You are on page 1of 28

J Intell Robot Syst

DOI 10.1007/s10846-010-9407-x

A Hybrid Control Approach for Non-invasive Medical

Robotic Systems

Swandito Susanto · Sunita Chauhan

Received: 25 May 2008 / Accepted: 22 February 2010

© Springer Science+Business Media B.V. 2010

Abstract In this paper, a hybrid supervisory control approach adopted for a non-
invasive medical robot called Focused Ultrasound Surgical Robot—Breast Surgery
(FUSBOT-BS) is elaborated. The system was built for the use in the breast surgery
with high intensity focused ultrasound (HIFU) as the means of the treatment. A
number of different control strategies such as PID and model-based control were
incorporated into a family of controllers to create the hybrid control. Depending on
the objective, the supervisory control determines the type of controller used for the
specified task so as to maximize the advantages of each of the controllers. Before it
was implemented into the actual robotic system the then proposed control approach
was modeled and simulated using Matlab®. This control approach was developed
based on a review of popular control approaches used in medical robotic systems, in
order to look at the feasibility of having a uniform control strategy for a spectrum of
medical robotic system. With unified control strategy it is possible to have a safety
standard regulation for the medical robotic systems which is currently difficult to be
done because of various control strategies adopted by each of the medical robotic

Keywords Hybrid control · Medical robotic systems · Non-invasive surgery ·

High intensity focused ultrasound

S. Susanto (B) · S. Chauhan

Department of Mechanical and Aerospace Engineering, Nanyang Technological University,
50 Nanyang Avenue, RRC, N3-01a-01, Singapore, Singapore
S. Chauhan
J Intell Robot Syst

1 Introduction

In the last decade, the development and implementation of minimally invasive and
non-invasive surgery methods in the operating room have tremendously changed the
ways that conventional surgery was carried out in the past [1, 2]. In a minimally
invasive or non-invasive surgery where great accuracy, repeatability, and stability
are needed, robotic technologies come in handy. The applications of the robots and
computer integration in these methods range from simple robotic arms to highly
complex surgical robots [1–3]. However, due to the delicate nature of the surgery, the
use of robots is highly restricted in the surgical room. A number a stringent require-
ments are also imposed to any medical robotic system available and/or under devel-
opment for safety reasons. Given this fact, numerous studies are done continuously
in order to improve the robotic systems.
There are many factors that affect the safe operation of medical robots such as
design limitations to do complex surgical manoeuvres, malfunction of the system’s
components and unpredictability of the working envelope especially for invasive
surgical applications. One measure that can be taken in order to improve the safe
performance of medical robots given these factors is to have a reliable control system.
Besides being used to move the system along a desired trajectory, control itself serves
a higher purpose such as to ensure the performance and stability of the robot during
operation thus enabling better outcome in its application in surgery. With appropri-
ate control architecture, a complex movement can be planned and executed more
accurately. The errors can be predicted and compensated prior to the application,
and in case of failure, it is possible to force the system to fail in a predicted and safe
manner. Hence, devising an optimum control strategy becomes a very important task
and should be dedicated to the objectives and design of robotic system [4, 5].
There are a number of medical robotic systems available either in the market or
used in research laboratories around the world. Various types of control strategies
are adopted by these medical robots [6–11]. However, most of them can be classified
into two types of control, model-based and adaptive control. Some of the medical
robots that adopt model-based control include ROBODOC, a robot for joint replace-
ment surgery which adopts a force feedback controller with velocity and position
reference [6]. Another one is Mitsubishi PA-10 robot arm, which is widely used in re-
search laboratories worldwide for application as robot assisted surgery whose control
is based on torque transmission [7]. Several model based controller approaches have
also been proposed and implemented into robotic system for laparoscopic manipu-
lation [8]. As for the adaptive control, a robotic prosthetic eye was reported to have
used a neural network based control system where array of sensors is used to detect
the movement of the natural eye and the control system coordinates the data to
give movement to the prosthetic eye accordingly [9].
Each of these medical robotic systems adopts a different control strategy to suit
the developer which contributes to the problem of unifying safety standards of
medical robotic systems. Safety standard is very important considering the stringent
safety requirements put on medical robots. Furthermore a considerable amount of
time is needed to develop a control strategy for a robotic system. It is desirable to
have a common control approach for medical robots similar to their industrial coun-
terpart, for instance proportional-integral-derivative (PID) control is used by about
80% industrial robotic systems available [12].
J Intell Robot Syst

In this paper, an optimal control hierarchy and strategy is proposed for non-
invasive applications that would help in deducing a unified safety regulation which is
highly desirable. Secondly, an optimal control architecture would help in making the
development phase of medical robots shorter leading to enhanced usage. A hybrid
supervisory control was chosen due to its capability in combining the advantages of
more than one control strategies for a particular application. The proposed control
approach consists of model-based robust control and PID control as its parameter-
ized controllers with a discrete time event controller as the supervisor. The control
was implemented into a non-invasive medical robotic system to justify its feasibility.
In doing so few experimental scenario were also generated. Section 2 of this paper
elaborates the development of the proposed control strategy. The details of the
derivation including the simulation are included. In Section 3, the implementation
and experimental result of the proposed control strategy is shown. Section 4 gives the
summary and conclusion for the applicability of this proposed control approach in
achieving the objectives.

