Annual Reviews in Control 43 (2017) 188–198

Contents lists available at ScienceDirect

Annual Reviews in Control

journal homepage: www.elsevier.com/locate/arcontrol

Review article

A review on model reference adaptive control of robotic manipulators

Dan Zhang∗, Bin Wei
York University, 4700 Keele Street, Toronto, Ontario, M3J 1P3, Canada

a r t i c l e i n f o a b s t r a c t

Article history: The accuracy of the motion control for robotic mechanisms will have an effect on their overall perfor-
Received 3 November 2016 mance. Under the condition where the robotic end-effector carries different loads, the motions for each
Revised 16 January 2017
joint of robotic mechanisms change depending on different payload masses. Conventional control sys-
Accepted 15 February 2017
tems possess the potential issue that they cannot compensate the load variation effect. Adaptive control,
Available online 22 February 2017
especially the model reference adaptive control (MRAC), has therefore been put forward to handle the
Keywords: above issue. Adaptive control is generally divided into three categories, model reference, self-tuning and
Model reference adaptive control gain-scheduled. In this study, the authors only focus on the model-reference approach. To the best of the
Robotic manipulator authors’ knowledge, very few recent research articles can be found in the area of MRAC especially for
Review robotic mechanisms since robotic system is a highly nonlinear system, and it is difficult to guarantee the
stability of MRAC in such system. This study presents a review and discussion on the MRAC of robotic
mechanisms and some issues of MRAC for robotic mechanisms are also demonstrated. This study can
provide a guideline for upcoming research in the field of MRAC for robotic mechanisms.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction
The control problem for robotic mechanism is usually described
as follows, provided an expected path, a numerical model of the
mechanism and its interactions with surroundings, finds a con-
trol algorithm that delivers force or torque signals to the actuators,
in this way the robotic mechanism is able to accomplish antici-
pated movement. The control design for a serial mechanical mech-
anism contains two major steps. A robotic end-effector travel path
is first specified, e.g. making the end-effector move from position
A to position B. From the movement trajectory of the end-effector
and also via resorting to the inverse kinematics, the motions of
the joints can therefore be determined so as to create the desired
trajectory for the end-effector. The next step is to figure out the
amount of torque that one has to apply to joints so that the joints
are able to achieve the desired motion. The torque can be calcu-
lated based on inverse dynamic equations.
Controlling the robotic manipulator to conduct in a designated
fashion is a tough matter since the robotic system is extremely
nonlinear. Regarding robotic systems, the dynamic equations’ co-
efficients contain joint and payload variables. These variables may
be unknown or may even alter during the task. When a robotic
mechanism is in motion, the joint variables change, and this can
make the robotic system’s dynamic equation alter throughout a
∗
Corresponding author.
E-mail addresses: dzhang99@yorku.ca (D. Zhang), binwei28@yorku.ca (B. Wei).
task. For the purpose of having a higher repeatability and accuracy
in the robotic system’s performance, one needs to employ a con-
trol system (e.g. MRAC) which can take in the dynamic character-
istic changes of the robotic system. Traditional control techniques
treat a robotic mechanism as uncoupled linear subsystems; they
are able to provide adequate performances at low speeds. However,
for applications that require high-speed actions, they are no longer
effective. The utilization of the PID control system for controlling
sometimes may not promise system stability or optimal control for
the system (Visioli, 2006). For the aim of addressing the above is-
sue, one can consider to employ the adaptive control. MRAC is one
of the most prevalent and well-used methodologies.
Adaptive control is able to adjust to a manipulated system with
parameters that change constantly, or are originally unknown. Re-
garding non-adaptive control system, the control system is mod-
elled on the basis of the system’s priori data, in other words,
we know the system and develop the controller only intends for
that system and presume the changing phenomenon does not ex-
ist within the system. With respect to the adaptive controller, on
the contrary, the controller does not have to rely on the prior data
from the system, and if some random changes happen in surround-
ings, the controller is able to handle it through adjusting to the
altered states. If one considers a system with a known transfer
function, and one designs a fixed classical control system, the con-
trol system will maintain the specified parameters all the way to
the point where it does not apply to the system anymore. It can
be claimed that this control system relies on its structure and is
modelled on a-priori knowledge, this is a non-adaptive control sys-

