Sunteți pe pagina 1din 38

Accepted Manuscript

Adaptive position/force control for robot manipulator in contact with a


flexible environment

Piotr Gierlak, Marcin Szuster

PII: S0921-8890(16)30561-9
DOI: http://dx.doi.org/10.1016/j.robot.2017.05.015
Reference: ROBOT 2856

To appear in: Robotics and Autonomous Systems

Received date : 15 September 2016


Revised date : 4 May 2017
Accepted date : 29 May 2017

Please cite this article as: P. Gierlak, M. Szuster, Adaptive position/force control for robot
manipulator in contact with a flexible environment, Robotics and Autonomous Systems (2017),
http://dx.doi.org/10.1016/j.robot.2017.05.015

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
Title. Adaptive position/force control for robot manipulator in contact with a flexible
environment
Authors. Piotr Gierlaka,1,*, Marcin Szustera,2
Affiliation. aRzeszow University of Technology, Department of Applied Mechanics and
Robotics, al. Powstańców Warszawy 12, 35-959 Rzeszów, Poland.
E-mail address: 1pgierlak@prz.edu.pl, 2mszuster@prz.edu.pl
*
corresponding author.

Abstract. The subject of the article is the adaptive position and force control of a robotic
manipulator in interaction with flexible environment. The aim of the study is to provide a
solution that takes into account the essential aspects of operation of the manipulator with the
environment and at the same time can be actually implemented. A manipulator-environment
system model taking into account motion resistance and environment elasticity. The position
and force control task has been defined considering the manipulator and environment models.
Asymptotic stability of the control system has been demonstrated considering the adaptation
of parameters of the manipulator and the environment. Practical stability of the system has
been demonstrated in the case of interference with the guaranteed stability of the adaptation of
parameters without requiring persistence of excitation. Numerical analysis and experimental
study of the issue has been presented.

Keywords. manipulator modelling, flexible environment, adaptive control, position/force


control, stability
1. Introduction
Nowadays, manipulators are becoming more and more frequently applied in the broadly
defined industry, which requires the realisation of a desired motion path of the manipulator
end-effector with simultaneous control of the forces of interaction with the environment. It is
no longer just about performing of assembly tasks - which was still the objective a few years
ago [1, 2]. Currently, attempts are being made to provide a precise control of forces in order
to ensure proper execution of processes such as grinding, chamfering, polishing, etc. [3, 4, 5].
The presented problem of contact between the manipulator and the environment is associated
with partial restrictions of the manipulator's motion, i.e. the existence of constraints. An
important and still current issue is the modelling of such manipulator-environment system and
its control.
The literature shows many different approaches to the modelling of manipulators that
work with the environment. The model selection is determined mainly by its application and
dominant phenomena. This article is focused on modelling for control purposes. In the case of
control-oriented models, usually the most important occurring phenomena are modelled, and
attempts are made to achieve a mathematical structure of the model that enables the
demonstration of the control system's stability. Three approaches are generally distinguished
when it comes to manipulator modelling: modelling of manipulator as a system of rigid
bodies without flexibility and backlash in joints [6], modelling that takes into account the
flexibility and/or backlash in the joints [6, 7, 8, 9] and modelling that takes into account the
flexibility of bodies [6, 8, 10]. Regarding the modelling of the environment operating with the
manipulator, there are also multiple approaches. Usually, the surface of contact between the
manipulator and the environment is introduced, and the environment characteristics are
modelled in selected directions related to the surface [8]. It is assumed that motion of the
manipulator end-effector is possible on the surface of contact along tangential directions
while applying pressure on the surface in normal directions. One of the following variants is
used in modelling of interactions in tangential directions:
- model ignoring motion resistance,
- model considering motion resistance.
In the case of normal directions, interaction is usually modelled with taking into account the
selected variants:
- rigid environment model [11, 12],
- model that takes into account only the environment flexibility [12],
- model that takes into account the environment flexibility and damping [8],
- model that takes into account the environment flexibility, damping and mass [8, 13].
Sometimes, the problem of lack of knowledge of the environment or its uncertainty is
considered [14, 15, 16, 17].
An important aspect is to determine the model of contact between the manipulator and
the environment. Usually, point contact between the manipulator end-effector and the
environment surface is adopted; contact in the form of a line or surface of contact is rarely
taken into account. In addition, it is assumed whether the contact is continuous, or whether
there are phases of coming into and out of contact.
Therefore, the modelling problem itself is complex. The most commonly used among
the discussed solutions is the model of manipulator as a rigid body system without flexibility
and backlash in the joints, whereas its infinite rigidity is assumed in the environment model or
only its most important feature (flexibility) is taken into account. This is reflected in actual
systems existing in the industry, where the manipulator is usually a very rigid system and the
environment is also rigid or flexible. The approach in which the manipulator and the
environment are treated as rigid is relatively well-known and widespread [11, 18, 19, 20,21],
in the world of both science and application. On the other hand, the approaches that take into
account the selected environment characteristics in the manipulator control rather not go
beyond academic considerations and require popularization. This is partly caused by the fact
that the practical implementation of such solutions requires the knowledge of parameters of
environment, which can be troublesome.
Most of papers on the control of interaction force between the manipulator and the
environment use the approach presented in the fundamental work [1]. It proposes that the
interaction force derivative considered in the control system should be determined
according to an equation where ke is the environment stiffness and is its
deformation rate. This approach is supported by the fact that in robotics the quality of velocity
measurements is relatively good, while the differentiation of the measured signal of force is
troublesome because of the noise in the signal. The disadvantage of this approach is the need
for knowledge of the environment stiffness (ke), which may be variable and is very difficult to
determine in real applications. As described by the author [1], if the stiffness is very high, the
control system parameters may be selected in such a way that it is insensitive to the
inaccuracies in the knowledge of the value of ke. However, if the environment stiffness is
relatively low and may be variable, such an approach becomes invalid. In addition, in real
applications of force control systems there are usually vibrations, because of which the
deformation rate values are high and oscillating. This may be the case when machining thin-
walled elements, for example - then the introduction of the derivative of force to the control
system is troublesome in general.
One of solutions to the problem of lack of knowledge of the environment stiffness may
be an attempt to adapt the coefficient ke [15, 22], but the properties of adaptive systems show
that the initial phase of adaptation is characterised by the biggest adaptation errors. In the case
of incorrectly specified stiffness and the associated interaction force, the system may operates
improperly or even be destroyed. Therefore, the above-mentioned papers contain only a
theoretical analysis of the issue and simulation studies, and are not supported by experimental
results. Often the force derivative is not considered at all in a control system, and the
interaction force is determined by measuring the manipulator's position [23].
An interesting approach to the question has been presented by the authors of the article
[24] in which a practical application of position-force control with the use of adaptive control
is discussed. The aim of the aforementioned article, similarly to the one at hand, is to describe
the cooperation of the manipulator and the flexible environment of unknown stiffness.
Nonetheless, two significant differences are observed. The first one is connected with the fact
that in the article in question the control compensating non-linearity is based on the
manipulator model devoid of adaptation of parameters. Thus it is possible that the difficulty
with estimating the object parameters may arise. The second difference is that the authors of
the aforementioned article use PID control in the feedback force loop in order to achieve the
convergence of the force error, whereas in the reviewed manuscript the compensatory control
for control in a normal direction towards the contact surface, which compensates the influence
of environment interaction on force control was used.
Taking into account historical advancement of robotics, in the writings published even
before the 90s [1] it was advised to determine the forces of interaction on the basis of the
measurement of displacement of the end-effector and arbitrarily assumed stiffness.
Implementation of such an approach in robotisation of processes, in which high precision is
necessary, requires also a detailed knowledge of location of the contact surface and
calibration of manipulator kinematics parameters. Nowadays, bearing in mind the
advancement of measurement systems, there is a more favourable approach based on a direct
measurement of forces/moments of forces applied to the end-effector of the manipulator. Due
to a relatively good quality of the signal generated by force sensors, it may be said that the
statement that measurements of forces are flawed with great errors and measurement noises,
which used to be true in the beginnings of robotics, has lost its relevance. Moreover, the
possibilities of force signal filtering have increased and arriving at a force signal differential,
inputted in the control signal, is not a serious practical problem. Basing on the article [25] it is
claimed that the majority of modern manipulators with force control packages are based on
the measurement of force with the use of force sensors in the end-effector, more rarely in the
joints. The approaches with force control option in which the force would be determined by
the measurements of the position of the end-effector of the manipulator are not present in the
writings on this subject matter. This confirms in practice the replacement of indirect force
measurement systems by direct force measurement systems. Such a force measurement does
not require adopting an environment model, which like every model is flawed by errors and
modelling uncertainty.
This paper proposes the second approach, which uses a direct measurement of the
interaction force, thus ensuring proper feedback in the force control loop and enabling safe
operation of the system. Control quality has been examined in the case of both occurrence and
lack of the interaction force derivative in control law.
This article focuses on providing a solution that takes into account the essential
aspects of operation of the manipulator with the environment and at the same time can be
actually implemented. Therefore, the manipulator model is assumed to be a system of rigid
bodies and the environment model takes into account the motion resistance in tangential
directions and flexibility in normal directions (Section 2). Section 3 discusses the position and
force control task while taking into account the manipulator and environment models. The
first considered example is an ideal case, when the manipulator is not affected by
interferences. Asymptotic stability of the control system has been demonstrated considering
the adaptation of parameters of the manipulator and the environment. Then the paper presents
a variant which takes into account active interference and demonstrates practical stability of
the control system with the guaranteed stability of the adaptation of parameters without
requiring persistence of excitation. Section 4 is dedicated to the presentation of a numerical
example, which illustrates the operation of the control system. Section 5 contains and
discusses the results of experimental studies. The paper is summarized in Section 6.