2 Control Derivations

2.1 Background

For verification purpose, the hybrid control was implemented into a non-invasive
medical robot called FUSBOT-BS, devised at the Biomechatronics Group, RRC,
NTU, Singapore [13, 14]. As shown in Fig. 1, a part of the robotic system is inside
the tank with the HIFU module as its end-effector. An ultrasound imaging probe
is located at the centre of the HIFU module and is used for online scanning. There
are five axes of movement: vertical axis (VA), rotational axis (RA), horizontal axis
(HA), orientation axis (OA), and jig axis (JA), which will be explained further in
Section 2.5 (Fig. 5).

2.2 Control Approach for FUSBOT-BS System

A hybrid control is characterized by interaction between continuous and discrete

dynamics [15, 16]. The advantage of having a hybrid control is that more than one
type of control strategies can be utilized to perform a specific application where in

HIFU module

Robotic arm
imaging probe

Water tank

Fig. 1 FUSBOT-BS System

J Intell Robot Syst

that application, each of the sub-application may have different dynamics requiring
different control strategies [15, 16]. Depending on its predefined strategy, only one
particular control from a family of controllers will be operating at a time. Thus, the
system needs to have a logic-based switcher to change the operating controller at
a specific condition. This switcher, based on its algorithm will switch the operating
controller upon either internal or external event. This pre-dominant logic component
is generally called a supervisor [17, 18]. The logic component could be either one
of the controls implemented into the system or an independent algorithm or even
a human operator. The overall responsibility of a multi-controller supervisor is to
decide when to switch controller (scheduling) and which controller to switch to next
(routing) [19].
Among various control approaches, proportional-integral-derivative (PID) was
chosen as one of the candidates given its strength such as ease of implementation,
able to be implemented into various systems, etc. PID controller is very popular
especially for industrial robotic systems. In fact, a wide range of PID controllers and
PID tuning software are available in the market. However, even though generally
the PID controller can give the desired result in terms of the system’s performance
(accuracy, repeatability), it also has some weaknesses especially in the situation
where rapid disturbances or time-varying parameters are expected [20, 21].
In a medical environment, rapid disturbances or time-varying parameters may
exist due to the presence of human as an operator and as a subject of operation even
though other operating condition e.g. position of the robot, room condition remain
the same. The possible threat from these unpredictable environments must be mini-
mized if not eliminated. Thus, having a different control approach either to be imple-
mented competitively or as complimentary to the PID controller is necessary. Some
control approaches from the model-based control and non-model based control are
the candidates.
The PID control adopted is a non-model based PID control. Hence, it is logical
to have a model-based control instead of another non-model based control as a
comparative or complimentary control approach. Model-based controls have been
proven to be able to give a predictable result to the control objectives (e.g. stability,
accuracy) provided that the system’s dynamics is known. Furthermore, most of the
medical robotic systems reviewed adopt a model-based control approach either
implemented alone or combined with other control approaches such as PID control
approach, showing its wide acceptance among control engineers [12, 20].
In modelling the control system, the most difficult part is modelling the dynamics
of the robotic system. Dynamic parameters such as friction, backlash, inertia, Cori-
olis, etc. mostly are unknown although some of them can be estimated in known
situations. Furthermore, system dynamics change over time and thus, it could affect
the performance of the control system. Hence, in medical robots where the safety
of the patient and the reliability of the robots are highly demanded, dynamics may
pose as a big problem. Fortunately, for medical robotic systems, the system dynamics
generally remain the same over time (no strong external and internal disturbances)
and this makes it suitable for a model-based controller. During its operation, a
medical robotic system most of the time does not repeat the same movement (e.g.
the trajectory of the arm) even though the operation procedure may be the same
(e.g. moving the arm to the target area). Thus, a non-model based control approach
such as adaptive-learning control is not suitable for a medical robotic system.
J Intell Robot Syst

