IRJET- Singular Identification of a Constrained Rigid Robot

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 08 | Aug 2018 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 941
Singular Identification of a Constrained Rigid Robot
Hossein Sharifi1
1PhD student, Dep. of Electrical and Computer Engineering, Lamar University, TX, USA
---------------------------------------------------------------------***------------------------------------------------------------------
Abstract - In this paper, a singular system identification
procedure based on Strong equivalency is proposed for a
constrained robot singular model. In fact, since constrained
robot model is considered as a singular system, it requires to
be transformed to an equivalent form before applying
identification on it. This equivalent model plays an
important role in the identification algorithm, since an
inappropriate equivalency causes large identification error
and sometimes identification divergence. The Strong
equivalency applied in this paper keeps the important
dynamics of the robot model while reducing the number of
singular system’s initial conditions. The Recursive Least
Square identification algorithm is also applied in the next
step on the Strong equivalent system. Indeed, Recursive
Least Square identification procedure based on strong
equivalent model is implemented for the first time for this
constrained robot singular model, and can be generalized to
all constrained robot models of this kind. Performance of
this method on the robot is also validated by illustrating the
simulations results of the technique on a robot singular
model. According to the results, the model identification
error convergence is improved significantly and also the
output tracking is more satisfactory compared to the
previous identification techniques for the constrained robot
model.
Key Words: Singular system identification,
constrained robot, Strong equivalency, Recursive Least
Square identification.
1. INTRODUCTION
Robot manipulators are widely being used in various
industrial engineering areas and therefore, robot systems
studies are under a significant attention by the scholars.
Robot arm movement can be considered constrained or
unconstrained. Unconstrained case happens when the
robot arm moves in a free space without interacting with
the environment. Constrained motion of robot case
happens when the robot end-effectors interact
mechanically with the environment. However, in practical
tasks, the mechanical interactions of the robot arm with
the environment or with the object being manipulated
need to be considered. Therefore, in manufacturing
environment, the constrained robot model is of greater
importance than the unconstrained model [1].
Indeed, constrained robot model identification and
estimation is regarded as an important issue in robotics
since the robot parameters are not easily available (they
are not usually provided by the robot manufacturers) and
they are not directly measurable in practice due to the
structural complexity and payload of robot manipulators
[2], [3]. A large number of theoretical and experimental
algorithms are done to deal with the identification of robot
model parameters in this area and they resulted with
different levels of accuracy. Although, in most of these
algorithms, some dynamics of the robot are neglected in
the process which is the main drawback of these
approaches. So, considering that the constrained robot
model is a singular model, the identification approach is
regarded as a singular identification method.
Singular systems identification is of immerse interest
among researchers, since these systems describe the
dynamics of extensive number of systems such as power
systems, electrical networks, chemical processes,
economical systems, robotic systems, mathematics, neural
networks, etc [4-8]. As a matter of fact, singular systems
(also known as generalized state space systems) present a
larger class of systems than the normal state space
systems, and therefore singular theoretical properties had
been studied as an active research topic by many
researchers in the last decades, in [9-11]. Indeed,
investigations on singular systems are being performed in
two main areas; differential-algebraic equations theory of
singular systems, and singular systems control theory. As
far as the main focus in this paper is the singular
equivalency and identification of robots, the researches in
the singular systems theory are emphasized.
Actually, singular equivalency is considered as a crucial
issue in singular systems studies, because the first step in
all techniques dealing with this kind of systems is to find
and equivalent regular state space model for them.
Singular equivalency has been introduced as reduction
methods for several years. These reduction methods have
been widely studied in theory of singular matrices in [12].
Later, Polak introduced an algorithm to reduce a
differential system to a linear time independent state form
[13]. Moreover, Fettweis, Desoer and Dervisoglu explained
a method for reducing the state equation of algebraic -
differential systems in circuit theory [14], [15]. In parallel,
Luenberger proposed an algorithm for the state equation
reduction of singular discrete systems. Besides, [16-22]
have made great progress in the field of controllability,
observability and stability of singular systems [23]. All the
studies mentioned used the regular theory to attain a
result. The drawback of the regular theory is that it keeps
the original system’s dynamics and ignore the infinite
impulse modes of the system. Neglecting the infinite
modes is an important shortcoming for the constrained
robot singular model, whereas it owns important infinite
modes that by neglecting them, the system would be
identified with low accuracy. Based on Strong equivalency
model developed in [24], the singular identification
through Least Square or any other on-line algorithms is