2. Manipulator and environment model


Dynamics equation of motion of the manipulator in the joint space takes the form [26]
, (1)
where qRn - the vector of generalised coordinates, M(q)Rnn - the inertia matrix,
, ∈ - the vector of centrifugal and Coriolis forces (moments), ∈ - the
n n
viscous friction vector, G(q)R - the gravity vector, ξ(t)R - the vector of disturbances
bounded by ||ξ||≤b, b>0, uRn - the control input vector, J(q)Rmn - an analytical Jacobian
matrix, λRm - an interaction force vector expressed in the task space, n - the number of
degrees of freedom of the manipulator, m – a workspace (task space) dimension.
The manipulator’s workspace is associated with the Cartesian coordinate system. The
kinematics of the manipulator in the Cartesian coordinates is described by the function
∈ (2)
The relationship between velocity in the Cartesian space and in joint space is as follows
(3)
where the analytical jacobian J is
(4)
On the basis of (3) the correlation between acceleration takes the form
(5)
Dependencies of (1), (3), (4) and (5) permit to represent the dynamics of the manipulator in
the space of the Cartesian coordinates {C}. It will be a convenient form of description of the
hybrid position-force control taking into account the characteristics of the environment with
which the robot interacts.
For this purpose, equation (1) was premultiplied by (JT)-1, in order that the interactive
force λ may be expressed directly in the Cartesian coordinates:
, (6)
where was taken into account that (JT)-1=(J-1)T which was written in short form as J-T. On the
basis of (3) and (5) it was determined
(7)
(8)
and substituted into equation (6), which took the form of
,

(9)
Assumed designations

, , , ∈
, ∈ (10)
, ∈

and the dynamics of the manipulator were expressed in the Cartesian space in the form of [6,
21]
, , , (11)
Divide the m-dimensional space of the Cartesian coordinates {C} to subspace {T} and {N}:
⨁ [13]. Subspace {N} is an r-dimensional space of directions ni normal to the
contact surface of the manipulator’s end-effector and the environment, whereas subspace {T}
is (m-r)-dimensional space of tangent directions τi. Therefore the vector of coordinates c can
be written as
(12)
where cτRm-r, cnRr. Assuming that the environment, with which the manipulator enters
contact (through the end-effector) is characterized by elasticity, then the said environment
feature is present in the subspace {N}. The environment will be described by the equation:
(13)
rr
where: KeR - diagonal matrix of the environment’s stiffness, that meets the dependency
.
In the tangential direction, the movement of the manipulator’s end-effector over the
contact surface will be accompanied by resistance. It is modeled with dry friction. In the case
of translation in the contact plane the elements of the vector Fe τ will take the form:
. . . (14)
where: Fe τ(.), Fe n(.) – are appropriate components of tangential and normal force, μ – the
coefficient of dry friction, vτ(.) – the velocity of movement along the direction τ(.). However,
in the instance of rotation in the contact plane the elements of vector Fe τ will take the form:
. . . (15)
where: Fe τ(.), Fe n(.) – are appropriate components of the tangential and normal force, in this
instance Fe τ(.) will be the momentum of resistance forces, μZ – substitute coefficient of dry
friction, ωn(.) – angular velocity of rotation in the contact plane, that is the turnover around the
axis n(.).
In relation to the above dissertations, the interaction force vector can be written as

(16)

where: Fe τRm-r - vector of tangential forces, Fe nRr - vector of normal forces.
Assuming, for simplification, that on a system described by equation (11)
interferences hold no effect, meaning Ψ(q,t)=0. Then we will write down the dynamics model
of the manipulator in the Cartesian space as
, , (17)
On the basis of the environmental description (13) we receive:
(18)
where ∈ is the environmental flexibility matrix, which is the diagonal matrix,
hence the dependence of arises. Taking into account (18) in (12), and then in
the equation (17), we obtain
, , (19)

or
, , (20)
where
∈ (21)

∈ (22)
Equation (20) describes the dynamics of the system in Cartesian space as a function of
movement parameters in the tangent plane and the forces on the normal directions.

2.1. Model properties


Structural properties of the mathematical model of the system, which are used in the analysis
of the system stability are as follows [6]:
Property 1: The inertia matrix A(q) is a positive definite symmetric matrix, provided that J is
a full-rank, which verifies
(23)
where and ∞ denote the strictly positive minimum and maximum eigenvalues
of A(q) for all q.
Property 2: The matrix , verifies
‖ , ‖ ‖ ‖ (24)
where 0 is some bounded constant.
Property 3: The matrix , verifies
‖ , ‖ ‖ ‖ ‖ ‖ (25)
where 0, 0 are some bounded constants.
Property 4: The vector , verifies
‖ , ‖ (26)
where 0 is some bounded constant.
Property 5: The matrix 2 , is skew-symmetric, provided that the matrix
, 2 , is skew-symmetric.
Proof: In the dynamics of rigid manipulators, a property is shown [6] stating that the matrix
, can be selected in such a way as to
, 2 , (27)
was a skew-symmetric matrix. On this basis, will be now shown the same property of the
matrix 2 , . By performing the differentiation matrix A(q) and
considering equation (10) we can write
2 , 2 ,

2 ,

2 , (28)

where the use is made of the property that ,


T
and M(q)=M(q) since the inertia matrix is a symmetric matrix. Expression (28) can be
written as
2 , 2 , (29)
since
(30)
Truth of the expression (30) can be demonstrated taking the equation
(31)
which after differentiating with regard to time gives:
(32)
-1
and after right-hand multiplication of (32) by J equation (30) is obtained. Let us now
premultiply every element of equation (29) by ET and postmultiply by E. We then receive
2 , 2 , , (33)
T -T
Multiplication of the skew-symmetric matrix , left-handedly by the matrix E J and
-1
by right-handedly by the matrix J E also yields a skew-symmetric matrix. Therefore, the
matrix 2 , is a skew-symmetric matrix.