2.3 Hybrid Control Modeling

A hybrid system consists of a continuous and a discrete part. In FUSBOT-BS, the

continuous part is represented by a set of nonlinear ordinary differential equations
(ODE) of the continuous dynamics. On the other hand, the discrete part is described
by the discrete event controller which is connected to the system in a feedback
configuration through an interface (i.e. partitioning of its state-space). The math-
ematical modelling of the hybrid supervisory control of the FUSBOT-BS which is
shown here is inspired by the modelling developed by Koutsoukos et al. [17], due to
its suitability to the system and relatively simple to be implemented. Figure 2 shows
the general block diagram of the FUSBOT-BS hybrid supervisory control.
The controller/supervisor is a discrete event system modelled as a deterministic
finite automaton. The automaton is specified by S = (  δ, φ), where 
S, X, S is the
set of states obtained from the system states condition (e.g. position, angle, etc.), X 
is the set of plant symbols generated by the plant, R  is the set of controller symbols
generated by the supervisor, δ is the state transition function, and φ is the output
function. Between the supervisor and the hybrid system, there is an “interface”.
The supervisor and the system communicate via the interface which converts the
continuous time signal to sequences of symbols understood by the supervisor and
vice versa. The interface system consists of generator and actuator sub-system. The
generator generates plant symbol to the supervisor upon plant event. On the other
hand, actuator converts supervisor’s decision into input signal for the plant [17].
In the case of FUSBOT-BS, the plant event is defined by specifying several
“hypersurfaces”. Hypersurface refers to the borderline/partitioning line that divides
set of state space into few regions. If the state crosses the hypersurface from one
partition to another partition which has been pre-determined, the event occurs. The
hypersurface state is specified based on the error data generated by experimental test
with stand alone PID position controller (as explained in Section 2.4). From these
data, position and velocity range of the manipulator is partitioned. The variation
of the state variables of the hybrid control strategy is chosen to be “steady state”

Fig. 2 Block diagram of FUSBOT-BS hybrid supervisory control

J Intell Robot Syst

variation. In this variation, the values of the state variables at the end of one control
section are used as the initial conditions for the following control section. The plant
symbol is generated by the interface to the supervisor according to the function
shown in the following:

αi : N (hi ) → X (1)

where, hi for i = 1, 2, · · · , m is the hypersurface specified by the designer of the

system. N(hi ) is the function that correlates the hypersurface with its plant symbol.
And the sequence of the plant symbols is described by:

x [n] = αi (x (τe [n])) (2)

where (τe [n]) is the time of the nth plant event, with τe [0] = 0.
After the supervisor receives the plant event symbol from the interface, it will
analyze it and then generate supervisor symbol to the interface which in turn will
generate appropriate command to the system. In between generation of the two
symbols there is a time delay. This is undesirable as it could cause a problem to the
controller especially when more than one plant events occur within the period of the
time delay. This can be avoided if the supervisor can estimate when the plant event
will take place [22, 23].
Once the plant symbol is received by the supervisor, the supervisor determines
which controller is suitable for the next system states. Considering the equations
for the nonlinear time-invariant system expressed by a set of ODE, the system,
x = X(x, u, v, t), y = H(x) with control input u, output y, with positioning error
[eT ≈ r − y] where r is reference point, is governed by a hybrid controller with
event driven logic switcher. The goal of the logic switcher is to change the controller
(generating output [φ]) in order to bring eT ≈ 0.
Let p is the ith number of the plant event for i = 1, 2, · · · , m. p is a subset of a
function of real, finite number P. X EP denotes the state-space variable at p. For
each p ∈ P, e p ≈ y p − y denotes the p output estimation error. π p is the “normed”
value of e p or a performance signal, which
 is used by the supervisor to assess the
potential performance of controller p. s is a switching logic whose function is to
determine σ based on the current value of π p . This estimator-based supervisor from
time to time when the plant event occurs selects for candidates’ control signal [ν p ]
whose corresponding performance signal π p is the smallest among the π p s, for [ p ∈
P]. Since FUSBOT-BS has only two continuous controllers (PID and model-based)
to be chosen it is relatively simpler and fast.
The automaton is specified by G = (  R,
S, X,  ψ, λ), where  S is the set of states
variables, X is the set of plant symbols, R is the set of controller symbols, ψ is
the state transition function, and λ is the output function. Since FUSBOT-BS has
only two continuous controllers, the controller symbol is R  = {r1 , r2 }. The actuator
generates two possible outputs to plant which are,

−1 if 
r = r1
λ (
r) = (3)
1 if r = r2

In which −1 refers to model-based robust controller and 1 refers to the PID con-
troller. After receiving the deciphered supervisor symbol, the plant will implement
J Intell Robot Syst

a suitable controller as instructed by the supervisor. In FUSBOT-BS the generator

recognizes six plant events in terms of region profile.
h1 (x) = x1 h2 (x) = x2 h3 (x) = −x2
h4 (x) = x3 h5 (x) = −x3 h6 (x) = −x4
These events are generated when the end-effector crosses either x1 , x2 , or x3 axis
which partitions the x–y plane region of the robot manipulator workspace as shown
in Fig. 3. The negative sign occurs when the state crosses the hypersurface from inside
and vice versa. Symbols generated by the plant are as follows:
α1 (x) = 
x1 α2 (x) = x2 α3 (x) = −x2
α4 (x) = 
x3 α5 (x) = −x3 α6 (x) = 