improved significantly. This is also proved in another
study on an electrical singular system in [23].
In this regard, the parameters identification of the
constrained robot singular model is performed in this
paper using Recursive Least Square algorithm as the
identification method and Strong equivalent model as the
equivalency approach. The objective is to show that the
minimum estimation error and the best tracking of the
parameters are obtained. The results of applying the
proposed approach on the constrained robot model are
found satisfactory in comparison with the previous robot
identification methods. Moreover, by using the proposed
approach for this model, the difficulties in singular
identification are handled to some degrees.
The rest of the paper is organized as follows: in section
2 the constrained robot model and its specifications are
presented. Section 3 describes the Strong equivalency
approach for singular systems. The Recursive Least Square
estimation algorithm is briefly explained in section 4, and
section 5 contains the proposed method simulations and
numerical results on the constrained robot model.
Eventually, the work is finished by the conclusion in
section 6.
2.CONSTRAINED ROBOT MODEL AND ITS
SPECIFICATIONS
Without considering friction and other disturbances,
the dynamics of an n-link constrained rigid robot
manipulator can be described as follows:
( ) ̈ ( ̇) ( ) (1)
In the above equation, denotes the vector of
generalized displacements in robot coordinates,
denotes the vector of generalized constraint forces in
robot coordinates, and denotes the vector of
control inputs in robot coordinates. ( ) is the
symmetric positive definite manipulator inertia matrix,
( ̇) is the vector of centripetal torques and
( ) is the vector of gravitational torques. It is
assumed that the robot is non-redundant and equipped of
joint position and velocity sensors and a force sensor at its
end-effector [25-27].
From a proper definition of matrix ( ̇), ( ), and
( ̇) in equation (1), the following equation is attained.
( ̇ ) (2)
So, ̇ is a skew-symmetric matrix.
The dynamic equation (1) can be expressed as linear in
terms of a suitable selected set of robot and load
parameters.
( ) ̈ ( ̇) ( ) ( ̇ ̈) (3)
where ( ̇ ̈) and are the vectors
including the unknown manipulator and load parameters.
Also, a positive constant exists in the following
equation.
( ) (4)
such that is the identity matrix. Matrix ( )
exists and is positive definite and bounded.
Because of the existence of the impulsive modes in the
model, the robot system is considered as a singular
system. One robot arm model can be linearized around the
operating point, to get the following linear singular state
space general formulation as its model representation.
( ) ̇( ) ( ) ( ) ( ) ( ) ( ) (5)
( ) ( ) ( ) ( )
In the above equation, ( ) is the output vector of length .
( ) is a vector of system state variables with dimension ,
and ( ) is an dimensional vector of input. , , , are
real matrices, considering that is the singular matrix.
is also the regression parameter vector of unknown
parameters of the system matrices. ( ) and ( ) are
process and output noises that are not considered in this
work.
The impulsive modes cause dependent state space
equations. Dependent states generate a big trouble in
identification process because this will result in initial
conditions in the algebraic differential equations and it is
required to estimate the initial conditions throughout the
identification procedure. Thus, estimating the initial
conditions randomly may result in high identification
error or even divergence.
3.STRONG EQUIVALENCY
To overcome the problem of initial conditions
mentioned in the previous section, the number of initial
conditions can be reduced; in other words, the number of
dependent state equations can be reduced. For this aim,
the original model needs to be transformed to an
equivalent reduced model in the first step before applying
identification algorithm. So, this section is mainly focused
on describing the Strong equivalency approach on a
singular system.
Most of the singular equivalency methods reduce the
model to an equivalent model without considering the
infinite dynamics of the model. Since most of these
approaches are developed based on the regular theory,
they are not the most efficient equivalencies to be chosen
for the singular model [12-22]. Based on these