3. Adaptive tracking control


The general approach used to derive the adaptive control algorithm is based on methods
known from the literature on rigid control of manipulators [27]. However, due to the nature of
the description of the system taking into account the characteristics of the environment,
known algorithms can not be applied directly, but must be significantly modified.
The proof of stability in the sense of Lyapunov for adaptive control is widely discussed in
the writings on this subject matter [28, 29, 30, 31], yet due to the specifics of the described
system the proof of stability causes many difficulties. The specifics of the problem is an
attempt to execute control both by means of position and force on the basis of the theory of
tracking control; moreover, additional coupling between the variables of the state of the
system takes place through forces of interaction with the environment, which is not stiff, but
flexible. In the case of a stiff environment the situation is more simple and there is a
possibility to divide the controller into a part controlling the position and a part controlling the
force, but assuming a stiff environment does not always correspond with reality and may lead
to big control errors. In the case of low stiffness of the environment it is necessary to take it
into consideration with regard to the model and control laws. Yet another challenge taken is
an attempt to use the theory of compensatory control not only to control the motion in
tangential directions, but also to control the force in normal directions. This allows for
compensating the influence of environment on the executed control of force. Yet another
assumption which makes the requirements more strict is assuming the lack of knowledge on
the stiffness of the environment and remaining parameters of the system, which gives rise to
the neceessity of usiing adaptivee control. Inn the literatuure studies the
t authors have not met m such
an approoach to thiss problem, leet alone thee proof of sy ystem stability.
For a system deescribed by equation (220) a limited d trajectory θ(t)Rm, , , must
m be
considered, which can be referrred to as a generalized d trajectory, since it connsists of a
movemeent trajectorry in a tangent plane, hhence cτd(t) Rm-r, , , aas well as th
he force
trajectorry in normaal directionss ∈ , , . Definingg the trackin ng error
as:
(34)
and a filltered trackking error
(35)
where
(36)
is a deriivative of thhe tracking error
e and Λ =ΛT>0 - design matrix, ΛRn. Equuation (35),, with
the use of (36), makkes it possib ble to transfform to the equation
(37)
where
(38)
is an auuxiliary variaable. Differrentiating (337) we receiive
(39)
Formulaas (37) and (39) we willl take into account in the t equation n (20) obtaiining
, , , (40)
Assuming the desiggnation
, , (41)
where ffRm is the function
f describing thee nonlineariity of the system. In thee end, the
descripttion of the system,
s in teerms of the filtered traccking error, takes the foorm of:
, (42)
Here apppears the fuunction f, whose
w structuure results from
f the mo
odel of the m manipulatorr and the
environnment. It also depends on o the impreecisely know wn parametters of the m manipulatorr as well
as the uunknown stifffness of thee environm ment. For succh a describ bed system, control sho ould take
into acccount a convventional PD D control, ccontrol comp pensating foor nonlinearrities of thee object
∈ as well as a control co ompensatingg for the inffluence of fo orces λ:
(43)
where approximaates f. The term t KDs is a form of PDP control, where
w 0,
KDRmm.

Fig. 1. T
The generall scheme of the closed--loop system
m

As clearrly follows from the sccheme, nonllinearity of the


t system and
a the influ
luence of
interactiion forces (lambda)
( aree compensaated by the inner
i ol loop, whille the outerr loop
contro
stabilizees error by using
u a PD controller.
Nonnlinear function can be written if thhe form of
(44)
where


(45)


Nonlinear functions fτ and fn described by formula (45) are linear in the parameters, that is
why they can be written in the following form:
, , , , , (46)
, , ,

, , , , , (47)

, , ,
In equations (46) and (47) the matrixes , , , and , , , are
bτ bni
the so called regression matrixes, pτR and pniR are the vectors of unknown parameters,
bτ, bni – parameter space dimensions, i=1,…,r. Function approximation of fτ and fn by
appropriately and depends on the estimation of unknown parameter vectors, which we
write as
, , ,
, , ,

(48)
, , ,

, , ,
where
, , , (49)
, , , (50)
and ∈ , ∈ are evaluations of vector parameters pτ and pni. Taking into account
the control law (43) as well as the formulas (44), (46), (47) and (48), in equation (42), we
obtain
, , ,
, , ,

, (51)
, , ,

, , ,
where
(52)
(53)
are errors of parameter estimates. Equation (51) is a closed-loop system description in terms
of the filtered tracking error and the parameter estimation error.

3.1. Asymptotic stability of the adaptive control


In the following section the asymptotic stability of the adaptive control tracking the desired
trajectory is presented.
Theorem 3.1: A limited position-force trajectory θ(t)Rm, , is given and no
disturbances are acting on the manipulator. Control in the form of (43) for the system (20) is
set and the vector of the filtered tracking error is decomposed in the following way
(54)
where sτR(m-r), sn=[sn1 … sni … snr]TRr. The parameters adaptation law is determined as
follows
, , , (55)
, , , (56)
where 0, 0 are the design gain matrixes. Suppose the system is
persistently exciting. Then the closed-loop system is asymptotically stable, i.e. s→0 when
t→∞.
Proof: Let’s assume the function:
∑ (57)
where Peii is an element of the flexibility matrix. Differentiating (57) with regard to time we
obtain
∑ (58)
Considering the equation (51) we can write
, , ,
, , ,

2 ,
, , ,

, , ,
∑ (59)
T
Given that E =E, third element of equation (59) we write as
, , , , , ,
, , , , , ,
⋮ ⋮
(60)
, , , , , ,
⋮ ⋮
, , , , , ,
Expression 2 , is a skew-symmetric matrix (c.f. Property 5), therefore, a
dependency arises
2 , 0 (61)
Considering the (54), (60) and (61), we write
, , ,
, , ,

∑ (62)
, , ,

, , ,
and further
, , , ∑ , , ,
(63)
Considering adaptation law (55) and (56) in (63) we obtain
(64)
In order to demonstrate that s→0 when t→∞, it will be sufficient to prove that → when
t→∞. Based on the Barbalat lemma [32] we can state the function → if it is uniformly
continuous, that is, the second derivative with respect to time is limited. The second derivative
is
2 (65)
and is limited, if the vectors s and are limited. Since V≥0 and , then the function V is a
Lyapunov function and is limited. From this it can be seen that vector s is limited, so then the
tracking error and its derivative are limited. The parameter estimation errors and
are also limited, thus the parameter estimation vectors and are limited.
The matrix A(q) is symmetrical and positive definite, hence an inverse matrix A(q)-1 exists,
whereas the inverse matrix of E is
(66)
so from equation (40) results
, , ,
, , ,

, (67)
, , ,

, , ,
The right side of equation (67) is limited and, consequently is also limited. That is why is
limited, → when t→∞ and s→0 when t→∞, which implies that → and → when
t→∞. The closed-loop system is asymptotically stable.

3.2. Practical stability of the adaptive control


Subsection 3.1 demonstrates the asymptotic stability of the system. This result is elegant, but
impossible to achieve from a practical point of view. The reason is interference affecting the
controlled object, for example. For this reason, this subsection will present the approach that
demonstrates the so-called practical stability [6]. Another problem in adaptive algorithms is
that persistency of excitation is generally needed for suitable performance of such systems
[33, 34, 35]. The possible unboundedness of the parameter estimates when persistency of
excitation fails to hold is known as “parameter drift.” To avoid the persistency of excitation
requirement a modified tuning algorithm of the parameter estimates is used [18, 33, 34, 35].
The proof of stability in this case is more complicated, but the result is more useful and makes
it possible to estimate the accuracy of the control system.