http://dx.doi.org/10.1016/j.arcontrol.2017.02.002
1367-5788/© 2017 Elsevier Ltd. All rights reserved.
 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
Fig. 1. Non-adaptive control.
Fig. 2. Adaptive control.
tem. Notwithstanding, if the control system is relying on posteriori
data, for instance, if the control system’s parameters are changing,
due to the system parameters’ changes or due to the interferences
originating from the surroundings, this control system is named
adaptive. If the system faces uncertain interferences, or it is an-
ticipated to endure changes in its parameters in a fashion that is
unpredicted from the start, under this situation we employ adap-
tive control. Nevertheless, in other situations where we know how
the system’s working state is going to alter, taking an aircraft as an
example, it is known that designing of the aircraft control system
is influenced by the speed and altitude, and one expects it to travel
at particular values for altitude and speed. Under this situation a
controller is able to be designed for every single anticipated oper-
ating point and the different controllers can be interchanged each
other, this is named gain-scheduling. In other scenarios where we
know the system’s parameters vary, but one also knows an extent
for the parameter change, in this situation it is feasible to develop
a fixed controller which is able to handle different changes of the
parameters, and promise the stability, this type of control system
is a robust control.
Regarding the non-adaptive control, from Fig. 1, it can be ob-
served that under the situation where the performance error is im-
proved, the modelling accuracy is also increased. Furthermore, it is
believed that future is same with the present, disregarding changes
in environment and dynamics, and structure damage.
For the adaptive control, it obtains a designated system perfor-
mance asymptotically, the modelling accuracy performance is not
compromised, i.e. adaptive control can guarantee the system per-
formance regardless of the accuracy of the model, as demonstrated
in Fig. 2, and more importantly, it can be self-improved when un-
foreseen and adverse conditions exist.
The adaptive control can be classified into the three types,
MRAC, self-tuning and gain-scheduled, as illustrated in Fig. 3. The
authors here primarily focus on the model-reference approach.
189
Regarding the MRAC, there is an input being given to the real
and the reference systems, and the error among the real output
and the reference model’s output is generated. The error is then
employed to modify the controller’s parameters in order to make
the error minimized. Fig. 4 illustrates such a control system.
As comparing to other control techniques, adaptive approach
is more likely to perform better for a broad array of movements
and loads. The asset of the MRAC is that one does not have
to completely know the plant parameters, rather, the plant pa-
rameters’ approximations can be employed and the adaptive con-
trol system resorts to prior input/output data to further improve
these approximations. On the downside, there are two disadvan-
tages for this type of control. The control system’s stability per-
formance is crucial because it is hard to develop a stable adapta-
tion rule. The other issue one faces is that MRAC system depends
on cancelling the non-linear segments via the reference model
(Sutherland, 1987). In real situation, fully cancellation cannot be
ensured, but the non-linear segments can be forged very small to
the point where one is able to neglect them. MRAC technique was
first presented in Whitaker, Yamron, and Kezer (1958), when the
authors examined adaptive flight control system, utilizing a refer-
ence model to create an offset among the real and ideal behav-
ior. The offset was utilized to alter the control system parameters
for the purpose of having desired outcome regardless of changing
system dynamics and uncertainties. The objective of employing an
adaptive controller is to reach and retain a satisfactory degree in
the control system performance when plant parameter variations
exist. While a traditional feedback controller is primarily custom
built for eliminating the impact of disturbances on the controlled
variables, an adaptive controller is primarily custom built for elim-
inating the impact of parameter disturbances on the control sys-
tem’s performance.
2. Model reference adaptive control of robotic manipulators
2.1. General adaptive control
For a conventional controller, feedback is employed to reject
the disruption impact that acts on the manipulated variables for
the purpose of bringing these manipulated variables back to the
ideal value. In order to do so, one determines the variables and
compares the variables to the ideal values and an error can be
generated. This error is provided to the control system. In these
feedback systems, one is able to modify the control system param-
eters so that an ideal control performance can be accomplished.
This is conducted on the basis of knowing a priori plant dynamics’
information. In the cases where the plant dynamic models’ param-
eters vary with respect to time, the traditional control is no longer
able to handle it because the control performance can be deterio-
rated. In cases like this, one can resort to the adaptive control. A
structured technique for designing distributed and adaptive control
systems is demonstrated in Valente, Carpanzano, and Brusaferri
(2011). Distributed architectures are visualized as separate units
with conventional interfaces that can be altered and re-utilized not
having the entire control framework been affected. While regard-
ing the centralized control structures, every single modification of
the machine configuration demands a large scale control system
changes. In reconfigurable manufacturing system, modular and dis-
tributed structure is crucial for ensuring the ability of every unit or
parts of the control to be adjusted in a situation where a hardware
reconfiguration happens.
In Bi, Liu, and Baumgartner (2015), the authors illustrated the
sustainable manufacturing based on the robotic system reconfig-
uration via employing robot modules. The modules are suction
pump and adapters, end-effector, PLC and robot controller, sen-
sors and power supply. Regarding the control, two different con-
190 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
Fig. 3. Adaptive control classification.
Fig. 4. A MRAC system.
trol systems need to be run. One control system is utilized to con-
trol the robotic mechanism and the other one is a PLC, which will
maneuver user inputs and sensor information for the purpose of
signalling the robotic mechanism’s control system which program
to operate. Under the condition where the robot is reconfigured,
the control system also has to be reconfigured. The software sys-
tem has to be modified in order to assist the interactions within
system modules.
In Wilson and Rock (1995), the neural networks (NN) are uti-
lized for modelling the control reconfiguration of a space robotic
mechanism. The conventional control system was displayed, and
through employing the NN, the conventional control system was
modified to a reconfigurable control system. Two NN control sys-
tems were put forward to accomplish rapid adaptation. Initially,
a fully-connected structure, which is able to integrate a priori
estimated linear solution instantaneously, was employed, this al-
lows speedy stabilization by an estimated linear control system.
After that, a back-propagation learning approach was employed
which enables back-propagation with discrete-valued functions. It
demonstrates an adjustable NN-based controller of the robotic
mechanism that used in space.
In Jung, Jeong, and Lee (2005), an adjustable flight controller
employing the mode switching of multiple models was investi-
gated. The fundamental concept is to employ the approximates
of the aircraft parameters to determine what control system one
needs to select under a specific flight state, which is associated
with the multi-mode adaptive control. In Landau, Lozano, Mo-
hammed and Karimi (2011), the fundamental notion of adaptive
control and different categories were presented, i.e. open-loop
adaptive, indirect adaptive, direct adaptive, and conventional con-
trols. The adaptive controller is able to be regarded as a traditional
feedback controller, however the difference is that the performance
index is regarded as the manipulated variable, which means the
adaptive control system contains the following two loops, the tra-
ditional feedback system (is regarded as one loop) and the adapta-
tion system (is regarded as a second one).
A control development approach is put forward in Valentea,
Mazzolinib, and Carpanzanoa (2015), which contains three main
procedures: initial control concept modelling, implementation ad-
vancement and robustness assessment of result. The control sys-
tem is viewed as a group of separate and distributed control units,
having the ability of nesting one another. The control logics’ forma-
tion lays a foundation for the overall control development process.
To realize the reconfiguration goal for the controller, a crucial at-
tribute for the control structure is the modularity and distribution
of the control decisions across the different entities.
The MRAC is widely used in numerous areas, for example, the
process control, which is a quasi-linear system. Robotic system is a
high nonlinear system, and it is difficult to guarantee the stability
of MRAC in nonlinear system, which is why very few recent re-
search articles can be found in the area of MRAC for robotic mech-
anisms.
2.2. MRAC used in nonlinear process control
The model reference adaptive control is widely used in process
control, which is a quasi-linear system, and there are tremendous
amount of studies for the MRAC being used in the nonlinear pro-
cess control. Here the authors just present some typical examples,
and will not list all of them. Interested readers can resort to related
references. In Avinashe, Akhil, Dhanoj, and Shinu (2015), an appli-
cation of a MRAC in a non-linear process was presented. It was
compared to the case where the non-linear process was under the
conventional proportional-integral control, and the simulation re-
sult indicates that the model reference adaptive control performs
better than that of the proportional-integral control in terms of
steady state error and settling time. Similarly, in Krishna, Kumar,
and Shaik (2012), a MRAC was simply applied to a spherical tank,
i.e. a nonlinear process control, and the model reference adaptive
control was designed based on the MIT rule, and the approach was
approved to be effective. In Joseph and Isaac (2013), a MRAC was
applied to a conical tank level control, and it was compared to a
 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
proportional-integral control based on the Ziegler Nichols method.
The LabVIEW simulation shows that the model reference adaptive
control is better than that of the proportional-integral control in
terms of error minimization. In Oltean, Dulau, and Duka (2016),
the authors presented a case study of MRAC used for a level pro-
cess control. Two model reference adaptive control schemes were
presented. The first scheme is based on the MIT rule and the sec-
ond scheme is based on the Lyapunov methodology. It was found
that the MRAC based on the MIT rule for the level process con-
trol is only locally stable, whereas the MRAC based on the Lya-
punov methodology is globally stable. In Ram and Lincoln (2013),
a model reference tracking based fuzzy adaptive PI control system
was put forward to deal with a non-linear process plant. The con-
trol parameters are estimated based on the difference between the
plant and reference model. The matlab simulation indicates that
the control system can generate proper signals to control the plant
under the case where plant being nonlinearity and measurement
noise and disturbance exist. In Babu and Lincon (2013), a MRAC
system was designed with control law that is to be combined with
an adaptive law for a nonlinear process. A spherical tank level pro-
cess was employed as an example to validate the design procedure
of the control system. The control system was compared to the
conventional PI control system, and the results illustrate that the
MRAC system has better rise time and settling time performance.
2.3. Model reference adaptive control for robotic manipulators
Designing of a non-adaptive control system usually does not
take in the effect of nonlinearities and dynamic couplings among
joints’ motions. In the situation where robotic mechanism mo-
tions demand high speeds and accelerations, the control perfor-
mance will be severely deteriorated. In addition, designing of a
non-adaptive control system requires the accurate data and spe-
cific use of the complicated system dynamics and system parame-
ters. Uncertainties are able to make dynamic performance deterio-
rate and system unstable. Numerous uncertainties do exist in every
kind of robotic mechanism’s dynamic models, for instance, length
of linkages, unpredictable loads, elasticities and backlashes of gear
trains are either not possible to identify exactly or vary in a non-
predictable fashion. This is the reason that the adaptive control is
necessary to resolve the above issue.
MRAC approach and its application to robotic mechanisms were
raised in Neuman and Stone (1983) and Amerongen (1981). Com-
monly design issues in adaptive robotic mechanism control are
concisely illustrated. S. Dubowsky (Dubowsky & Desforges, 1979)
is the pioneer researcher who employed the MRAC approach in the
robotic mechanism. The procedure closely follows the technique in
Donalson and Leondes (1963). A linear time-invariant differential
formulation was employed to function as the reference model for
every single DOF of the robotic mechanism. The author manipu-
lated the mechanism via modifying the gains of the position and
velocity feedback system to closely follow the model. One resorted
to a steepest-descent approach to update the feedback gains. The
dynamics of the reference model were first determined, followed
by determination of the nonlinear dynamic formulation for the
mechanism, yet how this formulation is associated with the La-
grange formulation did not illustrate clearly. Subsequently, an er-
ror function was determined and it follows the approach of the
steepest-descent. A group of formulations were obtained for the
parameter adjustment system, which makes the error that gener-
ated by the real response of the closed-loop system and the re-
sponse of the reference model minimized.
An adaptation calculation rule was designed in Horowitz and
Tomizuka (1986) for a serial robotic mechanism in order to com-
pensate the nonlinear segment in dynamic formulations and de-
191
couple the dynamic interaction between the joints. The following
figure shows such control system (Fig. 5).
One can see that the plant output is compared to the refer-
ence model, and an offset can therefore be generated. This off-
set is subsequently utilized by the adaptive algorithm. The output
from the adaptive algorithm becomes input elements to the plant.
Meantime, the plant output is compared to the desired model rp,
an offset can also be generated, and the offset is going through
the integration block and the result of which subsequently sub-
tract the feedback position and velocity processed by Kd and Kp
elements. The process bears a resemblance to the conventional
P-I-D control. This process’s output multiplies the adaptive algo-
rithm components and the result of which subsequently adds an-
other part of the adaptive algorithm components is the plant in-
put. The above procedure is not going to stop until the offset be-
tween the reference model and plant output approaches zero. The
ideal system is set apart from the plant, i.e. the plant variables’
feedback will not be utilized to associate with the input to the
reference model. For detailed information, readers can be referred
to Horowitz and Tomizuka (1986). The adaptation technique that
was put forward in this manner differs from the technique from
Dubowsky’s (Dubowsky & Desforges, 1979). Three major dissimilar-
ities are summarized as below: first of all, the entire controller in
Horowitz’s technique has an inner loop MRAC architecture and an
outer loop PID control system, while the controller in Dubowsky’s
technique is completely modelled on the basis of the MRAC sys-
tem; subsequently, the couplings between each joint and nonlinear
segments in the robotic mechanism’s formulations are disregarded
in Dubowsky’s approach while they are included in Horowitz’s ap-
proach; finally, the design technique in Horowitz’s study is on the
basis of the hyper-stability methodology while the adaptive sys-
tem model presented in Dubowsky and Desforges (1979) is based
on the steepest-descent method. Also in Horowitz and Tomizuka
(1986), there are some issues as follows. (1) The authors stated
in Horowitz and Tomizuka (1986) that “The overall control sys-
tem will have an inner loop MRAS controller and an outer loop
PID action controller with fixed gains”. The above statement is not
consistent with Figs. 4 and 5 in Horowitz and Tomizuka (1986).
According to Figs. 4 and 5 in Horowitz and Tomizuka (1986), the
controller contains an MRAC inner loop system, but it does not
have an outer loop PID action controller. (2) For Figs. 4 and 5 in
Horowitz and Tomizuka (1986), the outer loop controller is in the