approaches, the singular system is transformed to the
following equation [23].
̇ ( ) ( ) ( ) (6)
( ) ( ) ( ) ( )
with and ( ) as the polynomial part of G(s).
Therefore, the reduction method deletes some
information of the system which generates huge errors in
the original system identification process.
By the new Strong equivalency which is based on the
Canonical Kronecker form of the ( ), this problem is
settled. The new equivalency is based on the Restricted
System Equivalency, introduced by Rosenbrock [20]. The
Canonical Krocker form of ( ) can be described
based on the non-singular matrices M and N as follows.
( ) [
̃
] ,
[ ] , [ ] (7)
is the model regular degree of freedom and ̃ is the
nilpotent matrix with the singular index equal to
. Therefore, the state vector is divided into two parts;
regular and singular subsystem state vectors respectively,
as follows.
( ) [
( )
̃( )
]
(8)
The singular system definition (5) is then converted to
its Laplace form as follows.
[ ] [
( )
( )
]=[
( )
( )
]
(9)
From the Restricted System equivalency of [19],
equation (9) can be converted to equation (10) as the
equivalent model.
[ ̃ ̃
̃
] [
̅( )
̃( )
( )
] [
̅( )
̃ ̃( )
( )
]
(10)
In the above equation, ̃ and are the sub blocks of ,
and ̃ and are the sub blocks of .
Thus, in this way the original system is divided into
two sub systems and all the properties of the original
system are kept in the equivalent system.
Later, this approach is modified by the Strong
equivalency, since there is a need to estimate the two sub
systems parameters separately and it is inconvenient.
Based on the Strong equivalency, the Laplace
description of the model (9) can be converted to the
following model.
[ ] [ ] [ ] [ ] (11)
In the above equation, .
However, this transformation imposes extra
constraints (Q and R matrices) over the equivalency
method. This strong equivalency is more convenient
because this method provides a more integrated model
rather than two separate sub models as in equation (10),
and it is a significant superiority of this equivalency in
identification applications [29].
In this paper, the Strong equivalency described in this
section is used in the first step before applying Recursive
Least Square identification algorithm on the robot system.
4.IDENTIFICATION
The well-known Recursive Least Square (RLS) is
chosen as the identification algorithm in this work. This
algorithm is an adaptive filter that recursively updates to
find the coefficients that minimizes the weighted least
squares cost function. This algorithm is proved to
converge fast and this is the main reason for adopting it in
this research.
Here, the main stages associated with RLS for
identification of the strong equivalent model of equation
(11) are described:
First: Updating the unknown vector of parameters;
̂( ) ( ) ( )( ( ) ( ) ( ))
(12)
Second: Reconstructing the Gain matrix ( );
( ) ( ) ( ) ( ) ( )( ( ) ̂( ))
(13)
Third: Updating the Covariance matrix;
( ) ( ) ( ) ( )( ( ) (
) ( )) ( ) ( )) ( ( ) ( )) ( ))
(14)
The parameters summary used in the above algorithm
are as follows.
̂( )

( )
In this algorithm, the initial conditions for covariance
matrix and estimated parameters vector are defined in the
first stage. Then, P, K, and ̂( ) are updated in each
iteration through the above formulations until the
algorithm converges. Initial conditions of the states are
also involved in this algorithm, because the error of
estimation is required to be calculated in each stage
considering the output values. In the next section, it is
shown that the results of RLS on the Strong equivalent
model are satisfactory.
5.SIMULATIONS AND RESULTS
The original formulation (1) of the robot manipulator
is linearized around its operating points through MATLAB
code and the following model is attained in the form of a
singular system. Also, consider that the model is attained
for one arm of the robot [28].
[
( )
] [
̈
̇
] [
( )
]
(15)
So, with the aid of Strong equivalency, model of
equation (15) can be converted to its equivalent form by
the following process.
By choosing , ̇, ̈ as the singular system states, the
singular system model matrices A, B, C, E are stated as
equation (16).
( ) [ ] [ ]
[ ] [ ] (16)
( ) [ ̇
̈
]
With the aim of attaining the equivalent model of the
system, the following calculations for analyzing the modes
of the system are done.
[ ] [ ]
[ ] [ ]
So from the above ranks, it is resulted that the
impulsive modes of the system which arise because of the
dependent states are neither controllable nor observable.
These cause problem during the identification because of
the initial conditions arising.
From equation (11), the following equivalent model is
resulted for this system (the suffix SE in the equation (17)
is short for Strong equivalency).
( ) [ ] [ ]
[ ] [ ]
Also, note that in the above equation the state variables
change; the new state variables are ̇, ̈, and ̇ ̈.
Therefore, the initial conditions of the state become zero
in this case, and there is no need to estimate the initial
conditions value while applying the identification
algorithm.
Parameters of the Strong equivalent model are then
identified through the 3-steo identification algorithm of
section 4. The number of inputs and outputs used for the
identification process is 600, with P matrix equal to 1000I.
The initial states values are chosen as zero regarding the
Strong equivalency property which is independent of
initial values.
The following results are attained for eight model
parameters estimation. Also, the estimation error results
are shown in the continuation.
Fig -1: First four model parameters; identified values
and real values using the proposed method