Theorem 3.2: A limited position-force trajectory of the manipulator θ(t)Rm, , is


given and limited disturbances are acting on the system (20)
,
, (68)
,
where Ψτ(q,t)R(m-r), Ψn(q,t)Rr, and bτ is a known constant such that ||Ψτ(q,t)||≤bτ and bni is a
known constant such that |Ψni(q,t)|≤bni. Control in the form of (43) for the system (20) is set
and the vector of the filtered tracking error is decomposed according to the equation (54).
Assume the parameters are bounded by known values
‖ ‖ (69)
‖ ‖ (70)
The parameters adaptation law is determined as follows
, , , ‖ ‖ (71)
, , , | | (72)
where 0, 0 are the design gain matrixes, kτ>0 and kni>0 are design
parameters. Make no assumptions of any sort of persistent excitation condition. Then the
filtered tracking errors and and parameter estimates errors and are uniformly
ultimately bounded with practical bounds given respectively by the right-hand sides of (85)-
(88). Moreover, the tracking errors may be made as small as desired by increasing the
elements of gain .
Proof: Let us return to the equation (20) and consider dynamics of the manipulator with taking
into account interferences Ψ(q,t). Further analysis was performed in the same way as for the
system without interruption to yield a description of a closed-loop in the form
, , ,
, , ,

, , (73)
, , ,

, , ,
The difference in comparison to the description of the undisturbed system (51) lies in the fact,
that in equation (73) the term Ψ(q,t) appears. In order to prove the stability of the closed-loop
system the function (57) and its derivative (58) we assume. Considering the equation (73) into
(58) we can write
, , ,
, , ,

2 , ,
, , ,

, , ,
∑ (74)
Considering equations (54), (60), (61) and (22), we write
,
, , ,
, , ,


, , ,

, , ,
(75)
and further

,
, , ,
,
∑ , , , (76)
where the matrix KD was replaced by submatrixes
(77)
with ∈ , ∈ . Considering adaptation law in (76) we obtain
, , ‖ ‖
∑ | | (78)
and note that
‖ ‖‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ (79)
‖ ‖‖ ‖ ‖ ‖ ‖ ‖
‖ ‖ (80)
Therefore, the equation (78) was converted into the following form

‖ ‖ | | ‖ ‖ | |

‖ ‖ ‖ ‖ ‖ ‖ ∑ | | ‖ ‖ ‖ ‖ (81)
where KDτmin is the minimum singular value of KDτ. Formula (81) can be converted into the
following form
‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ∑ | | | |
‖ ‖ ‖ ‖ (82)
The function  is negative as long as the terms in square brackets are positive. These
expressions can be written as

‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖ ‖

(83)
| | ‖ ‖ ‖ ‖ | |
‖ ‖
(84)
They will be positive if

‖ ‖ ≡ (85)

| |
≡ (86)
or

‖ ‖
≡ (87)

‖ ‖
≡ (88)
Thus, is negative outside a compact sets defined by (85)-(88). According to a standard
Lyapunov theorem extension [21, 33, 34], this demonstrates that ||sτ||, |sni|, ‖ ‖ and ‖ ‖ are
uniformly ultimately bounded and control system is stable. The values bsτ, bsni, bpτ and bpni,
may be taken as practical bounds of the filtered tracking error and parameter estimation
errors. It follows that the tracking error and its derivative are limited, as well as the
parameter estimation vectors and .
 
3.3. Conclusion
In the following section the stability of the manipulator control system considering the
interaction of the robot and its environment is shown. At first, asymptotic stability was
observed which constitutes a neat result, yet of little practical meaning, since in order to
demonstrate the stability, two strict assumptions were made: there are no disturbances in the
system and system is persistently exciting. Then the practical stability of the system is shown,
which is characterised by the fact that the tracking errors and estimated system parameters are
limited; moreover, the tracking errors aim at a defined environment, the size of which may be
regulated by setting the control system gain. By appropriate choice of parameters adaptation
law, the stability of the system was achieved without the need to maintain persistency of
excitation of the system. A different research aim was achieved as well, that is the control law
does noot require thee knowledg
ge of environnment stiffn
ness, which is taken intto account in
i the
estimateed parameteers.

4. Nummerical exam mple


As an exxample, let''s consider control
c of a two-link pllanar manippulator (Figuure 2) whosse
dynamiccs in joint coordinates
c is describedd by equatio
on (1), in whhich the maatrixes and vectors
v
are as foollows
(89)
0
, (90)
0
(91)

(92)

(93)

(94)
(95)
The parrameters in matrixes
m nd vectors o f equation (1)
an ( are as fo
ollows:

(96)

where: mi – a masss of ith link, li – a lengthh of ith link


k, lci – the diistance betwween center of mass
of ith linnk and end of i-1link, Ii – a mass m
moment of inertia
i of ith
h link relatiive to its cen
nter of
mass, cvvi – coefficieent of visco
ous friction in ith kinemmatic pair. The
T model pparameter values
are show wn in Tablee 1 (subsectiion 4.2).

Fig. 2. T
The schemee of the 2-deegrees-of-frreedom man
nipulator with environm
ment

The mannipulator ennd-effector'ss coordinatees in Cartessian space arre:


(97)
The anaalytical Jacoobian (4) forr the manipuulator is:
(98)
The equation describing the dynamics of the manipulator in Cartesian coordinates is (11),
where
(99)
with

(100)

, (101)
with

(102)

(103)

, (104)

, (105)


(106)

  Non-linear function f can be expressed as dependencies (46) and (47), with the following
for the case under consideration
1 0
(107)
0
(108)

1 0 (109)
0 1 (110)
Moreover, taking into account (99)-(110), we write the functions in the linear form in relation
to the parameters as , , , and , , , , where:
, , , (111)

(112)

, , , (113)

(114)
Auxiliary signal υ defined by equation (27) in the case under consideration is as follows
(115)

which takes into account equation (108) and Λ=diag{Λτ, Λn}.
The desired system trajectory should be designed in such a way as to maintain the
constraints imposed on the system by the environment. In the analysed case the surface of
contact with the environment is one-dimensional and at the same time it is the trajectory of
the manipulator end-effector. The trajectory is described by straight line equation
h(cτ,cn)=cn=0. The velocity vector for the end-effector is as follows
(116)
whereass the velocitty value is defined
d by tthe followin
ng
(117)
The velocity vectorr must be taangent to thee trajectory, and thereffore conditioon 0
must bee met, so thaat
0 (118)
Taking into accounnt that ∂h/∂ccτ=0 and ∂h//∂cn=1, the result of eq quation (1188) is 0, which
indicatees that the ennd-effector motion is nnot planned in the normmal directionn, and the en nd-
effectorr velocity inn the tangenttial directioon may be of any limiteed value. Thhe trajectory y will
meet thee specified condition iff the movem ment of the manipulator
m r end-effecttor in the tan
ngential
direction is for example

0 ∑ (119)

were cτdd(0) is the innitial positio


on of the ennd-effector, vmax is the maximum
m vvelocity, wτ is the
coefficient correspoonding to th he velocity iincrease and decrease rate,
r t(0.5 0) [s]. Funcction
(119) annd its first and
a second derivative
d inn relation to
o time are constant andd limited.
  The trajectory of o the system m in the norrmal directiion is the prressure forcce on the surrface of
contact,, which shouuld be contiinuous, limiited and nott negative, and a should hhave contin nuously
the firstt and secondd derivative in relation to time. Su uch conditions are met bby the adop pted
functionn
∑ (120)
where Fe nmax is thee maximum pressure foorce, wn is th he coefficient corresponnding to thee force
increasee and decreaase rate, t((0.50) [s]. Itt is obviouss that as a reesult of the fforce trajectory
realisatiion with thee existing fleexibility of the environnment, cond dition 0 is not striictly
met.
In F
Fig. 3 the deesired path (Fig.
( 3a), thee desired diisplacementt along the ttangential direction
d
(Fig. 3bb) and the deesired normmal interactioon force (Fiig. 3c) are shown.

Fig. 3. D
Desired trajectory of manipulator
m eend-effector: a) the dessired path, bb) the desireed
displaceement alongg the tangen
ntial directioon, c) the deesired normal interactioon force

The assiigned trajecctory is plannned in a waay which makes


m the end-effector oof the manippulator
push aggainst the suurface approopriately durring motion n, whilst theere is no preessure again
nst the
surface made whilee the end-efffector returrns. Therefo ore, three cy
ycles of mottion can be
distinguuished. Thannks to the ad daptation off the contro
ol system in each subseequent cyclee, the
quality of control should
s increease.