form of Ki e − K p x p − Kd xv , but in the similar paper (Horowitz,
Tsai, & Anwar, 1987) by the same authors, the outer loop controller

is in the form of Ki e + K p x p + Kd xv . It is not consistent. (3) For
the Eq. (13’) in Horowitz and Tomizuka (1986), the adaptive algo-
rithm are all positive, but in Horowitz (1983) (note that Horowitz
and Tomizuka, 1986 is part of the dissertation Horowitz, 1983), the
adaptive algorithms are all negative, it is also not consistent.
Model reference, self-tuning and linear perturbation adaptive
controls are concisely discussed in Hsia (1986). Regarding the
MRAC, the fundamental concept is to develop a control signal to
the robotic mechanism’s dynamic formulation, which is able to
make the robotic mechanism to act in a designated way defined by
the reference model, and the adaptation calculation rule is devel-
oped from the Lyapunove stability principle. The model reference
adaptive control technique illustrated in Srinivasan (1987) is on the
basis of the partitioning control ideology, which gives them capa-
bility to compensate for non-linear segments in the dynamic for-
mulations and decouple dynamic interactions among the linkages.
It followed and employed the approach presented in Horowitz
(1983). Upcoming study will be putting an emphasis on simplify-
ing MRAC methods, as applying them to robotic mechanisms’ real
time control is a daring work.
A MRAC structure for a three degree-of-freedom serial robotic
mechanism was illustrated in Horowitz (1983). It is about apply-
192 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
Fig. 5. A MRAC control system.
ing the MRAC system to serial robotic mechanisms. Because the
dynamic formulations of serial robotic mechanisms are extremely
nonlinear and the load from time to time changes or is not known,
the MRAC system was utilized to the serial robotic mechanisms.
An adaptive algorithm was put forward to compensate nonlinear
segments that exist in the dynamic formulations and to decouple
dynamic interactions among joints. In the end, a three degree-of-
freedom serial robotic mechanism was employed in a numerical
simulation and the outcomes indicate that by using the adaptive
control method the sensitiveness of the mechanism’s performance
to configuration and load changes can be successfully reduced
to an acceptable level. The primary subject of the approach in
Horowitz (1983) is summarized in three major steps: deterministic
nonlinearity compensation and decoupling control need to be first
conducted since one is required to determine the inertia matrix
and nonlinear segment, after that, the adaptive nonlinearity com-
pensation and decoupling control process are therefore proposed
and performed. The purpose of this step is to modify the inertia
matrix and nonlinear segment adaptively rather than calculating
them; and the adaptation algorithm was designed; lastly, one can
finalize the entire control system through including the feedback
gain. In Horowitz (1983), the author did not completely employ
the Landau’s hyper-stability technique (Landau, 1979), the author
employed parts of the technique, and they themselves designed
the adaptation algorithm. As stated by Hsia (1986), Horowitz’s ap-
proach is separate from Landau’s hyper-stability approach. In ad-
dition, according to Sutherland (1987), it was claimed that “While
Landau’s method replied on a pre-specified parameter matrix for a
model and continuous adaptation of the plant parameters, it will
be seen later that it is possible to estimate the model parameters
and adapt them continuously”, based on the above expression, one
can see that Horowitz did not employ Landau’s technique in de-
termining the adaptation calculation rule, rather, the author devel-
oped the adaptation algorithm rule through a different approach,
but how the adaptation algorithm rule was determined was not
specifically illustrated. In Sutherland (1987), the author applied the
Horowitz’s technique (Horowitz, 1983) to a two degree-of-freedom
serial robotic mechanism and a flexible mechanism.
In Tomizuka and Horowitz (1988), and Tomizuka, Horowitz, and
Anwar (1986b), the experimental studies of the continuous and
discrete time adaptive control for a one degree of freedom robotic
mechanism and a Toshiba TSR-500 V robotic mechanism were
performed. Horowitz et al. (1987) is the extension of Tomizuka,
Horowitz, and Anwar 1986a) for one-axis direct drive robot ma-
nipulator. A two axes direct drive robot manipulator was employed
and analyzed instead of a single axis case. In Tomizuka, Horowitz,
and Teo (1985), the authors illustrated the experimental assess-
ment of MRAC system and robust control system for a robot arm
positioning under the cases where load variations exist. The exper-
iment outcome indicates that these two methods are able to be
insensitive with regards to the load changes effect. There are four
adaptive control techniques (i.e. computed torque method, variable
structure method, adaptive linear model following method, and
adaptive perturbation method) that are able to be applied to the
robot arm being outlined in Hamadi (1989). The adaptive nonlinear
model following method was designed after. It mainly incorporated
the self-tuning regulator and the MRAC.
An improved form of the method in Horowitz (1983) was put
forward in Sadegh and Horowitz (1987). The assumption that the
inertia matrix and nonlinear term are constant throughout adap-
tation can be got rid of through modifying the control rule and
parameter adaptation rule. The above adaptive control system’s
stability is verified. The details can be referred to Sadegh and
Horowitz (1987). Fig. 6 shows such modified control system ver-
sion. The above is named “exact compensation adaptive control
law (ECAL)”.
However, it was realized that this approach consumes an exces-
sive amount of time because calculations of extremely nonlinear
functions of joints’ positions and velocities are involved. In order to
resolve this problem, in Sadegh and Horowitz (1990) and Sadegh
(1987), a more improved model was proposed. The improvement
is composed of employing the ideal joint’s positions and velocities
when determining the nonlinearity compensation control system
and the parameter adaptation rule rather than employing the ac-
tual ones, and it is named “desired compensation adaptive control
law (DCAL)”. The entire evolvement procedure can be briefly illus-
trated in Fig. 7.
Because Nader incorporated Craig’s approach (Craig, Hsu, & Sas-
try, 1986) into the Horowitz’s approach, the assumed constant in-
ertia matrix and nonlinear segment throughout adaptation is able
to be got rid of. Craig’s approach is about re-parametrization, i.e.
breaking down the nonlinear parameters that are in the robotic
 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198 193

Fig. 6. An modified MRAC control system.

