initialized. the effectiveness of the approach.
The result of the whole human-robot interaction sys-
tem is shown in Fig. 11. The task trajectory (with a References
dashed line) gives the target point locations, which cy- [1] J. Wainer, D. J. Feil-Seifer, D. A. Shell, et al. The role of
cle every 5 seconds. The task reference model output physical embodiment in human-robot interaction. The 15th
80 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

IEEE International Symposium on Robot and Human Interactive [20] S. Suzuki, K. Kurihara, K. Furuta, et al. Variable dynamic assist
Communication, Hatfield: IEEE, 2006: 117 – 122. control on haptic system for human adaptive mechatronics.
[2] F. L. Lewis, D. Dawson, M. Abdallah, et al. Robot Manipulator Proceedings of the 44th IEEE Conference on Decision Control/
Control: Theory and Practice. Boca Raton: CRC Press, 2003. European Control Conference, New York: IEEE, 2005: 4596 –
4600.
[3] J. J. E. Slotine, W. Li. Applied Nonlinear Control, Englewood Cliffs:
Prentice Hall, 1991. [21] S. Suzuki, K. Furuta. Adaptive impedance control to enhance
human skill on a haptic interface system. Journal of Control
[4] M. Jamshidi, B. J. Oh, H. Seraji. Two adaptive control structures
Science and Engineering, 2012: DOI 10.1155/2012/365067.
of robot manipulators. Journal of Intelligent and Robotic Systems,
1992, 6(2/3): 203 – 218. [22] F. C. Chen, H. K. Khalil. Adaptive control of nonlinear systems
using neural networks. International Journal Control, 1992, 55(6):
[5] N. Hogan. Impedance control: an approach to manipulation.
1299 – 1317.
Proceedings of the American Control Conference, San Diego: IEEE,
1984: 304 – 313. [23] F. L. Lewis, S. Jagannathan, A. Yesildirek. Neural Network Control
of Robot Manipulators and Nonlinear Systems. London: Taylor
[6] R. Anderson, M. W. Spong. Hybrid impedance control of robotic
and Francis, 1992.
manipulators. Journal Robot Automation, 1988, 4(5): 549 – 556.
[24] S. S. Ge, T. H. Lee, C. J. Harris. Adaptive Neural Network Control
[7] H. Kawasaki, R. Taniuchi. Adaptive control for robotic
of Robotic Manipulators. Singapore: World Scientific, 1998.
manipulators executing multilateral constrained task. Asian
Journal of Control, 2003, 5(1): 1 – 11. [25] G. A. Christodoulou, M. A. Rovithakis. Adaptive control of
unknown plants using dynamical neural networks. IEEE Trans-
[8] H. Wu, W. Xu, C. Cai. Adaptive impedance control in robotic cell
actions on Systems, Man, and Cybernetics – Part A: Systems and
injection system. The 17th International Conference on Methods
Humans, 1994, 24(3): 400 – 412.
and Models in Automation and Robotics, Miedzyzdrojie: IEEE,
2012: 268 – 275. [26] A. Yeşildirek, F. L. Lewis. Feedback linearization using neural
networks. Automatica, 1995, 31(11): 1659 – 1664.
[9] S. S. Ge, C. C. Hang, T. H. Lee, et al. Stable Adaptive Neural
Network Control. Boston: Kluwer Academic. 2001. [27] M. M. Polycarpou. Stable adaptive neural control scheme for
nonlinear systems. IEEE Transactions on Automatic Control, 1996,
[10] M. K. Vukobratovic, A. G. Rodic, Y. Ekalo. Impedance control
14(3): 447 – 451.
as a particular case of the unified approach to the control of
robots interacting with a dynamic known environment. Journal [28] A. S. Poznyak, W. Yu, E. N. Sanchez, et al. Nonlinear
of Intelligent & Robotic Systems, 1992, 18(2): 191 – 204. adaptive trajectory tracking using dynamic neural networks. IEEE
Transactions on Neural Networks, 1996, 10(6): 1402 – 1411.
[11] E. Gribovskaya, A. Kheddar, A. Billard. Motion learning and
adaptive impedance for robot control during physical interaction [29] G. A. Rovithakis. Performance of a neural adaptive tracking
with humans. IEEE International Conference on Robotics and controller for multi-input nonlinear dynamical systems. IEEE
Automation, Shanghai: IEEE, 2011: 4326 – 4332. Transactions on Systems, Man, Cybernetics – Part A: Systems and
Humans, 2000, 30(6): 720 – 730.
[12] L. Huang, S. S. Ge, T. H. Lee. Neural network adaptive impedance
control of constrained robots. IEEE International Symposium on [30] R. Hunt, K. J. Zbikowski. Neural Adaptive Control Technology.