4.1. Casse studies


Control law (43) and parameter adaptation law (71) and (72) have been applied to achieve the
desired trajectory. Parameters used in the simulation study are shown in Table 1. In addition,
adaptation gain matrixes for parameter adaptation are assumed in the following form:
Γτ=diag[0.9, 6.75·10-6, 1.8, 1.53·10-5, 0.9, 6.75·10-6, 15, 15, 0.09, 0.09], and
Γn=diag[3·10-7, 2.25·10-10, 2·10-3, 4.5·10-11, 3·10-7, 2.25·10-10, 3·10-3, 3·10-3, 3·10-5, 3·10-5].

Table. 1. Parameters used in numerical tests


1 2 3 4 5 6 7 8 9
parameter unit value parameter unit value parameter unit value
2
a1 kgm 0,036 Λτ s-1 3 tF11 s 3
a2 kgm2 6·10-5 Λn s-1 34.6 tF21 s 16
2
a3 kgm 0,031 KDτ11 kg/s 8 tF12 s 3
a4 Nms 0,54 KDn11 s 0,00173 tF22 s 28
a5 Nms 0,51 kτ - 0,2 tF13 s 3
a6 Nm 0,05 kn1 - 0,2 tF23 s 41
a7 Nm 0,025 vmax m/s 0,03 tc11 s 5
l1 m 0,22 wτ s-1 5 tc21 s 10
l2 m 0,22 Fe nmax N -10 tc31 s 12
Ke11 N/m 1333 wn s-1 5 tc41 s 17
Ce11 kg/s 0.7 cτd(0) m 0,22 tc12 s 19
μ - 0.04 ̂ 0 - 0 tc22 s 24
̂ 0 - 0 tc32 s 26
tc42 s 31
tc13 s 33
tc23 s 38
tc33 s 40
tc43 s 45

Numerical tests were performed for different options, taking into account the instances of
disturbances acting on the system or the lack of them as well as the cases of modifying control
law. In Table 2 the tested case studies and short descriptions of differences which are
elaborated on in the subsequent part of the subsection are presented.

Table 2. Case studies


Case study
1 2 3 4
Varying - no disturbances - disturbances in - no disturbances - disturbances in
factors - PD controller in the form of - PD controller in the form of
the form of dampening the the form of dampening the
environment environment
- PD controller in - PD controller in
the form of the form of
Estimated bsτ=0,0035 bsτ=0,0035 bsτ=0,0035 bsτ=0,0035
accuracy of bsn1=16,1194 bsn1=20,1656 bsn1=16,1194 bsn1=20,1656
the system bpτ=0,7468 bpτ=0,7468 bpτ=0,7468 bpτ=0,7468
bpn1=0,7468 bpn1=0,7911 bpn1=0,7468 bpn1=0,7911

In case studies 1 and 3 is assumed that the system is not affected by interference, that is bτ=0
and bni=0, so the practical bounds of the filtered tracking error and parameters estimation
errors calculated baased on equuations (85)--(88) and daata from Taable 1 are bssτ=0,0035,
bsn1=16,,1194, bpτ=00,7468 and bpn1=0,74688.
In case studies 2 and 4 dammping of ennvironment was w taking into accounnt. It can be
includedd in the matthematical description
d of the systeem and treatted as a distturbance. Damping
may be written as resistance,
r which
w is lineear in veloccity of end-eeffector at thhe normal
direction. Then thee vector of disturbances
d s can be wriitten as:
0
, (121)
where Ce11 – the cooefficient off environmeent damping g. The damp ping coefficcient Ce11 is adopted
based onn experimenntal study. InI this case,, the follow
wing limitations of distuurbances aree
assumedd: ||Ψτ||=bτ=0
= and ||Ψn1||≤bn1= . Bassed on equattions (85)-( 88) and datta from
Table 1, practical bounds
b are bsτ=0,0035, bsn1=20,165 56, bpτ=0,74
468 and bpn11=0,7911.
To eevaluate thee quality of control, thee following quality ratinngs, writtenn in the geneeral
form, arre assumed
max | | (122)
∑ | | (123)
where x – selected variable, k - an index oof sample, N - a numbeer of samplees, xmax – maaximum
of x, xRM
MS – root me
ean square of
o x.

4.2. Ressults of nummerical simmulations


In this ssubsection selected
s resu
ults of numeerical simullations perfo formed in acccordance with
w four
case stuudies describbed in subseection 4.1 aare presented
d. Due to sm mall differennces in timee
courses of the signaals in the co
ontrol system m in particu
ular tests, th
he first one iis described
d
complettely, whereaas in the remmaining desscriptions on nly most rellevant differerences are
describeed.

4.2.1. C
Case study 1
Overall control signnals (Fig 4aa) are comp osed of PD control (Fig 4b), comppensatory co ontrol
(Fig 4c)), the normaal force conttrol λ2 and ffriction forcce compensaation λ1 (Figg. 4d).

Fig. 4. C
Control signnals associaated with thee workspacee: a) the oveerall controll signals, b)) the PD
control signals, whhere UPD1= KDτ11sτ, UPDD2= KDn11sn1 c) the comp pensatory coontrol, d) th he
normal force controol (λ2) and friction
f forcce compensaation (λ1)
In the innitial movem
ment phase,, compensattory controlls are mismaatched, becaause estimaates of
parametters are equual zero. Thee signals geenerated by the PD conttroller takess majority meaning
m
at the beeginning, annd then the influence oof the PD siggnals decreaases, becausse the parammeters
estimatees adaptatioon, and the meaning
m of the compen nsatory conttrol increasees. At the
beginninng the normmal force control fluctuuates (Fig. 4d) due to laack of enviroonment flex xibility
compennsation by thhe compenssatory controol. In Fig. 5 the realised trajectoryy and path of o end-
effectorr are presentted. Due to the flexibiliity, deformaation of env
vironment aalong normaal
direction occurs (Fig. 5b) and path of endd-effector (F Fig. 5c) does not satisfyy condition cn=0.

Fig. 5. T
The realisedd trajectory: a) displaceement of maanipulator end-effector
e r along the
tangentiial directionn, b) displaccement of m
manipulator end-effectoor along the normal direection,
c) path oof the end-eeffector, d) normal inteeraction forcce

The chaaracteristic feature


f of th
he adaptive system is thhat, in the in
nitial robot movement phase,
errors arre the biggeest, and then
n are successsively redu
uced, as a reesult of the pparameter estimates
e
adaptatiion. In the Fig.
F 6 error of displacem ment (Fig. 6a)
6 and velo ocity (Fig. 66c) and erro
or of
Fig. 6b) and its time derrivative (Figg. 6d) are prresented.
force (F
Fig. 6. T
The trackingg errors: a) error of dispplacement along
a the taangential dirrection, b) error
e of
normal interaction force, c) errror of veloccity along th
he tangentiaal direction, d) error of time
derivative of normaal interactioon force

In the ppresented sim mulation th on of the paarameters estimates


he worst-casse scenario in estimatio e
is takenn, i.e. their initial valu
ues are assuumed to zerro. Fig. 7 an
nd Fig. 8 ppresent estim
mates of
parametters of the system.
s All estimates arre bounded..

The parameeter estimatees ̂ for noonlinear fun


Fig. 7. T nction
The parameeter estimatees ̂
Fig. 8. T for nnonlinear fun
nction

For the practical immplementatiion of the coontrol system, the know wledge abouut control siignals
and trajectory in thhe joints spaace are requiired. In Fig. 9a overall control signnals in jointt space
are pressented, and in
i Fig. 9b and
a Fig. 9c aangles of lin nks’ rotation
n and angullar velocitiees of
links aree shown.

Fig. 9. C
Control andd motion in thet joint spaace: a) the overall
o conttrol signals - torques in
n joint,
b) anglees of rotatioon of links, c)
c angular vvelocities off links

Filteredd tracking errrors sτ, sn1 (Fig.