2.4 PID Gains for the FUSBOT-BS System

Different gains were generated through trial and error (heuristic) approach for
the PID controller based on its position, velocity, and acceleration profiles. The
procedure for gain generation was as follows,
1. All the possible movements of the individual motors were determined,
2. A range of possible distances for each configuration for each of the available
possibility in point 1 was defined (e.g. 2,500, 10,000, 50,000 encoder counts, etc).
The range of distances was chosen based on the anthropomorphic human data
and the common location of the breast cancer occurrences,
3. The tolerable position error and motor stability (by observation) were defined
and based on that the PID terms were varied one by one to obtain the optimum
PID set-up,
4. Using the PID gains obtained, the speed and acceleration were varied. The
range of speed and acceleration where the motor can give optimal performance
according to the objectives were then determined.
Figure 4a–f show the movement error of each of the axes in various distances using
the obtained optimum PID gains.
The PID gains generated through the testing and the movement error of each axis
with respect to the travel distance for different directions of movement are shown in
Fig. 4a–f. Qualitatively there was no instability observed during test movements in

Fig. 3 Annular regions of FUSBOT-BS workspace (transverse view)

J Intell Robot Syst

Fig. 4 a Movement errors vs.

distance (JA). b Movement a
errors vs. distance (OA). 0.6
c Movement errors vs. distance 0.5

Error (mm)
(HA). d Movement errors vs. 0.4
distance VA. e Movement 0.3
errors vs. distance (RA-CCW). 0.2
f Movement errors vs. distance Limit
(RA-CW) 0
0 5 10 15
Travel Distance(mm)

Error (degrees)

0.3 Clockwise
0.25 Counterclockwise
0.15 Cw-Ccw
0.1 Limit
0 10 20 30 40 50 60
Travel Distance (degrees)

Error (mm)

0.4 Backwards
0.3 Fw-Bw
0.2 Limit
0 50 100 150
Travel Distance (mm)

d 0.6
Error (mm)

-10 10 30 50 70
Travel Distance (mm)

e 0.8
Error (degrees)

0.6 1st 120 deg

2nd 120 deg
0.3 3rd 120 deg
0.2 Limit
0 40 80 120
Travel Distance (degrees)
J Intell Robot Syst

Fig. 4 (continued)
f 1.2

Error (degrees)
1st 120 deg
2nd 120 deg
3rd 120 deg
0 40 80 120
Travel Distance (encoder counts)

the defined domain. One of the reasons for this might be due to a sufficiently low
optimum PID gains. Based on the actual movement errors, it can be found that for
certain movements a PID controller is not sufficient to give the accuracy needed, for
instance, the rotational axis as shown in Fig. 4f doesn’t satisfy the required accuracy
of system. Thus, for such cases there is a need to minimize its inaccuracy to acceptable
tolerance level.

2.5 Modeling of the FUSBOT-BS System

In order to derive the model-based control, the kinematics and dynamics of the
system need to be modeled. For derivation of kinematics equations of the FUSBOT-
BS system, modified Denavit–Hartenberg (D-H) method was adopted [24, 25]. D-H
has become the standard robot kinematics model because of its physical interpreta-
tion, strict definition and multiplicative structure. Figure 5 shows the modified D-H
coordinate systems. The transformation matrix from the base of the robot to the end
effector can be derived by multiplying all of the transformation matrices from the
first joint to the last joint. The expression for the transformation T, can be written as:

⎡ ⎤
S (θ2 + θ4 ) −C (θ2 + θ4 ) 0 d2
⎢ C (θ2 + θ4 ) S (θ2 + θ4 ) 0 0 ⎥
TBase ⎢
=⎣ ⎥ (6)
0 0 1 − (r1 + r2 + r3 + r4 ) ⎦
0 0 0 1

where, S stands for Sine, C stands for cosine, θ is the angular displacement, d is the
linear distance and r is the linear displacement.
In order to mathematically describe the dynamic properties of a manipulator,
a classical method from analyses of mechanics, Lagrange equation of motion, was
chosen due to its simplicity and systematic nature [26]. There are two types of
models, which are differential model and integral model. The first model requires the
measurement of positions, velocities and accelerations while the latter only requires
the measurement of positions and velocities [26, 27].
The differential model was adopted and the system’s dynamics is derived using the
generalized notion of relative position and orientation between links according to the
J Intell Robot Syst

Fig. 5 Line diagram of

4 HIFU axis


OA 2, 3



0, 1
XZ - Plane

modified D-H parameters. Lagrange equation in generalized Lagrange coordinates

for a kinematics chain can be written as [27]:

d ∂L ∂L ·
− = τi − Qi qi (7)
dt dq̇i dqi

where τi is the generalized force/moment which acts along the zi axis of the ith
coordinate frame, L is the Lagrange function (the
· difference between total kinetic
and potential energy of a manipulator), and Qi qi is the friction force acting on the
ith joint. Qi qi can be calculated from:
· · ·
Qi qi = Fiv qi + Fis sgn qi (8)

where, Fiv is the viscous friction coefficient of the ith link while Fis is the coefficient
of the Coulomb friction. The generalized coordinate which describes the movement
of the ith link is denoted by qi , and it can be calculated as:

qi = (1 − σi ) θi + σi ai (9)

where σi = 0 if the joint i is rotational and σi = 1 if joint i is translational.