Intelligent Control, Vancouver: IEEE, 2002: 615 – 619. Singapore: World Scientific, 1996.
[31] P. V. Osburn, H. P. Whitaker, A. Kezer. New Developments in
[13] C. Wang, Y. Li, S. Ge, et al. Continuous critic learning for
the Design of Model Reference Adaptive Control Systems. Easton:
robot control in physical human-robot interaction. Proceedings
Institute of Aeronautical Sciences, 1961.
of the 13th International Conference on Control, Automation and
Systems, Gwangju: IEEE, 2013: 833 – 838. [32] P. Parks. Lyapunov redesign of model reference adaptive control
systems. IEEE Transactions on Automatic Control, 1966, 11(3):
[14] Y. Li, S. S. Ge, C. Yang. Impedance control for multi-point
362 – 367.
human-robot interaction. Proceedings of the 8th Asian Control
Conference, Kaohsiung: IEEE, 2011: 1187 – 1192. [33] K. J. Åström, B. Wittenmark. A survey of adaptive control
applications. Proceedings of the 34th IEEE Conference on Decision
[15] T. Tsuji, Y. Tanaka. Tracking control properties of human-robotic
and Control, New Orleans: IEEE, 1995: 649 – 654.
systems based on impedance control. IEEE Transactions on
Systems, Man and Cybernetics – Part A: Systems and Humans, [34] J. Aseltine, A. Mancini, C. Sarture. A survey of adaptive control
2005, 35(4): 523 – 535. systems. IEEE Transactions on Automatic Control, 1958, 6(1): 102
– 108.
[16] K. Doya, H. Kimura, M. Kawato. Neural mechanisms of learning
and control. IEEE Control Systems Magazine, 2001, 21(4): 42 – [35] H. Unbehauen. Adaptive dual control systems: A survey. Adaptive
54. Systems for Signal Processing, Communications, and Control
Symposium, Lake Louise: IEEE, 2000: 171 – 180.
[17] D. Wolpert, M. Miall, R. Chris, et al. Internal models in the
cerebellum. Trends in Cognitive Sciences, 1998, 2(9): 338 – 347. [36] N. M. Filatov, H. Unbehauen. Survey of adaptive dual control
methods. IEEE Proceedings Control Theory and Applications,
[18] D. Kleinman, L. Baron, S. Baron, et al. An optimal control model
2000, 147(1): 118 – 128.
of human response – Part I: Theory and validation. Automatica,
1970, 6(3): 357 – 369. [37] K. J. Åström, B. Wittenmark. Adaptive Control. India: Pearson
Education, 2001.
[19] R. C. Miall, D. J. Weir, D. M. Wolpert, et al. Is the cerebellum
a smith predictor. Journal of Motor Behavior, 1993, 25(3): 203 – [38] I. D. Landau. A survey of model reference adaptive techniques
216. theory and applications. Automatica, 1974, 10(4): 353 – 379.
 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016 81

Bakur ALQAUDI received the B.Sc. de- He is currently pursuing his Ph.D. at University of Texas at Arlington,
gree in Electronics Commination and Elec- Texas, U.S.A. E-mail: [email protected].
trical Automation from the Yanbu Indus-
trial College, Yanbu, Saudi Arabia, and Frank L. Lewis (S’70-M’81-SM’86-F’94) re-
M.Sc. degree in Electrical Engineering fo- ceived the B.Sc. degree in Physics and Elec-
cusing in Biorobotics, Control and Cyber- trical Engineering and the M.S.E.E. degree,
netics from Rochester Institute of Technol- both from Rice University, Houston, TX,
ogy, Rochester, NY, U.S.A., in 2008 and U.S.A., the M.Sc. degree in Aeronautical
2012, respectively. He is currently pursuing Engineering from the University of West
the Ph.D. degree with the University of Texas at Arlington, Arling- Florida, Pensacola, FL, U.S.A., and the Ph.D.
ton, TX, U.S.A. He joined Yanbu Industrial College as an Instructor, degree from the Georgia Institute of Tech-
from 2008 to 2009, and received the King’s scholarship for Gas and nology, Atlanta, GA, U.S.A. He is a Univer-
Petroleum track in 2009. His current research interests include physi- sity of Texas at Arlington Distinguished Scholar Professor, a Teaching
cal human-robot interaction, adaptive control, reinforcement learning, Professor, and the Moncrief-O’Donnell Chair with the University of
robotics, and cognitive-psychological inspired learning and control. E- Texas at Arlington Research Institute, Fort Worth, TX, U.S.A. He is the
mail: [email protected]. Qian Ren Thousand Talents Consulting Professor with Northeastern
University, Shenyang, China. He is a Distinguished Visiting Profes-
Hamidreza MODARES received the B.Sc.
sor with the Nanjing University of Science and Technology, Nanjing,
degree from the University of Tehran,
China, and the Project 111 Professor with Northeastern University.