Nader Sadegh’s
Nader Sadegh’s first
second modified
modified version
version
(Exact
Horowitz’s method (Desired
compensaon
compensaon
adapve control law
adapve control law
ECAL)
DCAL)

Fig. 7. Modification process.

dynamic equation into two components multiplication: one com-

ponent being a constant unknown lump, which contains the link-
age dimensions and payload, and the other component being a
known nonlinear expression of the robotic system structural dy-
namics. The parameter adaptation rule is geared to the estimation
of the constant quantities that are not known. One approach to
reparametrize the mechanism’s dynamic formulations is to decom-
pose every single element of inertia matrix, nonlinear term and
gravity term to the multiplication of constant terms that are not
known and known expressions of the joint displacement vector.
Another approach is to re-parametrize the dynamic formulation to
the multiplication of the constant vector that are not known and a
matrix formulated by known expressions of joint position.
To give a general mathematical description of the model ref- Fig. 8. Two-link manipulator.
erence adaptive control algorithm for a general robotic system in
order to help clarify the above concepts and show the structure  
of the controller, here a 2-DOF robotic arm, as shown in Fig. 8, is +m2 l1 l2 cos θ2 θ˙ 1 θ˙ 1 + θ˙ 2 (4)
used as a case study. The Lagrange equation can be derived as follows:
The torques applied to the joints can be determined by using
the Lagrange method: L = K−P
1 1  2
d∂L ∂L = (m1 + m2 )l1 θ˙ 12 + m2 l2 θ˙ 1 + θ˙ 2
2 2
τ1 = − (1) 2 2
∂ θ1
dt ˙ ∂ θ1  
+m2 l1 l2 cos θ2 θ˙ 1 θ˙ 1 + θ˙ 2
d ∂L ∂L
τ2 = − (2) −(m1 + m2 )gl1 sin θ1 − m2 gl2 sin(θ1 + θ2 ) (5)
dt ∂ θ˙ 2 ∂ θ2
Thus:
The total kinetic and potential energy are expressed as:
d∂L ∂L
P = (m1 + m2 )gl1 sin θ1 + m2 gl2 sin(θ1 + θ2 ) (3) τ1 = −
∂ θ˙ 1 ∂ θ1
dt
= ((m1 + m2 )l1 + m2 l2 + 2m2 l1 l2 cos θ2 )θ̈1
2 2
1 1  2
K = (m1 + m2 )l1 2 θ˙ 12 + m2 l2 2 θ˙ 1 + θ˙ 2
2 2
194 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198

Fig. 9. Joints 1 and 2 output under PID.

+(m2 l2 + m2 l1 l2 cos θ2 )θ̈2

2 the load is 0, the joint output is shown in a solid blue line in Fig.
9(a); when the load is increased to 5 and 15, it can be observed
+(−2m2 l1 l2 sin θ2 )θ˙ 1 θ˙ 2 + (−m2 l1 l2 sin θ2 )θ˙ 2
2 that joint 1 output deviates from the previous case, as shown in
+((m1 + m2 )l1 cos θ1 + m2 l2 cos(θ1 + θ2 ))g (6) Fig. 9(a), and also the joint output fluctuates. The same goes to
joint 2, as shown in Fig. 9(b). Fig. 10(a) and (b) shows joints 1 and
d ∂L ∂L 2 output under the MRAC case. By using the MRAC approach, three
τ2 = − lines agree with one another for different payloads, i.e. the load
dt ∂ θ˙ 2 ∂ θ2 variation effect is compensated.
= (m2 l2 + m2 l1 l2 cos θ2 )θ̈1 + (m2 l2 )θ̈2 + (m2 l1 l2 sin θ2 )θ˙ 12
2 2
In Yuh and Holley (1988), a direct approach to discrete-time
+m2 l2 cos(θ1 + θ2 )g (7) MRAC was designed, and the authors employed the hyperstability
theory to make the controller stable, and the designed controller
If these are expressed in a matrix form, the following is ob- was applied for an industrial robotic manipulator. Through sim-
tained:
       ulation, it was found that the proposed MRAC could provide ro-
τ1 m11 m12 θ̈1 n bust properties under the poor a priori information in terms of the
= Mθ̈ + N = + 11 (8)
τ2 m12 m22 θ̈2 n21 robot dynamics and circumstances. The above study was merely a
simplified version that was originally proposed by Tomizuka and
where m11 = (m1 + m2 )l1 + m2 l2 + 2m2 l1 l2 cos θ2 , m12 = m2 l2 +
2 2 2
Horowitz (1983), the only difference is that only the linear terms
m2 l1 l2 cos θ2 , m22 = m2 l2 , n11 = 2(−m2 l1 l2 sin θ2 )θ˙ 1 θ˙ 2 + (−m2 l1 l2
2
for the manipulator were considered, which allows simplified dy-
sin θ2 )θ˙ 22 , n21 = m2 l1 l2 sin θ2 θ˙ 12 namic equations to be obtained.
After applying the MRAC approach, one has: In Ampsefidis and Bialasiewicz (1993), a MRAC system that is
∧ ∧ based on the Lyapunov’s direct methodology was designed for the
Control l erOut = τ = M u + V −Fp e − Fv e˙ (9)
 SISO system, and this MRAC does not need the perfect model fol-
where u = KI (r p − x p ) − K p x p − Kd xv . Since the mechanism dy- lowing conditions. Because the direct Lyapunov’s technique is em-
namic equation is: ployed, the stability conditions are ensured. The proposed MRAC
τ = Ma + V + Gg (10) was subsequently applied to a manipulator. In Maliotis (1991), a di-
rect MRAC strategy was proposed for a robotic manipulator based
Hence, the output from the mechanism is: on the Lyapunov stability criterion. The designed control rule is on
∧ ∧
the basis of the formulation of a plant state model which has all
M u + V −Fp e − Fv e˙ = τ = Ma + V (11)
the robot manipulator non-linear characteristics included. A mod-
Thus, the accelerations of the joints are as follows: ified Lyapunov function was employed to derive the adaptation
   