( 10a, bb) and param
meters estim
mation errorss , (F
Fig. 10c,
d) tend to determinned boundarries which cconfirms thee practical stability of tthe system.
Fig. 10. The analyssis of signaals boundednness: a) the norm of filtered trackiing error alo ong the
tangentiial directionn, b) the norrm of filtereed tracking error
e along the normal direction, c)
c the
norm off the parameeter estimattes error , d) the norm m of the parrameter estiimates errorr

4.2.2. C
Case study 2
The intrroduced distturbance affffects the boounds bsn1, and
a bpn1 and d slightly inffluences thee
waveforrm of the filltered tracking error sn11 (Fig. 11). All errors are
a boundedd, which praactically
proves tthat the systtem is stable. The charaacteristics of
o the achiev ved signal ccourses are very
similar tto those achhieved in the case studyy 1.

Fig. 11. The analyssis of signaals boundednness in the case


c of a disturbance: aa) the norm of
filtered tracking errror along th
he tangentiall direction, b) the normm of filteredd tracking errror
along thhe normal direction, c) the norm off the param meter estimattes error , d) the norm m of the
parametter estimatees error

The dammping phenomena is no ot taking intto account in


i control laaw. Despite this, the im
mpact of
the distuurbance (daamping) on thet control quality is not
n significant, which caan be shown n by
using quuality ratinggs (Table 3)).
4.2.3. C
Case study 3
In this ccase the appproximationn of PD conttrol was useed and the operation
o off the system
m without
disturbaances was teested. Fig. 12
1 shows thhe norm of errors.
e The graphs
g showw that the sy
ystem is
stable. QQuality of control
c is deetermined byy quality raatings presen
nted in Tablle 3.

Fig. 12. The analyssis of signaals boundednness in the case


c of PD control in thhe form : a)
the normm of filteredd tracking error along tthe tangential direction, b) the normm of filtered
d
trackingg error alongg the normaal direction,, c) the norm
m of the parrameter estimmates error , d)
the normm of the parrameter estimates errorr

4.2.4. C
Case study 4
In this ccase the appproximation
n of PD conttrol was useed and the operation
o off the system
m in the
circumsstances of diisturbances was tested.. The normss of errors presented
p onn Fig. 13 aree
limited and the system remain ns stable. Quuality of con
ntrol is deteermined by qquality ratin
ngs
presenteed in Table 3.
Fig. 13. The analysis of signals boundedness in the case of a disturbance and PD control in
the form : a) the norm of filtered tracking error along the tangential direction, b) the norm
of filtered tracking error along the normal direction, c) the norm of the parameter estimates
error , d) the norm of the parameter estimates error

4.3. Discussion on the results


In order to compare the quality of control, quality ratings defined by the formulas (122) and
(123) were set. The ratings were not set for the whole time horizon, but separately for each
cycle (the cycles are presented in Fig. 3), in order to illustrate the increase in quality of control
resulting from the adaptation of the system. In the last line of the table a percentage change in
quality of control in the case study 3 with regard to the case study 1 and in the case study 4
with regard to case study 2 are presented. The difference in quality of control results from the
use of PD control approximation in case studies 2 and 4. Quality ratings are presented in
Table 3.

Table. 3. Quality of control


Quality rating Case study 1 Case study 2
Cycle 1 Cycle 2 Cycle 3 Cycle 1 Cycle 2 Cycle 3
sτ,max 0,0136 0,0013 0,0005 0,0136 0,0013 0,0005
sn1,max 42,2116 20,1518 16,6356 42,2126 20,1647 16,6503
sτ,RMS 0,0323 0,0032 0,0017 0,0323 0,0032 0,0017
sn1,RMS 82,4389 70,4405 71,4704 82,4319 70,4911 71,5293
Case study 3 Case study 4
Cycle 1 Cycle 2 Cycle 3 Cycle 1 Cycle 2 Cycle 3
sτ,max 0,0133 0,003 0,0005 0,0133 0,003 0,0005
sn1,max 44,1253 5,2929 4,689 44,0854 5,2504 4,6159
sτ,RMS 0,0325 0,0178 0,0014 0,0325 0,0178 0,0014
sn1,RMS 95,2093 23,3805 33,2959 94,3135 23,1092 32,7147
Improving -31,33% -98,44% 27,18% -30,85% -98,22% 27,50%
the control
quality

From the analysis of the quality ratings results, that:


- the introduced damping has a marginal impact on the control quality,
- the application of PD control approximation in the form of results in percentage
averaged increase in quality ratings of more than 27% in the third cycle, whereas in the initial
cycles, when the fastest adaptation of parameters takes place, the quality of control
deteriorated.

5. Experimental study
The experimental research concentrates on the application of the adaptive control
system for two degrees of freedom manipulator. The laboratory stand [36] for this research
features a robotic manipulator, the dSPACE control and measurement system and a PC
workstation with Matlab, Simulink and ControlDesk. The manipulator with open control
system is equipped with DC gear-motors with incremental encoders and end-effector with the
ATI FTD-Gamma SI-130-10 force/torque sensor. The sensor measures the three components
of the force and three components of the torque and provides the feedback necessary for the
implementation of the force control. Computations connected with the control signals
generating and data acquisition were realised using the dSPACE digital signal processing
boards. The system consists of a set of the dSPACE control and measurement boards: the
pragrammmable proccessor card DS1006 wiith three exp pansion cards: DS40022 (PWM and d digital
I/O), DS S3001 (incrremental enccoders), DS
S2002 (analo og inputs). For program
mming the
processoor card DS11006 the Maatlab with SSimulink software is ussed. Controll algorithm is built
in Simuulink environnment and then
t compilled. The alg
gorithm saved in the coorrespondingg code
is loadeed to memorry card and can be donee in real-tim
me. To manaage the expeeriment, reccording
and visuualization of data, the ControlDesk
C k software is
i applied. The
T schemee of the labo oratory
stand is shown in Fig.
F 14.

Fig. 14. The schem


me of the lab
boratory stannd

5.1. Casse studies


In the exxperimentall study, the same condiition as in th
he simulatioon study aree applied, i.e. the
desired trajectory described
d by
y equations (119) and (120),
( the coontrol law ((43), the parrameter
tuning aalgorithm giiven by equ uations (71) and (72) an
nd the parammeters from m Table 4 aree used.
Scenarioos of controol with and without norrmal force derivative
d iss applied (T
Table 5).

Table. 44. Parameterrs used in experimentaal study


parametter unit value
v
Λτ s-1 3
-1
Λn s 3
34.6
KDτ11 kg/s 8
KDn11 s 0
0,00173
kτ - 0,2
kn1 - 0,2
vmax m/s 0,03
wτ s-1 5
Fe nmax N -10
wn s-1 5
cτd(0) m 0,22
̂ 0 - 0
̂ 0 - 0

Table 5. Case studies - experiment


Case study
1 2
Varying factors PD control in the form of PD control in the form of
Conditions of practical application
Differentiation of force Not required Required
signal
Force signal filtering Not required Required
before differentiation
Time of control 6,04e-5 s 6,08e-5 s
calculatio
Estimated accuracy of Due to the lack of knowledge on i.a. parameters of the real
the system object, it is not possible to estimate the accuracy of the system
precisely

Control law (32) with a PD term in the form of KDs (case study 2) may be difficult to
realisation because of the need for determining the pressure force derivative. The
measurement of forces, even those obtained using high quality systems, are subject to
measurement noise, which in the case of small interaction forces has a large share in the total
measurement signal. Introduction of derivative of such signal to the control system often
causes system instability, as the oscillation of the control signal stimulates eigenfrequencies of
the robot-environment system. In order to circumvent this inconvenience, it is possible to omit
the interaction force derivative and assume the approximation of PD control, where the
filtered tracking error s is replaced by in the form of:
(124)

With such an approach the selection of filter for force signal and differentiation of force signal
are not required. In the case of exercising the PD control in the form of KDs force signal
filtering with low-pass filter is necessary.