Some of the actuators of FUSBOT-BS are not directly coupled to their respective
motors. Instead, they are connected to belt chains. Thus, FUSBOT-BS is classified
under a class of geared robots. Aside from dealing with the dynamics of friction which
J Intell Robot Syst

is also found in the direct drive robots, geared robots have to deal with transmis-
sion dynamics too. The friction force is hardly linear and in order to identify the
coefficient of frictions, identification testing is needed. It is highly difficult to achieve
an accurate dynamic model with the entire dynamics variables and its nonlinearity.
Fortunately, some of the variables is less significant, thus can be eliminated and some
can be linearized or treated as constant. Following are certain assumptions made for
FUSBOT-BS differential model:
1. Dynamics of transmissions are neglected.
2. Gear ratios are constant.
3. Friction force acting at a joint is linear with respect to the viscous friction
coefficient and coulomb friction coefficient.
4. The robot model is canonical.
The Lagrange function of a manipulator can be calculated as [27]:
L = EKc − EPc (10)
where, EKc is the total kinetic energy and EPc is the total potential energy. The
LaGrange equation for the FUSBOT-BS system can be written as:
·· · ·
τ1 = (m1 + m2 + m3 + m4 ) q1 − (m1 + m2 + m3 + m4 ) g0 + F1v q1 + F1s sgn q1
τ2 = I2zz + η2 Im2 + I3yy + η2 Im3 + I4zz + η2 Im4 + 2q3 3 cz q2
  ··   ·· · ·
+ m4 4 cx Cq4 − m4 4 c y Sq4 + 3 cx q3 + I4zz + η2 Im4 q4 + 23 cz q2 q3
 · · · ·
− m4 4 c y Cq4 + m4 4 cx Sq4 + 3 cx q3 q4 + F2v q2 + F2s sgn q2
3  ··   ··
τ3 = cx − m4 4 c y Sq4 q2 + m3 + m4 + m4 4 cx Cq4 q3
  ·· · · · ·
+ −m4 4 c y Sθ4 + 2m4 4 cx Cθ4 q4 − m4 4 c y Cq4 q2 q4 − m4 4 cx Sθ4 q3 q4
·2   ·2 · ·
−3 cz q2 + m4 4 c y Cq4 − 2m4 4 cx Sq4 q4 + F3v q3 + F3s sgn q3
  ··   ··   ··
τ4 = I4zz + η2 Im4 q2 + 2m4 4 cx Cq4 − m4 4 c y Sq4 q3 + I4zz + η2 Im4 q4
 · · · ·
− m4 4 c y Cq4 + 2m4 4 cx Sq4 q3 q4 + F4v q4 + F4s sgn q4 (11)

where mii ci is the first moment of ith link, g0 is the gravitational constant, q is the
· ··
generalized joint coordinates, q is the generalized joint velocity, q is the generalized
joint acceleration, η is the inertia ratio, and i Ii is the (3 × 3) vector of inertia of the
ith link.

2.6 Dynamics Parameters Identification

For the model-based control the identification or a good approximation of the

dynamics parameters is of significant importance. Some dynamics parameters (e.g.
inertia) can be easily calculated and some such as friction and beam effect parameters
need to be validated through experiments. These dynamics were modelled and
J Intell Robot Syst

validated due to their significant effects on the accuracy and movement of the
manipulator. The nominal parameters such as the friction parameters are obtained
from experiments with respect to the system’s dynamics mentioned in Section 2.5.
The experiments are designed in such a way that the desired dynamics parameters
could be excited at a given time while keeping other parameters constant. A number
of experiments were done for each of the dynamic parameters obtain for more
accurate results.

2.6.1 Friction Modeling

Friction is a common phenomenon that occurs in any mechanical system. It is unde-
sirable in most of the cases because it causes errors in position and induces tracking
inaccuracy. In modelling the friction, there are two widely used models, static friction
model and dynamic friction model [28, 29]. In a static friction model, the friction is
modeled into few different components such as coulomb friction, viscous friction,
stiction and Stribeck with each of the components representing different aspects
of the friction force. On the other hand, in the dynamic friction model, instead of
having linear or constant friction components, its friction changes as a function of
variations in temperature, position, etc. Dynamics model has an advantage in terms
of its preciseness compared to the static model especially when the variation of the
variables is significant during the operation of the mechanical systems. However, it
has to be noted that with the increasing complexity of the dynamic friction model
there is a consequent increase in the computational complexity. Furthermore, in a
lot of cases, static friction model is more than enough to give the estimation of the
friction in a system [29, 30].
LuGre model of the dynamic friction model was chosen due to its completeness
and reasonable complexity compared to other friction models [28]. In fact it is
the most widely used model for modelling friction in mechanical systems. LuGre
dynamic friction model can be reduced to static friction model by omitting the
dynamic parameters from its equation [28]. For FUSBOT-BS system, the LuGre
model used is the static LuGre model where the dynamic parameters were excluded.
One of the reasons for this type of the LuGre model to be chosen is that the friction
is affected by a number of nonlinear dynamic parameters. Hence, it is difficult to
model every parameter accurately and since FUSBOT-BS system operates in a
stable environment, most of the nonlinear dynamics parameters such as the tem-
perature, position of the system have little or no significant change. The parameter
identification procedure of the LuGre model is based on similar friction parameters
identification as reported in [28, 30] and is elaborated in the following.
The general LuGre model is linearized by giving it a linear viscous friction and
constant damping which is called “standard parameterization” and has the form,