a1 θ̈1 ∧ ∧ laws. The designed control system has two parts: one is that the
⇒ = = M−1 (M u + V −Fp e − Fv e˙ − V ) (12) known dynamics were extracted to fulfill the feedback lineariza-
a2 θ̈2
tion for the non-linear system, and another one is that a MRAC
After deriving the accelerations of the joints, a time integral is based on a modified Lyapunov function was designed on the un-
taken to have the velocities of joints 1 and 2, and a second integral known part of the plant. In Yang (1992), and Yang, Woo, and Wang
is taken to have the positions of joints 1 and 2. (1996, 1993), two discrete-time model reference adaptive control
As an example and comparison purpose, here we apply differ- methods were proposed and used in a three-link PUMA manipula-
ent payload masses at the end-effector of the robotic mechanism, tor, and the stability of the discrete-time MRAC system was proved
and apply a PID and a MRAC. For the PID control case, joint 1 mo- based on the Lyapunov theorem. In Schwartz (1994), a non-linear
tion output is shown in Fig. 9(a) and joint 2 motion output is pre- MRAC for robot manipulators was illustrated. The control adaptive
sented in Fig. 9(b). Regarding joint 1, under the condition where algorithm designed here did not require one to have the measure-
 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
Fig. 10. Joints 1 and 2 output under MRAC.
ment or estimation of the joint velocities or accelerations of the
robotic manipulators. In Tam and Thomopoulos (1992), a MRAC by
employing a modified output error method was presented and ap-
plied to a robotic manipulator. The proposed regressors and aug-
mented error in the modified output error approach lead to an
asymptotically globally stable system when the nonlinearities of
the manipulator system dynamic model were absent. In Maeda and
Totani (1990), an approach based on Popov’s hyper-stability was
employed to design the MRAC for a two-link SCARA robotic ma-
nipulator in order to achieve the goal of trajectory tracking control;
a linearized state equation with a state feedback was employed to
be the reference model. The adaptive system contains a linear and
a non-linear control, which can have the existing frequency chat-
tering in the control inputs decreased. In Kamnik, Matko, and Bajd
(1998), the MRAC was applied to the robot impedance control. The
new control system consists of a traditional manipulator position
control and an added force control. The force control scheme em-
ployed adaptive terms in order to cope with the unknown environ-
mental and robotic manipulator dynamic parameters. In Bolourchi
and Hess (1992), a tap delay filter based MRAC approach was stud-
ied and applied to a 2-DOF planar robot with revolute joints. The
tap delay filter was part of the MRAC and it allows the simple im-
plementation of an inverse modeling control scheme.
In Kirecci, Topalbekiroglu, and Eker (2013), the authors imple-
mented the model reference adaptive control to a dynamically un-
certain hydraulic robot for position tracking. A recursive parame-
ter estimator was utilized for the parameter estimation. The robot
was trialed under the case where a step input signal change oc-
curs, and the authors discovered that the MRAC is able to give
better tracking as comparing to a PID control system, and the
MRAC is adequate enough to substitute the classic PID control
system. On the other hand, the minor issue was found that a
small oscillation exist at the start of the movement when MRAC
is employed. The above study is merely a case study. In Ulrich
and Sasiadek (2010), the study illustrated the application of the
MRAC composite system used for tracing the endpoint of a flex-
ible joint robotic mechanism that is utilized in space. In Sharifi,
Behzadipour, and Vossoughi (2014), a new nonlinear model refer-
ence adaptive impedance control system was developed via com-
bining the MRAC and impedance control to handle the tracking
control issue in human–robot interaction, and the Lyapunov sta-
195
bility theorem was applied for the stability analysis. The experi-
mental results showed that the newly designed model reference
adaptive impedance control system performs better than simple
adaptive control systems in terms of tracing an ideal reference
model trajectory. Similarly, in Huh and Bien (2007), a sliding mode
based MRAC was proposed for a robotic manipulator. The novelty
of this study is that by introducing the sliding model control ap-
proach to the MRAC, the control system allows the manipulator
to follow its nominal dynamics. The control system was proved
stable through resorting to the Lyapunov approach. The simula-
tion results showed that the stiffness control by using the slid-
ing mode based MRAC approach yields improved performance over
the PID control, and the sliding mode based MRAC has smaller
tracking error compared to that of the PID control approach. In
Prakash and Anita, 2011b, a), a neural network-based MRAC intel-
ligent control system was developed by introducing an intelligent
supervisory loop to the traditional MRAC framework to overcome
the difficulty of employing MRAC for controlling the nonlinear sys-
tem in real time. The response of the traditional MRAC system
was compared with the P-I controller-based MRAC and the neu-
ral network-based MRAC. When compared, the results indicated
that the transient performance is improved substantially through
using the neural network-based MRAC and the neural network-
based MRAC presents more excellent tracking performance as com-
paring to the traditional MRAC. Also in Kamalasadan (2004) and
Kamalasadan and Ghandakly (2004), a new type of adaptive con-
trol system was developed by combining an intelligent supervi-
sory loop and the MRAC, and based on this, four different con-
trol strategies were derived, and they were applied to a single
link flexible robotic manipulator. The feasibility of the design ap-
proach was verified. In Kamalasadan and Ghandakly (2008) and
Kamalasadan, Ghandakly, and Al-Olimat (2004), a fuzzy multiple-
reference-model generator-based MRAC scheme was proposed by
combining a fuzzy logic switching strategy and a direct MRAC al-
gorithm, the stability was proved based on the Lyapunov approach,
the strength of which was illustrated by applying to a single-link
manipulator for position tracking. The simulation results showed
that the designed multiple MRAC could perform very positively at
different operating modes. In Su (2007), a model reference tracking
based adaptive PID control system was put forward through intro-
ducing a PID control system to the feedback path for robot motion
196 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
control, and the control system enables us to maneuver the mo-
tions of an unstable robot. In Suboh, Rahman, and Arshad (2009),
a fuzzy MRAC was proposed by introducing the Takagi-Sugeno-
fuzzification control and P-I control to the MRAC. It was applied to
a 2-dof planar manipulator that is used underwater. Takagi-Sugeno
fuzzification was employed for the fuzzy system and the P-I update
law was employed in the adjustment mechanism to obtain a fast
parameter adaptation. The Lyapunov approach was used for the
stability analysis. Through simulation, it was found that the robotic
manipulator’s joints asymptotically followed the desired trajecto-
ries, which is defined by the reference model, in spite of the fact
that the manipulator had outer disturbances and input variations.
In Zhang and Wei (2016), a PID-MRAC hybrid control system was
designed by considering the hyper-stability theorem for multiple
degrees of freedom serial mechanical mechanisms via incorporat-
ing a PID control system and MRAC. Through simulations and ex-
periments, it was found out that the convergence speed and its
performance for the PID-MRAC control system is better than that
of the MRAC and PID controllers. Based on the studies in Sharifi et
al. (2014), Huh and Bien (20 07), Kamalasadan (20 04), Kamalasadan
and Ghandakly (20 08), Su (20 07), Suboh et al. (2009), and Zhang
and Wei (2016), it can inspire us that some approaches for new
controller design for robotic manipulators can be realized by com-
bining MRAC and other control system to design advanced MRAC
system, which is worth exploring in the near future.
In Park and Cho (2004), a model reference adaptive fuzzy con-
trol system was designed for the multi-input and multi-output
plant model for controlling an uncertain flexible joint mechanism,
and the control system was proved stable by resorting to the Lya-
punov approach. The novelty of the above study is that by intro-
ducing the fuzzy modelling and fuzzy model based controller into
the MRAC, the new controller design can achieve the trajectory