Fig -2: Second four model parameters; identified
values and real values using the proposed method
From the results of Fig -1, it can be realized that the
first four identified parameters of the system converged
soon after less than 3 seconds to the real values. However,
by zooming the figure it is obvious that the estimated
values fluctuate around the real value with very small
fluctuations and eventually converges exactly to the real
point after less than 30 seconds. The same thing happened
also for the second four model parameters in Fig -2. This is
considered as a significant improvement in robot model
identification compared to the previous works. Moreover,
using Strong equivalency in identification process through
the proposed method, the estimation convergence is
guaranteed, which is not done in previous studies. Also,
computation time of the simulations is significantly less
than identification process using previous singular
equivalencies.
Fig -3: Estimation error using the proposed method
Fig -4: Model real and identified outputs using the
proposed method
According to Fig -3, the estimation error of the system
output is not even large at the beginning of the
identification. This estimation error converges to exactly
zero after around 37 seconds. Also, in Fig -4 the estimated
system output tracks the real system output with high
accuracy level. The estimated output even merges with the
real output after some seconds, at time 35.
Overall, from the results, it can be realized that the
estimation error for the robot model is significantly small
and the identification results are satisfactory. In fact, the
estimated parameters of the system are attained with a
high accuracy close to the original parameters very fast.
6.CONCLUSION
In this paper, the problem of identification of a
constrained rigid robot arm is studied. The constrained
robot is a singular system and therefore the singular
equivalency method is applied in the first stage of the
identification algorithm. The proposed method in this
work is the Strong equivalency combined with the
Recursive Least Square algorithm. The simulations on the
robot model are performed in MATLAB and the resulted
figures proved the efficiency of the proposed approach.
The benefit of this approach compared to the previous
identifications on this robot model is that the Strong
equivalency in this work keeps all the system dynamics
without any loss of information in the system. Meanwhile,
the number of non-zero initial conditions is reduced in the
equivalent model, which is the main objective of all the
equivalency approaches. Therefore, the estimation
accuracy is high with low estimation error. Besides, the
identified parameters of the model and its estimated
output converged exactly to real values after short
computation time.