dz |v|
= v − σ0 z (12)
dt g (v)
g (v) = α0 + α1 e−(v/v0 )
F = σ0 + σ1 z +α2 v (14)

where denotes the average bristle deflection which is determined by the velocity.
Bristle deflection in this case refers to the force that originates from the random
J Intell Robot Syst

distribution of interacting asperities on a surface. is the velocity, is the stiffness

coefficient of the microscopic deformations of during the pre-sliding displacement
of the bristles, is the damping coefficient associated with, model the Stribeck effect,
is the Coulomb friction force, corresponds to stiction force, is viscous friction, and is
the Stribeck velocity.
From Eqs. 12, 13, and 14, the steady-state friction characteristics for constant
velocity motions are given by,

Fss = α0 + α1 e−(v/v0 ) sgn (v) + α2 v

The Eq. 15 is characterized by four static parameters (α0 , α1 , α2 , v0 ) which were

estimated from the experimental data. The manipulator motions were performed
under constant velocity, in which the input torque values were recorded. The friction-
velocity data were then obtained from averaging those data using Eq. 8. Different
constant velocity values ranging from −15 to 15 mm/s for linear motion and −15◦ /s
to 15◦ /s for rotational motion were used with at least 20 data points collected. This
was done in order to identify closely the Stribeck velocity, Coulomb and Stiction
friction force. The equations for friction-velocity graph are as follows [28]:
d2 x
J =u− F (16)


min ss (vi ) 2
Fss (vi ) − F (17)

α0 ,
α1 ,
α2 ,

where, J is the total motor and load inertia, x is the position, v = dx/dt is the velocity,
u is the DC motor torque, F is the friction torque, Fss (vi ) are the friction values
measured during constant velocity (uss = Fss ) at velocities vi , and Fss (vi ) is given as,

ss =  α1 e−(v/v0 ) sgn (v) + 
F α0 +  α2 v (18)

Matlab (version 7.0) was used to find out the four static parameters (α0 , α1 , α2 , v0 )
of Eq. 15. Using Matlab optimization toolbox, the static parameters of the friction
model were calculated by minimizing the Eq. 15. An initial solution of zero was used
for each of the linear variables for initial datum. The optimization function found
the closest variables to minimize the function. These nominal values obtained were
used in the control as the nominal static friction parameters. The frictions vs. velocity
graphs are shown by Fig. 6a and b. The nominal friction parameters for each of the
sub-systems are listed in Table 1.
In Table 1, v < 0 and v > 0 represents the negative and positive velocity/direction
respectively. It was noticed that the Stribeck effect was very small and thus was taken
as zero. The friction forces of the axes with respect to the velocities are shown in
Fig. 6a and b. The general graphs after smoothing (curve fitting) were found to be
similar to stiction + coulomb + viscous friction graph (Fig. 6c). As can be seen from
Fig. 6a and b, for FUSBOT-BS system, the direction of the manipulator’s motion has
little or no effect towards the value of absolute friction forces value.

2.6.2 Backlash Modeling

The backlash for each of the gears mechanism used in the system was found through
manual calculation from the technical specification of the gears provided by the
J Intell Robot Syst

Fig. 6 a Friction–velocity
graph for VA, HA and JA. a 4
b Friction–velocity graph for
RA and OA. c Combined 3
general friction model for 2
static, coulomb and viscous

Friction (V)
-16 -12 -8 -4 0 4 8 12 16



Velocity (mm/s)

Vertical Axis (VA) Horizontal Axis (HA) Jig Axis

b 4
Friction (V)

-18 -14 -10 -6 -2-1 2 6 10 14 18

Velocity (deg/s)

Rotational Axis (RA) Orientation Axis (OA)


-v +v


Table 1 Nominal friction parameters

Friction Vertical axis Rotational axis Horizontal axis Orientation axis Jig axis
parameter v < 0 v<0 v<0 v<0 v<0 v<0 v<0 v<0 v<0 v<0
α0 (V) 0.711 0.9027 0.439 0.3912 0.1796 0.2421 −2.119 −1.408 −1.8046 −1.330
α1 (V) −1.761 0.4473 −0.899 0.1878 −0.4294 0.0079 −1.081 4.208 −1.3964 3.730
α2 0.0064 0.0019 0.0026 0.003 0.0007 0.0012 0.0065 0.0051 0.0165 0.0133
v0 (ec*/s) 0 0 0 0 0 0 0 0 0 0
ec* encoder counts
J Intell Robot Syst

manufacturer. This was later confirmed through experimental means from the input
and actual movement graph. A very small input (in terms of encoder counts) was
given to each of the motors and was increased until the movement can be seen from
the corresponding sub-system. In order to achieve an accurate measurement several
tests were done and the results were averaged. The mean value from both theoretical
and experimental tests was then calculated for each of the sub-systems (Table 2).