tracking effectively in spite of parameter perturbation and manip-
ulator model uncertainty. In Vempaty, Cheok, and Loh (2009), the
authors applied the MRAC to actuators of a robotic biped walker
to guarantee the actuators being able to generate the ideal torques
for the walking robot. The stability was ensured based on the
Lyapunov technique. The above study is merely a case study. In
Jurkovic (2004), the MRAC that uses only the joint position of the
robot manipulator in the adaptation law, the MRAC that uses both
joint position and velocity of the robot manipulator in the adapta-
tion law and the output feedback control are compared in detail.
The general results showed that the MRAC that uses both joint po-
sition and velocity of the robot manipulator in the adaptation law
outperforms the other two algorithms when noise and uncertainty
are considered. In Alqaudi, Modares, and Ranatunga (2016), a new
type of MRAC was presented by including a neural network for the
application of human-robot interaction. The control system con-
tains two control loops, one is an inner loop, it is used to learn the
robot dynamics and make the robot act as a prescribed impedance
model, another one is a task-specific outer loop, which takes the
human dynamics into account and adjusts the robot impedance
model in order to achieve the desired characteristics for human-
robot task performance. In Tar, Kovacs, and Takacs (2014), a vari-
ant MRAC which is based on robust fixed point transformations
was presented for the usage of a master–slave tissue cutting sys-
tem. The advantage of the variant MRAC system is that it does not
need to use the complicated Lyapunov 2nd method. In Zachi, Hsu,
and Lizarralde (2006), a novel MRAC was designed for robot mech-
anisms to conduct 3D Cartesian tracking without requiring depth
measurement. The control system was developed in a way to in-
clude the common situation of robotic systems in which every sin-
gle image feature can be extracted from a spherical object. The
stability was demonstrated based on the Lyapunov approach. In
Tsai, Wang, and Lin (20 0 0), a fuzzy model-following control sys-
tem was designed for controlling a planar 2-DOF robot manipula-
tor. The output of the unknown robotic manipulator and the ref-
erence model are the input to a self-tuning mechanism, the out-
puts of which are given to the fuzzy system, while the output of
the fuzzy system is the input of the unknown robotic manipula-
tor system. The control system was designed in such a way that
the adaptation process can be able to make the output tracking er-
ror converge to a residual set. In Chiou and Huang (2005), a MRAC
fuzzy sliding mode control system was developed to cope with im-
plementing in a multivariable robotic system. The control law was
created from the MRAC fuzzy sliding mode control system out-
put. An on-line parameter tuning algorithm was proposed based on
the Kalman-Yacubovich lemma, and the stability of the controller
was guaranteed by resorting to the Lyapunov approach. The pro-
posed control system was also compared with the SISO adaptive
fuzzy sliding mode control system for controlling a 5-DOF robot.
In Goléa, Goléa, and Benmahammed (2002), a fuzzy MRAC for con-
tinuous time MIMO nonlinear systems was proposed and used in a
two-link robot. The adaptation method employed a Takagi–Seguno
fuzzy adaptive controller, which can include a priori information,
and the P-I update rule was employed to have a fast adaptation
rate. The system was proved stable based on the Lyapunov sta-
bility analysis. In Xiao, Li, and Liu (2012), a MRAC PID controller
was developed based on the simple approximated model to con-
trol a compliant flexures-based XY micro-positioning stage. The ad-
justment mechanism of MRAC components was designed based on
the MIT rule. In Tung, Wang, and Hong (20 0 0), a compliant control
system was designed based on the MRAC for robotic manipulators.
When the robotic manipulator conducted compliant motion, the
path was made from the control system on the basis of the tan-
gential direction of the calculated contact force. The adaptive con-
trol law was employed to satisfy environmental constraints so that
the robotic manipulator end-effector can travel alongside the con-
tour of the surrounding workpiece and not lost contact with the
surface. In Al-Olimat and Ghandakly (2002), a MRAC algorithm us-
ing fuzzy logic adjustment for the reference model parameters was
designed and used in robotic manipulators. The algorithm was de-
signed in a way in order to deal with complex environmental con-
ditions that the robotic manipulator could possibly go through. The
intention of the designed control system is to guarantee that the
control system parameters can be changed automatically so that
they match with the present robotic manipulator environment. In
Tran, Ge, and He (2016), the control issue for an uncertain robotic
mechanism with input saturations and unknown input scaling was
studied by resorting to the MRAC. The MRAC was employed to
handle the input saturations, unknown input scaling was rejected
by the non-regressor procedure, and the robotic manipulator’s un-
certain dynamics were managed by the known function regressor.
In Yao and Wang (2012), the MRAC was used for dealing with
the underwater robotic manipulator nonlinear dynamics and hy-
drodynamics problem. The control system made the joint motion
on-line track the reference model output and against parameter
variations. In Doghiem (2014), the MRAC was applied to a 2-DOF
flexible joint robotic manipulator for position control under the
cases where uncertainties in parameters and load disturbances ex-
ist, and stability was presented on the basis of the Lyapunov tech-
nique. The above is a case study. In Santhakumar and Kim (2014),
a tracking control system for underwater vehicle-manipulators was
designed based on the MRAC. The designed adaptation law was
employed to approximate the parameters that are not known and
compensate for the remaining manipulator dynamics. The manip-
ulator mass’s influence on the control behaviour was illustrated
through the adaptive control strategy. This adaptive control strat-
egy allows one to prevail over the hard situation due to param-
eter uncertainties, payload variations and underwater current. In
TungLe, Leeand, and Lee (2011), an adaptive global asymptotic sta-
ble control method was designed for compensating the friction and
 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
disturbance effects on the robotic manipulators. The control ap-
proach combined the MRAC and exact linearization. In Cho and Seo
(2007), a direct MRAC based fuzzy control for the MIMO Takagi-
Sugeno model whose parameters are unknown was presented and
applied to a 2-DOF two-link robot manipulator. The adaptive law
calculates the fuzzy logic system error in order to achieve the
goal that asymptotic tracking of the reference signal for the sys-
tem with uncertainty can be provided, and the plant state fol-
lows the reference model state for bounded reference input sig-
nal. In Garg (2001), the MRAC was compared with the traditional
proportional-velocity control and linear quadratic regulator track-
ing control for 2-DOF two link robotic manipulators. The MRAC can
provide stable performance while the proportional-velocity control
cannot keep tracking errors when the robotic system experienced
inertia changes. The MRAC met design requirements with only par-
tial robotic system knowledge, and the linear quadratic regulator
tracking control is easier to implement than MRAC.
In Dias, Queiroz, Araujo, and Dias (2016), a control system that
consists of a left inverse system and a variable structure MRAC
for robotic manipulators was presented. The MRAC structure was
applied for each decoupled joint of the robotic system in order
to control the double integrator, it can also provide robust per-
formance in regards to external disturbance, and the MRAC con-
trol component only needs input torque and output joint position
knowledge. In Alinia, Hemmatian, Xie, and Zeng (2015), the pos-
ture control of the moving platform for a three degree-of-freedom
parallel robotic mechanism by employing the feedback lineariza-
tion control and the MRAC was investigated and compared be-
tween these two controllers. The simulation outcomes showed that
the MRAC performs better in the aspect of robustness than that
of the feedback linearization control when noise exists. In Lee,
Lim, Park, and Kim (2004), a control method by employing Takagi-
Sugeno fuzzy based MRAC was developed to track a reference sig-
nal for flexible joint manipulators with uncertain parameters. The
adaptation mechanism adjusted the parameters of the controller to
achieve the goal that the output from the plant can keep tracking