5.2. Results of experiments


In this subsection, selected results of tests performed on the real object are presented. The
results of the first test were described completely, whereas the description of the second one
focuses only on the most relevant differences.

5.2.1. Case study 1


Waveforms of signals recorded during experimental study are similar to those obtained in the
simulation. The generated control signals (Fig 15a) consists of PD control (Fig 15b),
compensatory control (Fig 15c), the normal force control λ2 and friction force compensation
λ1 (Fig. 15d). In Fiig. 16 the reeal trajectorry and path of end-effeector are shhown. In thee Fig. 17
movemeent errors (Fig.
( 17a,c) and force errors (Fig.. 17b,d) aree presented.. The only essential
differennce compareed to the sim t normal fforce error does not
mulation is that the derrivative of the
decreasee significanntly during the movem ment of thee manipulator. This is due to thee applied
control law, wherrein the forrce error dderivative is omitted. Fig. 18 annd Fig. 19 present
estimatees of parammeters. In Fig. 20 overrall control signals in joint
j space are presen nted. The
norms oof filtered trracking erro
or in the Figg. 21 are sh
hown. In thee real case ththe practicall bounds
of the ffiltered traccking error and param meters estim
mation errors (bsτ, bsn11, bpτ and bpn1) are
unknow wn. Moreoveer, the parameter estim mates error and can not bee calculated because
the real parameter ofo the manipulator andd the environ nment are unknown.
u

Fig. 15. Control siggnals generaated during experimentt associated d with the w workspace: a)
a the
overall control signnals, b) the PD
P control signals, wh here UPD1= KDτ11sτ, UPDD2= KDn11sn1 c) the
compennsatory conttrol, d) the normal
n forcee control (λ2) and frictiion force coompensation
n (λ1)

Fig. 16. The real trajectory exeecuted by thhe system: a)


a displacem
ment of mannipulator en nd-
effectorr along the tangential
t direction, b) displacemeent of manip
pulator end--effector alo
ong the
normal direction, c)
c path of the end-effecttor, d) norm
mal interactiion force
Fig. 17. The trackinng errors occcurring durring the exp periment: a) error of dissplacement along
the tanggential direcction, b) erro
or of normaal interaction
n force, c) error
e of veloocity along the
tangentiial directionn, d) error of time derivvative of normal interacction force

meter estimattes ̂ for nnonlinear fu


Fig. 18. The param unction
meter estimattes ̂
Fig. 19. The param for nonlinear function
fu

Fig. 20. The overalll control sig


gnals in thee joint spacee

Fig. 21. The normss of filtered tracking errror: a) along


g the tangen
ntial directioon, b) along
g the
normal direction

The envvironment damping


d is not
n taking innto accountt in control law. Despitte this, all signals in
the clossed-loop sysstem are bo
ounded, so iit can be staated that thee control syystem is staable. The
quality rratings are summarized
s d in the Tabble 6.
5.2.2. C
Case study 2
In this ccase the PD control witth the derivaative of the force signaal and low-ppass filtering
g was
d in Fig. 22 are limited and the system
used. Thhe norms off filtered traacking errorrs presented
remainss stable. Nonnetheless, one
o may nottice higher amplitudes
a and oscillattions of sign
nals than
in the caase study 1,, which negatively influuences the quality
q of coontrol (Tablle 6).

Fig. 22. The normss of filtered tracking errror in case study


s 2: a) along
a the taangential dirrection,
b) alongg the normaal direction

5.3. Disscussion on results of experimentts


In orderr to comparee the qualityy of controll, the quality y ratings (Table 6) desccribed by means
m of
formulaas (122) andd (123) weree set. In ordder to illustrate the increease in quallity of contrrol
resultingg from adapptation of th
he system, thhe ratings were
w set for each motioon cycle sep parately
(the cyccles are pressented in Fig
g. 3). In thee last line off the table, the
t percentaage change in
quality of control inn case studyy 1 with reggard to case study 2 wh hich results ffrom
approximation of PD P control in case studyy 1 is presented.

Table. 66. Quality of control in experimenttal studies


Qualityy rating Caase study 1
C
Cycle 1 Cycle
C 2 CCycle 3
sτ,max 0,,0205 0,0084
0 00,0094
sn1,max 422,8731 32,8411
3 332,8766
sτ,RMS 0,,0503 0,0312
0 00,0252
sn1,RMS 1227,5768 99,7368
9 112,6121
Caase study 2
C
Cycle 1 Cycle
C 2 CCycle 3
sτ,max 0,,0208 0,0118
0 00,0112
sn1,max 744,0151 36,1359
3 440,7902
sτ,RMS 0,,0497 0,0342
0 00,0276
sn1,RMS 218,2805 106,5392
1 134,5473
Improvving 299,66% 10,73%
1 12,01%
the conntrol
qualityy

From thhe analysis ofo the experriments resuults, that:


- the usee of PD conntrol approxximation in tthe form off causes percentagge shorteninng of the
time reqquired for exxercising th
he control allgorithm in real time off 0,66%; in absolute vaalues
this givees the increase of 0,4 μs,
μ thus from m practical point of vieew it is of nno meaning.
- the usee of PD conntrol approxximation creeates percenntage averagged increasee in control quality
ratings iin subsequeent cycles reespectively of more thaan 29%, 10% % and 12%..
6. Summary
The approach to control of an object operating with flexible environment presented in the
paper is based on the knowledge of mathematical structure of the system model. Because the
system model is used in the control algorithm, the adopted mathematical model takes into
account the most important physical phenomena, and omits those with little quantitative
impact, such as damping of an environment. A common approach in such cases is arbitrary
assumption of environmental parameters in order to take them into account in the control
system. This paper presents an algorithm for adaptation of parameters throughout the system
while taking into account the environment characteristics. The article analyses the behaviour
of a closed-loop control system and demonstrates its stability, also confirmed through an
experimental study. The advantage of the present algorithm is the fact that unknown
environment stiffness is considered in the control system.
  Experimental studies show a control variant without the pressure force error derivative.
This simplification was made due to the oscillations of the normal force derivative caused by
differentiation affected by measurement noise. This shows that force control in the normal
direction is not of as good quality as according to theoretical considerations and simulation
studies, but the control system remains stable. Naturally, another method which can be
successfully used primarily in constant or slowly variable desired interaction forces is to
introduce an integral element to the force control system. 

Acknowledgements
This work was supported by Rzeszow University of Technology under statutory activities
[DS.MA.17.001].