2.6.3 Cantilever Beam Ef fect Modelling

The HA module is supported at one end of the module while the other end is
unsupported. Thus HA shaft may behave like a cantilever beam. The deflection of
the HA shaft is expected due to the weight of the end-effector module during its
movement to the extreme end on the unsupported side of HA. This deflection will
affect the accuracy of the system and therefore requires to be compensated (Fig. 7).
The value of deflection (in millimeter) against the end-effector distance from
the central axis is shown in Fig. 5. The value was obtained through experimental
means where the height of the end-effector was measured against a fixed coordinate
system for some points along the horizontal axis. A polynomial curve was set to fit
all the measured points. The bending of the shaft affected the accuracy in vertical
and horizontal planes. However, the error due to bending was only compensated in
vertical direction since the error in the horizontal direction was very small even at
the extreme end of the axis (<<0.01 mm).

2.7 Modeling and Simulation of FUSBOT-BS Control System

Matlab software developed by MathWorks, Inc. was used for simulation of the
proposed control strategy which was subsequently implemented into the FUSBOT-
BS. There are several stages in system simulation which include, building the user
interface, modelling of the system structure and modelling of the control strategy.
These stages were implemented by using various Matlab toolboxes. In this section,
the system modelling and its results are presented.
Graphical User Interface (GUI) allows the user to choose the type of controller
to be used and to give certain inputs to the control, such as the gains for PID
controller. The GUI provides an easy layout for user to specify certain inputs to
the Matlab model for both the test system specifications as well as the intended
control approach. The second stage is numerical modelling of FUSBOT-BS where
the kinematics, dynamics and control strategies are modelled using the modelling
blocks provided by SimMechanics and Simulink toolboxes and through the m-files
of Matlab (version 7.0).

Table 2 Nominal backlash Axis Backlash width (encoder counts)

Vertical axis 127
Rotational axis 216
Horizontal axis 232
Orientation axis 222
Jig axis 218
J Intell Robot Syst

Fig. 7 Nominal value of the Cantilever Beam Effect

deflection 5

Deflection x 0.01 (mm)

-50 -30 -10
-5 10 30 50 70 90 110 130
y = 7E-08x4 - 2E-05x3 - 0.0009x2 + 0.0298x - 0.1661
Distance (mm)

The system is divided into five sub-systems. Each of the first four sub-systems
correspond to the number of DOFs namely VA, RA, HA, and OA. The last sub-
system is the end-effector module. Individual links and their joints are modelled
accordingly including the sensors and the drive as can be seen in Fig. 8. The
properties of the model were generated by Pro/E (based on its exact part model
created in Pro/E space). The properties include the mass, centre of gravity (CG) and
inertia tensor of the actual part with respect to its CG or a chosen coordinate system
for control modelling purpose and could be obtained by precise modelling of the
parts and inputting the available information of the part (such as the material used
for each of the respective parts and their material properties). Machine visualization
as generated by Matlab is shown in Fig. 9. The body blocks, coordinate frames, CGs,
etc. are shown in different shade.
The dynamics model of the FUSBOT-BS was modelled using Simulink blocks
based on the derivation as elaborated earlier. By using the Matlab model, the actual

Fig. 8 SimMechanics blocks for FUSBOT-BS base sub-system

J Intell Robot Syst

simulation structures

end position of the end-effector after a series of movements, given specific inputs,
could then be predicted. Thus, based on this information, the actual required torque
can be calculated and means can be derived to compensate the dynamic errors.
Matlab robust control toolbox was used in the robust controller modelling and
performance analysis of the axis [31]. It was inspired by the method developed by
Gu, Petkov and Konstatinov for general mechanical system [32]. For robust control
derivation, finding the control’s stabilization parameters is the most crucial part [34].
Assessment of the controller was done by using frequency response analysis with
robust performance and robust stability as the main focus in accessing the controller.
As each of the axes is similar and cascaded in structure and each of them is having its
own control input, a control simulation for one axis is similar for the rest of the sub-
systems with little modifications such as the value used for the weighting functions
and the possible perturbation parameters. Based on the kinematics and dynamics
derivation for the FUSBOT-BS given in Section 2.5, the robust controller derived
for the base sub-system is presented in the following.
The general expression of the dynamics of the system’s axis as derived in
Section 2.5:

  ··  · ·  ·  
A q q +B q q q +C q q 2 + G q + τ f v + τ f s = τ (19)

where q is the (nx1) vector of generalized joint coordinates, A is the (nxn) manipu-
lator inertia matrix, B is the (nx(n(n − 1)/2) Coriolis coefficient matrix and C is the
(nxn) centrifugal coefficient matrix, G is the (nx1) vector of gravitational forces, τ f v
is the (nx1) viscous friction, τ f s is the (nx1) Coulomb friction, τ is the vector of joint
control input torques to be designed.
J Intell Robot Syst