the output from the reference model, but a large amount of com-
putation is required for this MRAC fuzzy control. In Santhakumar
and Kim (2011), a MRAC strategy was simply employed to com-
pensate the linear and angular position errors of an underwater
vehicle-robotic mechanism, however, the payload and interaction
among the vehicle and robotic were not considered in the MRAC
process. In Kim, Oh, and Shin (2001), by employing the MRAC, the
dynamic response of each joint of a 2-link under-actuated robotic
manipulator was predetermined without the exact system param-
eters knowledge, but the generated torque at the active joint was
subject to serious chattering. One possible way to address the
above issue, for example, can be resorting to the fuzzy sliding
mode. In Zhao, Liu, and Yang (2006), a diagonal recurrent cere-
bellar model articulation based MRAC was presented for the 6-
PRRS parallel robotic manipulator trajectory tracking in order to
make the robotic system attain high precision positioning perfor-
mance. A discrete-type Lyapunov function was employed to calcu-
late the control system learning-rate parameters. However, the de-
signed control algorithm implementation needs fast control hard-
ware to be installed.
3. Conclusion
This study reviews the model reference adaptive control of
robotic mechanisms initially raised by the scholar Horowitz and
its later development by other authors. Some issues of MRAC for
robotic manipulators are also presented. Very few recent papers
can be found in the area of MRAC of robotic manipulators, and
this paper can provide general information in the field of MRAC
for robotic manipulators.
197
Acknowledgements
The authors would like to acknowledge the financial support
from the Natural Sciences and Engineering Research Council of
Canada (NSERC) and the Canada Research Chairs (CRC) program.
References
Alinia, S., Hemmatian, M., Xie, W., & Zeng, R. (2015). Posture control of 3-DOF paral-
lel manipulator using feedback linearization and model reference adaptive con-
trol. In Proceeding of the IEEE 28th Canadian conference on electrical and computer
engineering (pp. 1145–1150). May 3-6.
Al-Olimat, K., & Ghandakly, A. (2002). Multiple model reference adaptive control
algorithm using on-line fuzzy logic adjustment and its application to robotic
manipulators. In 37th IAS annual meeting. Conference record of the industry appli-
cations conference (pp. 1463–1466).
Alqaudi, B., Modares, H., & Ranatunga, I. (2016). Model reference adaptive
impedance control for physical human-robot interaction. Control Theory and
Technology, 14(1), 68–82.
Amerongen, J. (1981). MRAS: Model reference adaptive systems. Journal A, 22(4),
192–198.
Ampsefidis, L., & Bialasiewicz, J. (1993). Lyapunov design of a new model reference
adaptive control system using partial a priori information. Kybernetika, 29(4),
337–348.
Avinashe, K., Akhil, J., Dhanoj, M., & Shinu, M. (2015). Design of adaptive controller
for a level process. International Journal of Scientific and Engineering Research,
6(4), 1830–1833.
Babu, S. S., & Lincon, S. A. (2013). Design and implementation of model reference
adaptive controller using coefficient diagram method for a nonlinear process.
Journal of Electrical and Electronics Engineering, 6(4), 70–76.
Bi, Z., Liu, Y., & Baumgartner, B. (2015). Reusing industrial robots to achieve sustain-
ability in small and medium-sized enterprises. Industrial Robot: An International
Journal, 42(3), 264–273.
Bolourchi, F., & Hess, R. (1992). Nonlinear model reference adaptive control using
tap-delay filters. IEEE Transactions on Systems, Man and Cybernetics, 22(2).
Chiou, K., & Huang, S. (2005). An adaptive fuzzy controller for robot manipulators.
Mechatronics, 15(2), 151–177.
Cho, Y., & Seo, K. (2007). A Direct adaptive fuzzy control of nonlinear systems with
application to robot manipulator tracking control. International Journal of Con-
trol, Automation, and Systems, 5, 630–642.
Craig, J. J., Hsu, P., & Sastry, S. S. (1986). Adaptive control of mechanical manipu-
lators. In Proceedings of the 1986 IEEE international conference on robotics and
automation.
Dias, S., Queiroz, K., Araujo, A., & Dias, A. (2016). Robust control of robotic ma-
nipulators based on left inverse system and variable structure model reference
adaptive control. International Journal of Adaptive Control and Signal Processing,
30, 1389–1407.
Doghiem, H. (2014). Higher performance adaptive control of a flexible joint robot
manipulators. Journal of Mechanical and Civil Engineering, 11(2), 131–142.
Donalson, D., & Leondes, T. (1963). A model referenced parameter tracking tech-
nique for adaptive control systems. IEEE Transactions on Applications and Indus-
try, 82(68), 241–252.
Dubowsky, S., & Desforges, D. (1979). The application of model-referenced adaptive
control to robotic manipulators. Journal of Dynamic Systems Measurement and
Control, 101, 193–200.
Garg, A. (2001). Adaptive and optimal tracking control of electromechanical servo sys-
tems Master thesis. Queen’s University.
Goléa, N., Goléa, A., & Benmahammed, K. (2002). Fuzzy model reference adaptive
control. IEEE Transactions on Fuzzy Systems, 10(4), 436–444.
Hamadi, J. (1989). Adaptive control methods for mechanical manipulators: A compara-
tive study Master thesis. Naval Postgraduate School.
Horowitz, R. (1983). Model reference adaptive control of mechanical manipulators PhD
thesis. University of California.
Horowitz, R., & Tomizuka, M. (1986). An adaptive control scheme for mechanical
manipulators—compensation of nonlinearity and decoupling control. Journal of
Dynamic Systems, Measurement, and Control, 108(2), 1–9.
Horowitz, R., Tsai, M. C., & Anwar, G. (1987). Model reference adaptive control of a
two axis direct drive manipulator arm. In Proceedings of 1987 IEEE international
conference on robotics and automation (pp. 1216–1222).
Hsia, T. (1986). Adaptive control of robot manipulators – A review. In Proceedings of
1986 IEEE international conference on robotics and automation (pp. 183–189).
Huh, S., & Bien, Z. (2007). Robust sliding mode control of a robot manipulator based
on variable structure-model reference adaptive control approach. IET Control
Theory & Applications, 1(5).
Joseph, A., & Isaac, J. (2013). Real time implementation of model reference adap-
tive controller for a conical tank. International Journal on Theoretical and Applied
Research in Mechanical Engineering, 2(1), 57–62.
Jung, B., Jeong, S., & Lee, D. (2005). Adaptive reconfigurable flight control system
using multiple model mode switching. In Proceedings of the 16th IFAC world
congress: 16 Part 1.
Jurkovic, D. (2004). Output feedback adaptive control for robot manipulators Master
thesis. Carleton University.
Kamalasadan, S. (2004). A new generation of adaptive control: An intelligent supervi-
sory loop approach PhD thesis. The University of Toledo.
198 D. Zhang, B. Wei / Annual Reviews in Control 43 (2017) 188–198
Kamalasadan, S., & Ghandakly, A. (2004). A neural network based intelligent model
reference adaptive controller. In CIMSA 2004 - IEEE international conference on
computational intelligence for measurement system and applications (pp. 174–179).
Kamalasadan, S., & Ghandakly, A. (2008). A novel multiple reference model adaptive
control approach for multi modal and dynamic systems. Control and Intelligent
Systems, 36(2), 119–128.
Kamalasadan, S., Ghandakly, A., & Al-Olimat, K. (2004). A fuzzy multiple reference
model adaptive control scheme for flexible link robotic manipulator. 2004 IEEE
international conference on computational intelligence for measurement systems
and applications.
Kamnik, R., Matko, D., & Bajd, T. (1998). Application of model reference adaptive
control to industrial robot impedance control. Journal of Intelligent and Robotic
Systems, 22, 153–163.
Kim, M., Oh, S., & Shin, J. (2001). Robust model reference adaptive control of under-
actuated robot manipulators. In Proceedings of IEEE international symposium on
industrial electronics (pp. 1579–1584).
Kirecci, A., Topalbekiroglu, M., & Eker, I. (2013). Experimental evaluation of a model
reference adaptive control for a hydraulic robot: A case study. Robotica, 21,
71–78.
Krishna, K., Kumar, J., & Shaik, M. (2012). Design and development of model based
controller for a spherical tank. International Journal of Current Engineering and
Technology, 2(4), 374–376.
Landau, I., Lozano, R., Mohammed, M., & Karimi, A. (2011). Introduction to adaptive
control. Adaptive control, communications and control engineering. Springer-Ver-
lag London Limited.
Landau, Y. (1979). Adaptive control-the model reference approach. CRC Press.
Lee, J., Lim, J., Park, C., & Kim, S. (2004). Adaptive model reference control based