References
[1] J.J. Craig, Introduction to robotics: mechanics and control, third ed., Upper Saddle River,
N.J, Pearson/Prentice Hall, 2005.
[2] D. Whitney, Historical perspective and state of the art in robot force control, Robotics and
Automation. Proceedings, 2 (1985) 262–268.
[3] M. Lotz, H. Bruhm, A. Czinki, A New Force Control Strategy Improving the Force
Control Capabilities of Standard Industrial Robots, J. Mech. Eng. Autom., 4 (2014) 276-283.
[4] F. Tian, C. Lv, Z. Li, G. Liu, Modeling and control of robotic automatic polishing for
curved surfaces. CIRP J. Manuf. Sci. Technol. 14 (2016) 55–64.
[5] F. Yu, M. Minami, T. Maeba, A. Yanou, Constraint-combined Force/Position Hybrid
Control Method with Lyapunov Stability, SICE Annual Conference, Japan, (2011) 671-676.
[6] C.A. Canudas de Wit, B. Siciliano, G. Bastin (Eds.), Theory of robot control, Springer,
New York, 1996.
[7] P.B. Goldsmith, B.A. Francis, A.A. Goldenberg, Stability of hybrid position/force control
applied to manipulators with flexible joints, Int. J. Robot. Autom., 14 (4) (1999) 146-160.
[8] M. Vukobratovic, V. Potkonjak, V. Matijevic, Dynamics of Robots with Contact Tasks,
Springer, Dordrecht, 2003.
[9] Q. Zhu, Y. Mao, R. Xiong, J. Wu, Adaptive Torque and Position Control for a Legged
Robot Based on a Series Elastic Actuator, Int. J. Adv. Robot. Syst. 13:26 (2016) 1-9, doi:
10.5772/62204
[10] T. Endo, H. Kawasaki, Bending moment-based force control of flexible arm under
gravity, Mech. Mach. Theory 79 (2014) 217–229.
[11] M. Dapper, R. Maass, R. Eckmiller, Neural Velocity Force Control for Industrial
Manipulators Contacting Rigid Surfaces, in: L. Niklasson, M. Bodén, T. Ziemke (Eds.).
ICANN 98. Springer, London (1998) 887–892.
[12] L. Villani, J. De Schutter, Force Control, in: B. Siciliano, O. Khatib (Eds.), Springer
Handbook of Robotic. Springer, Berlin, Heidelberg (2008) 161–185.
[13] M. Vukobratovič, Y. Ekalo, A. Rodič, How to Apply Hybrid Position/Force Control to
Robots Interacting with Dynamic Environment, in: G. Bianchi, J.-C. Guinot, C. Rzymkowski
(Eds.), Romansy 14. Springer , Vienna (2002) 249–258.
[14] A. Fanaei, M. Farrokhi, Robust adaptive neuro-fuzzy controller for hybrid position/force
control of robot manipulators in contact with unknown environment, J. Intell & Fuzzy Syst.
17 (2) (2006) 125-144.
[15] F. Ferguene, R. Toumi, Dynamic External Force Feedback Loop Control of a Robot
Manipulator Using a Neural Compensator—Application to the Trajectory Following in an
Unknown Environment, Int. J. Appl. Math. Comput. Sci. 19 (1) (2009) 113–126.
[16] H. Liu, L. Wang, F. Wang, Fuzzy force control of constrained robot manipulators based
on impedance model in an unknown environment, Front. Mech. Eng. China, 2 (2) (2007)
168–174.
[17] J. Pliego-Jiménez, M.A. Arteaga-Pérez, Adaptive position/force control for robot
manipulators in contact with a rigid surface with uncertain parameters, Eur. J. Control 22
(2015) 1–12.
[18] P. Gierlak, Hybrid Position/Force Control of the SCORBOT-ER 4pc Manipulator with
Neural Compensation of Nonlinearities, Lecture Notes in Comput. Sci. 7268 (2012) 433-441.
[19] P. Gierlak, Hybrid position/force control in robotised machining, Solid State Phenom.
210 (2014) 192-199.
[20] Z. Hendzel, A. Burghardt, P. Gierlak, M. Szuster, Conventional and fuzzy force control
in robotised machining, Solid State Phenom. 210 (2014) 178-185.
[21] F.L. Lewis, C.T. Abdallah, D.M. Dawson, Control of robot manipulators, Macmillan,
New York, 1993.
[22] N. Kumar, V. Panwar, N. Sukavanam, S.P. Sharma, J.-H. Borm, Neural network based
hybrid force/position control for robot manipulators, Int. J. Precis. Eng. Manuf. 12 (3) (2011)
419–426.
[23] Y. Wu, S. Chen, Adaptive neural motion/force control of constrained robot manipulators
by position measurement, In IEEE; 2011 [cited 2016 Apr 21]. p. 498–502. Available from:
http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6021902
[24] L. Villani, C. Natale, B. Siciliano, C.C. de Wit, An experimental study of adaptive
force/position control algorithms for an industrial robot, IEEE Trans. Control Syst. Technol. 8
(5) (2000) 777-786.
[25] J. Marvel, J. Falco, Best practices and performance metrics using force control for
robotic assembly. US Department of Commerce, National Institute of Standards and
Technology, 2012.
[26] F.L. Lewis, S. Jagannathan, A. Yeşildirek, Neural network control of robot manipulators
and nonlinear systems, Taylor & Francis, London, 1999.
[27] J.-J.E. Slotine, W. Li, On the Adaptive Control of Robot Manipulators. Int. J. Robot. Res.
6 (3) (1987) 49–59.
[28] R. ul Amin, L. Aijun, M.U. Khan, S. Shamshirband, A. Kamsin, An adaptive trajectory
tracking control of four rotor hover vehicle using extended normalized radial basis function
network, Mech. Syst. Signal Process., 83 (2017) 53-74.
[29] M. Shojafar, N. Cordeschi, E. Baccarelli, Energy-efficient adaptive resource
management for real-time vehicular cloud services, IEEE Trans Cloud Comput. (2016) 1-1
[30] E. Baccarelli, P.G.V. Naranjo, M. Shojafar, M. Scarpiniti, Q*: Energy and delay-efficient
dynamic queue management in TCP/IP virtualized data centers, Comput Commun 102 (2017)
89-106.
[31] S. Sajjadi, S. Shamshirband, M. Alizamir, L. Yee, Z. Mansor, A. Manaf, T.A. Altameem,
A. Mostafaeipour, Extreme learning machine for prediction of heat load in district heating
systems, Energy and Buildings 122 (2016) 222-227.
[32] J.-J.E. Slotine, W. Li, Applied nonlinear control, Prentice Hall, Englewood Cliffs, N.J,
1991.
[33] F.L. Lewis, K. Liu, A. Yesildirek, Neural Net Robot Controller with Guaranteed
Tracking Performance, IEEE Trans. Neural Networks 6 (3) (1995) 701-715.
[34] K. Narendra, A. Annaswamy, A new adaptive law for robust adaptation without
persistent excitation. IEEE Trans. Autom. Control 32 (2) (1987) 134–145.
[35] M.M. Polycarpou, P.A. Ioannu, Identification and control using neural network models:
design and stability analysis, Dept. Elect. Eng. Syst., Univ. S. Cal., Tech. Report 91-09-01,
Sep. (1991).
[36] M. Szuster, P. Gierlak, Approximate Dynamic Programming in Tracking Control of a
Robotic Manipulator. Int. J. Adv. Robotic Syst. 13:16 (2016) 1-18, doi: 10.5772/62129
An interaction of robot manipulator with flexible environment is considered.
A force/position tracking controller is proposed.
No information on robot parameters is required.
Practical stability is guaranteed by the adaptive controller.
System stability is proved by using Lyapunov stability theory.
 
Marcin Szuster, Maaster of Scieence, PhD iin Mechaniccs. He work ks in Departtment of Ap pplied
Mechannics and Robbotics, Rzesszow Univeersity of Tecchnology fo or more thann 9 years. Inn 2012
he has ccompleted hish Ph.D. theesis: Generaating and reealization off the wheeleed mobile roobot
trajectorry using neuural dynamiic programm ming. Winn ner of three awards natiional and
internattional. Authhor or coauth hor of over 40 publicattions, membber of the PT TMTS. He is
involvedd in researcch on prototy yping and immplementattion of mob bile robots aand robotic
manipullators contrrol systems, path planniing tasks, behavioural control andd robots form mation
control, artificial inntelligent methods
m and optimizatio
on methods in robotics..
Piotr Giierlak, Mastter of Science, PhD in M Mechanics.. He works in
i Departm ment of Appllied
Mechannics and Robbotics, Rzesszow Univeersity of Tecchnology sin nce 2008. Inn 2011 he has h
completted his Ph.DD. thesis: Neeural controol of manipu
ulator’s mov
vement. Auuthor or coau uthor of
over 30 publicationns, member of Polish S Society of Theoretical and
a Appliedd Mechanicss. He is
involvedd in researcch on modellling of mannipulators, adaptive
a con
ntrol, artificcial intelligeent
methodss in robotics, signal proocessing, m
mechanical vibrations.
v He
H has workked on mech hanics-
related R
R&D projeccts with sev veral compaanies.

S-ar putea să vă placă și