Tehran, Iran, and the M.S. degree from
His current research interests include feedback control, intelligent
the Shahrood University of Technology,
systems, cooperative control systems, and nonlinear systems. He has
Shahrud, Iran, in 2004 and 2006, respec-
authored numerous journal special issues, journal papers, 20 books,
tively. He is currently pursuing the Ph.D.
including Optimal Control, Aircraft Control, Optimal Estimation, and
degree with the University of Texas at Ar-
Robot Manipulator Control, which are used as university textbooks
lington, Arlington, TX, U.S.A. He joined the
worldwide and he holds six U.S. patents. Dr. Lewis was a recipi-
Shahrood University of Technology as a Uni-
ent of the Fulbright Research Award, the National Science Foundation
versity Lecturer, from 2006 to 2009. Since 2012, he has been a Re-
Research Initiation Grant, the American Society for Engineering Educa-
search Assistant with the University of Texas at Arlington Research
tion Terman Award, the International Neural Network Society Gabor
Institute, Fort Worth, TX, U.S.A. His current research interests include
Award, the U.K. Institute of Measurement and Control Honeywell
optimal control, reinforcement learning, distributed control, robotics,
Field Engineering Medal, the IEEE Computational Intelligence Society
and pattern recognition. E-mail: [email protected].
Neural Networks Pioneer Award, the Outstanding Service Award from
Isura RANATUNGA (S’09) received the Dallas IEEE Section, and selected as an Engineer of the Year by the
B.Sc. degree from the University of Texas Fort Worth IEEE Section. He was listed in Fort Worth Business Press
at Arlington, Arlington, TX, U.S.A., in 2010, Top 200 Leaders in Manufacturing and Texas Regents Outstanding
where he is currently pursuing the Ph.D. Teaching Award in 2013. He is a PE of Texas and a U.K. Chartered
degree with a focus on robotics and au- Engineer. He is a member of the National Academy of Inventors and
tomation, both in Electrical Engineering. He a fellow of International Federation of Automatic Control and the U.K.
is a Graduate Research Assistant with the Institute of Measurement and Control. He is a Founding Member of
University of Texas at Arlington Research the Board of Governors of the Mediterranean Control Association.
Institute, Fort Worth, TX, U.S.A., and the Board of Governors of the Mediterranean Control Association. E-mail:
Next Generation Systems Research Group. His current research inter- [email protected].
ests include force control, physical human-robot interaction, bipedal
Dan O. POPA (M’93) received the B.A.
walking, adaptive robot control, and autonomous navigation. E-mail:
degree in Engineering, Mathematics, and
[email protected].
Computer Science and the M.S. degree in
Shaikh M. TOUSIF received the B.Sc. de- Engineering, both from Dartmouth College,
gree in Electrical and Electronic Engineer- Hanover, NH, U.S.A., and the Ph.D. degree
ing from American International University- in Electrical, Computer and Systems Engi-
Bangladesh, Bangladesh in 2009, and the neering from Rensselaer Polytechnic Insti-
M.Sc. degree in Electrical Engineering from tute (RPI), Troy, NY, U.S.A., in 1998, focus-
University of Texas at Arlington in 2014. ing on control and motion planning for non-
From 2009 to 2012 he was a lecturer holonomic systems and robots. He is an Associate Professor with the
in the Department of Electrical Engineer- Department of Electrical Engineering, University of Texas at Arling-
ing in American International University- ton, and the Head of the Next Generation Systems Research Group.
Bangladesh and responsible for teaching many engineering courses. He joined the Center for Automation Technologies at RPI, where he
82 B. Alqaudi et al. / Control Theory Tech, Vol. 14, No. 1, pp. 68–82, February 2016

was a Research Scientist until 2004, for over 20 industry-sponsored of several prestigious awards, including the University of Texas Re-
projects. He was an Affiliated Faculty Member of the University of gents Outstanding Teaching Award. He serves as an Associate Editor
Texas at Arlington Research Institute, Fort Worth, TX, U.S.A., and a of the IEEE Transaction On Automation Science and Engineering and
Founding Member of the Texas Microfactory Initiative, in 2004. His the Journal of Micro and Bio Robotics (Springer). He is an active mem-
current research interests include the simulation, control, packaging ber of the IEEE Robotics and Automation Society Conference Activities
of microsystems, the design of precision robotic assembly systems, Board, the IEEE Committee on Micro-Nano Robotics, and the ASME
and control and adaptation aspects of human-robot interaction. He Committee on Micro-Nano Systems and a member of ASME. E-mail:
has authored over 100 refereed publications. Dr. Popa was a recipient [email protected].