on Takagi-Sugeno fuzzy models with applications to flexible joint manipulators.
KSME International Journal, 18, 337. doi:10.1007/BF02996099.
M, Tomizuka, & Horowitz, R. (1983). Model reference adaptive control of me-
chanical manipulators. In IFAC adaptive systems in control and signal processing
(pp. 27–32).
Maeda, T., & Totani, T. (1990). An experimental study on model reference adaptive
control for robot manipulators. Advanced Robotics, 5(1), 25–38.
Maliotis, G. (1991). A hybrid model reference adaptive control/computed torque
control scheme for robotic manipulators. Proceedings of the Institution of Me-
chanical Engineers, Part I: Journal of Systems and Control Engineering, 205(3),
215–221.
Neuman, C., & Stone, H. W. (1983). MRAC control of robotic manipulators. In
K. S. Narendra (Ed.), 3rd Yale workshop on applications of adaptive systems theory
(pp. 203–210). New Haven, CT: Yale Univ.
Oltean, S. M., Dulau, M., & Duka, A. V. (2016). Model reference adaptive control de-
sign for slow processes: A case study on level process control. Procedia Technol-
ogy, 22, 629–636.
Park, C., & Cho, Y. (2004). Adaptive tracking control of flexible joint manipulator
based on fuzzy model reference approach. IEEE Proceedings, Control Theory and
Application, 150(2), 198–204.
Prakash, R., & Anita, R. (2011). A new approach to model reference adaptive con-
trol using fuzzy logic controller for nonlinear systems. International Journal of
Computer Science and Information Security, 9(2), 86–93.
Prakash, R., & Anita, R. (2011). Design of model reference adaptive intelligent con-
troller using neural network for nonlinear systems. Journal of Electrical Engineer-
ing, 1–8.
Ram, A. G., & Lincoln, S. A. (2013). A model reference-based fuzzy adaptive pi con-
troller for non-linear level process system. International Journal of Recent Re-
search and Applied Studies, 14(2), 477–486.
Sadegh, N. (1987). Adaptive control of mechanical manipulators: Stability and robust-
ness analysis PhD thesis. University of California.
Sadegh, N., & Horowitz, R. (1987). Stability analysis of an adaptive controller for
robotic manipulators. In Proceedings of 1987 IEEE international conference on
robotics and automation (pp. 1223–1229).
Sadegh, N., & Horowitz, R. (1990). Stability and robustness analysis of a class of
adaptive controllers for robotic manipulators. International Journal of Robotics
Research, 9(3), 74–92.
Santhakumar, M., & Kim, J. (2011). Modelling, simulation and model reference adap-
tive control of autonomous underwater vehicle-manipulator systems. In 2011
11th international conference on control, automation and systems (pp. 643–648).
Oct. 26-29.
Santhakumar, M., & Kim, J. (2014). Robust adaptive tracking control of autonomous
underwater vehicle-manipulator systems. Journal of Dynamic Systems, Measure-
ment, and Control, 136(5), 10.
Schwartz, H. (1994). Model reference adaptive control for robotic manipulators
without velocity measurements. International Journal of Adaptive Control and Sig-
nal Processing, 8(3), 279–285.
Sharifi, M., Behzadipour, S., & Vossoughi, G. (2014). Model reference adaptive
impedance control in Cartesian coordinates for physical human–robot interac-
tion. Advanced Robotics, 28(19), 1277–1290.
Srinivasan, R. (1987). Adaptive control for robotic manipulators Master thesis. Carleton
University.
Su, W. (2007). A model reference-based adaptive PID controller for robot motion
control of not explicitly known systems. International Journal of Intelligent Con-
trol and Systems, 12(3), 237–244.
Suboh, S., Rahman, I., & Arshad, M. (2009). Modeling and control of 2-DOF under-
water planar manipulator. Indian Journal of Marine Sciences, 38(3), 365–371.
Sutherland, J. (1987). Model reference adaptive control of a two link manipulator Mas-
ter thesis. Carleton University.
Tam, R., & Thomopoulos, S. (1992). Model reference adaptive control of robotic manip-
ulators using a modified output error method, 10, 143–150.
Tar, J., Kovacs, L., & Takacs, A. (2014). Novel design of a model reference adaptive
controller for soft tissue operations. In 2014 IEEE international conference on sys-
tems, man, and cybernetics, San Diego, CA, USA (pp. 2476–2481).
Tomizuka, M., & Horowitz, R. (1988). Implementation of adaptive techniques for mo-
tion control of robotic manipulators. Journal of Dynamic Systems, Measurement,
and Control, 110(1), 62–69.
Tomizuka, M., Horowitz, R., & Anwar, G. (1986). Adaptive techniques for motion con-
trols of robotic manipulators. In Japan - USA symposium on flexible automation
(pp. 117–224).
Tomizuka, M., Horowitz, R., & Anwar, G. (1986). Adaptive techniques for motion con-
trols of robotic manipulators. Japan - USA symposium on flexible automation.
Tomizuka, M., Horowitz, R., & Teo, C. L. (1985). Model reference adaptive controller
and robust controller for positioning of varying inertia. In Proceedings of confer-
ence on applied motion control (pp. 191–196). University of Minnesota.
Tran, T., Ge, S., & He, W. (2016). Adaptive control for an uncertain robotic manipu-
lator with input saturations. Control Theory and Technology, 14(2), 113–121.
Tsai, C., Wang, C., & Lin, W. (20 0 0). Robust fuzzy model-following control of robot
manipulators. IEEE Transactions on Fuzzy Systems, 8(4), 462–469.
Tung, P., Wang, S., & Hong, F. (20 0 0). Application of MRAC theory for adaptive con-
trol of a constrained robot manipulator. International Journal of Machine Tools
and Manufacture, 40(14), 2083–2097.
TungLe, H., Leeand, S., & Lee, C. (2011). Integration model reference adaptive control
and exact linearization with disturbance rejection for control of robot manipula-
tors. International Journal of Innovative Computing, Information and Control, 7(6),
3255–3267 June.
Ulrich, S., & Sasiadek, J. (2010). Direct model reference adaptive control of a flexi-
ble joint robot. 2010 AIAA Guidance, Navigation, and Control Conference. Toronto,
Canada.
Valente, A., Carpanzano, E., & Brusaferri, M. (2011). Design and implementation of
distributed and adaptive control solutions for reconfigurable manufacturing sys-
tems. CIRP sponsored ICMS. International conference on manufacturing systems.
Valentea, A., Mazzolinib, M., & Carpanzanoa, E. (2015). An approach to design and
develop reconfigurable control software for highly automated production sys-
tems. International Journal of Computer Integrated Manufacturing, 28(3), 321–336.
Vempaty, P., Cheok, K., & Loh, R. (2009). Model reference adaptive control for actu-
ators of a biped robot locomotion. In Proceedings of the world congress on engi-
neering and computer science: Vol II. San Francisco, USA: WCECS.
Visioli, A. (2006). Practical PID control. London: Springer.
Whitaker, H., Yamron, J., & Kezer, A. (1958). Design of model reference adaptive control
systems for aircraft, report R-164, instrumentation laboratory. Cambridge, Mas-
sachusetts: M. I. T. Press.
Wilson, E., & Rock, S. (1995). Reconfigurable control of a free-flying space robot us-
ing neural networks. In Proceedings of the 1995 American control conference: 2
(pp. 1355–1359).
Xiao, S., Li, Y., & Liu, J. (2012). A model reference adaptive PID control for electro-
magnetic actuated micro-positioning stage. In 2012 IEEE international conference
on automation science and engineering (CASE) (pp. 97–102).
Yang, S. (1992). Discrete-time model reference adaptive controller design for robotic
manipulators Master thesis. Northern Illinois University.
Yang, S., Woo, P., & Wang, R. (1993). Discrete-time model reference adaptive con-
troller designs for robotic manipulators. In 1993 American control conference, CA,
USA (pp. 1145–1149).
Yang, S., Woo, P., & Wang, R. (1996). Discrete-time model reference adaptive con-
troller designs for robotic manipulators. The International Journal of Robotics Re-
search, 5(3), 280–289.
Yao, J., & Wang, C. (2012). Model reference adaptive control for a hydraulic under-
water manipulator. Journal of Vibration and Control, 18(6), 893–902.
Yuh, J., & Holley, W. (1988). Application of discrete-time model reference adaptive
control to industrial robots: A computer simulation. Journal of Manufacturing
Systems, 7(1), 47–56.
Zachi, A., Hsu, L., & Lizarralde, F. (2006). Adaptive control of nonlinear visual
servoing systems for 3D Cartesian tracking. Sba Controle & Automação, 17(4),
381–390.
Zhang, D., & Wei, B. (2016). Design, analysis and modelling of a hybrid controller for
serial robotic manipulators. Robotica., 1–18. doi:10.1017/S02635747160 0 0564.
Zhao, J., Liu, Y., & Yang, Y. (2006). A diagonal recurrent CMAC model reference
adaptive control for parallel manipulators trajectory tracking. In Proceedings of
the sixth international conference on intelligent systems design and applications
(ISDA’06) (pp. 157–161). doi:10.1109/ISDA.2006.253824.