REFERENCES
[1] Yan, D., Lu, Y., and Levy, D. 2015. Parameter
Identification of Robot Manipulators: A Heuristic Particle
Swarm Search Approach. PLoS ONE, 10(6): e0129157. DOI:
10.1371/journal.pone.0129157.
[2] Carelli R., and Kelly R. December 2016. ADAPTIVE
CONTROL OF CONSTRAINED ROBOTS MODELED BY
SINGULAR SYSTEMS. the 28th Conterence on Decision and
Control. Tampa, Florida.
[3] Eini R., and Abdelwahed S. 2018. Indirect Adaptive
fuzzy Controller Design for a Rotational Inverted
Pendulum, 2018 Annual American Control Conference
(ACC), Milwaukee, WI, USA, pp. 1677-1682.
doi: 10.23919/ACC.2018.8431796
[4] Chou J. H., and Liam W. H. 2009. Stability robustness of
continuous-time perturbed descriptor systems. IEEE
Transactions on Circuits and Systems, 46(9), 1153-1155,
[5] Dai L. Singular control systems. 2002. Lecture Notes in
Control and Information Sciencs, Springer Verlag, Berlin.
[6] Campbell S L.1980. Singular Systems of Equations.
Research Notes in Mathematics. New York, Pitman.
[7] Lewis F. L. 1986. A survey of linear singular systems.
Circuits Syst., Signal Processing, 5(1), 3-36.
[8] Newcomb R. W., Dziurla B. 1989. Some Circuits and
Systems Applications of Semistate, Theory, Circuits,
Systems and Signal Processing, 8(3): 235-260.
[9] Rosenbrock H. H. Structural properties of linear
dynamic systems, Int. J. Control, 20(2), pp: 191-202.
[10] Luenberger D. G. 1977. Dynamic equations in singular
form. IEEE Trans. on Automatic Control, 22(3): 312-321.
[11] Ishihara J. Y., and Terra M. H. On the Lyapunov
theorem for singular system, IEEE Transactions on
automatic control, 47(11), pp:1926-1930.
[12] Gantmatcher F. R. 1959. The theory of matrices, vols.
1 and 2, Chelsea, New York.
[13] Polak E. 1966. An algorithm for reducing a linear,
time-invariant differential system to state form. IEEE
Tarns. Automat. Contr., vol. AC-11, pp. 577-579.
[14] Fettweis A. 1969. On the algebraic derivation of the
state equations. IEEE Trans. Circuit Theory, vol. CT-16, pp.
171-175.
[15] Dervisoglu A., and C. A. Desoer C. A. 1985. Degenerate
networks and minimal differential equations. IEEE Trans.
Circuit and Syst., vol. CAS-22, pp. 769-775.
[16] Rosenbrock H. H. 1970. State-space and multivariable
theory. New York, Wiley.
[17] Pernebo L. 1977. Notes on strict system equivalence.
Int. J. Contr., 25(1), pp. 21-38.
[18] P. A. Fuhrmann, “On strict system equivalence and
similarity,” Int. J. Contr., vol. 25, no. 1, pp. 5-10, 1977.
[19] B. Levy, S-Y. Kung, M. Morf, and T. Kailath, “A
unification of system equivalence definitions,” in Proc.
IEEE Conf. Decision and Control (New Orleans, LA), pp.
795-801, 1977.
[20] H. H. Rosenbrock, “Structure properties of linear
dynamical systems,” Int. J. Contr., 1974.
[21] Cobb D. 1984. Controllability, observability and
duality in singular systems. IEEE Trans. Automat. Contr.,
vol. 29, pp. 1076-1082.
[22] Verghese G. C., Levy B. C., and Kailath T. 1981. A
generalized state space for singular systems. IEEE Trans.
Automat. Contr., vol. 26, pp. 811-831.
[23] Eini R., and Ranjbar Noei A. 2013. Identification of
Singular Systems under Strong Equivalency. International
Journal of Control Science and Engineering, 3(3): 73-80
DOI: 10.5923/j.control.20130303.01.
[24] Verghese G. C., Levy B. C., and Kailath T. 1981. A
generalized state space for singular systems. IEEE Trans.
Automat. Contr., vol. 26, pp. 811-831.
[25] Ortega R., and Spong M. 2008. Adaptive motion
control of rigid robots: a tutorial. Proc. of the 27th. IEEE
CDC, Austin, TX.
[26] Slotine J. J., and Li W. 2007.Adaptive manipulator
control: a case study. Proc. of the IEEE Int. Conf. on Robotics
and Automation, Raleigh, C.N.
[27] Slotine J.J., and Li W. 1987. Adaptive strategies in
constrained manipulator. Proc. of the IEEE Int. Conf. on
Robotics and Automation, Raleigh, C. N.
[28] Eini R. 2014. Flexible Beam Robust H-infinity Loop
Shaping Controller Design Using Particle Swarm
Optimization. Journal of Advances in Computer Research,
5(3), Quarterly pISSN: 2345-606x eISSN: 2345-6078.
[29] RakhtAla S. M., Eini R. 2014. Nonlinear Modeling of a
PEM Fuel Cell System; a Practical Study with Experimental
Validation. International Journal of Mechatronics,
Electrical and Computer Technology, 4(10), pp. 1272-
1296, ISSN:2305-0543.

IRJET- Singular Identification of a Constrained Rigid Robot

More Related Content

What's hot (17)

Similar to IRJET- Singular Identification of a Constrained Rigid Robot (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Singular Identification of a Constrained Rigid Robot