Despite the intensive measurements and tests, the physical parameters of the
system can not be known exactly. Thus, approximation of the parameterized uncer-
tainties was introduced to the axis at certain intervals as:
mc = mc (1 + pm δm ) ; α2 = α 2 (1 + pα2 δα2 ) ; k = k (1 + pk δk ) (20)
where, mc , α 2 , k are the nominal values of the mc , α2 , and k. pm , pα2 , pk and δm ,
δα2 , δk represent the degree of possible perturbations on the dynamics parameters.
In this case, a nominal value of 0.2 was chosen for the uncertainty ( p) while
possible perturbations were chosen to be −1 ≤ δ ≤ 1. These values were chosen by
approximation based on the degree of preciseness in modelling the system. The block
diagram of the FUSBOT-BS VA with the upper LFT blocks of uncertain parameters
is shown by Fig. 10. It is derived based on the base sub-system’s dynamics (such as
the inertia, friction) as represented by Eq. 19.
The transfer function matrix G f b s1 in Fig. 11 represents the nominal transfer
function of the base sub-system including model uncertainties. Δ is unknown, but
is assumed to be stable because it is parameterized and satisfies the condition
A∞ < 1. d is the external disturbances acting on the system. The transfer functions
from d to e as depicted by Eq. 19 have to be small in terms of the norm (·∞ ) for
every possible uncertainties represented by matrix Δ.
ep W p I + G f b s1 K
=  −1 d (21)
eu Wu K I + G f b s1 K
The block diagram of the VA with model-based controller and associated uncertain-
ties is shown in Fig. 11. The input to the transfer matrix G f b s1 has five elements
in it (three input uncertainty parameters, external disturbance and control) and the
output has six elements in it (three output uncertainty parameters, system error
e p , control error eu and controller input). The value of the weighting function was
obtained through trial and error. Finding the value for the weighting functions is a
crucial step in designing a robust controller especially in a complex system. There
is no unique solution to the weighting functions. The number of trials done and
the experience of the control engineers are contributing factor in choosing a good
weighting function value.
For the VA, the μ-synthesis was done using D-K iteration method of robust
controller due to its simplicity and good performance. Following the closed-loop
structure of the VA as shown in Fig. 11, P(s) denotes the transfer function of the

Fig. 10 Block diagram of the

base sub-system with uncertain
J Intell Robot Syst

Fig. 11 Base sub-system

closed-loop structure

matrix system G f b s1 with five inputs and six outputs. The block structure of P(s),
which is  F , can be written as follows:
 p := :  ∈ 3×3 ,  F ∈ C1x2 (22)
0 F
The uncertainty block  represents the uncertainties of the parameters in the
dynamics modelling of the system, the second block  F is the fictitious uncertainties
block introduced to represent the performance requirement for the μ-synthesis of
the robust controller. Following that, an optimization to reduce the maximum value
of μ is derived as follows:
min min  
K Dl (s) , Dr (s)  Dl (s) F L (P, K) Dr−1 (s)∞ (23)
stabilizing stable,min .phase

⎡ ⎤ ⎡ ⎤
d1 (s) 0 0 0 d1 (s) 0 0 0
⎢ 0 d2 (s) 0 ⎥ ⎢ ⎥
Dl (s) = ⎢
0 ⎥ Dr (s) = ⎢ 0 d2 (s) 0 0 ⎥
⎣ 0 0 d3 (s) 0 ⎦ ⎣ 0 0 d3 (s) 0 ⎦
0 0 0 d4 (s) I2 0 0 0 d4 (s) I2
where, d1 (s), d2 (s), d3 (s), d4 (s) are the scaling transfer functions. μ-synthesis is about
finding the smallest construction of stabilizing controller K in which the desired
performance can be achieved over a range of frequencies. This can be obtained
provided the following conditions are satisfied:

μ P F L (P, K) ( jw) < 1 (24)
 W p I + FU G f b s1 ,  K 
    −1  <1 (25)
 Wu K I + FU G f b s1 ,  K 

The μ-synthesis using D-K iteration was executed by using Matlab Robust Control
Toolbox. Table 3 shows the summary of the μ-controller D-K iteration. At the third
iteration, the peak-value of μ was equal to 0.983 for frequency range of 10−2 –104 .
The peak-value of μ is smaller than one indicating that the system is able to achieve

Table 3 μ-Controller D-K D-K iteration summary

iteration summary
Iteration number 1 2 3
Controller order 4 14 16
Total D-scale order 0 10 12
Gamma achieved 11.136 1.021 0.983
Peak μ value 2.613 1.019 0.983
J Intell Robot Syst

Fig. 12 Nominal and robust performances of the μ-controller

the required robust performance using the chosen controller (Eq. 24). The μ value is
achieved by sixteenth order controller which is pretty high. As one of the objectives
of the controller is to find the simplest and smallest stabilizing parameters, there is a
need to reduce the order of the controller to simplify the computations.

2.8 Analysis of Robust Performance of the Controller

Once the controller is derived, its robust stability and performance need to be
assessed. The robust stability of the controller is shown in Fig. 12. From the figure, it
can be seen that the μ-controller was robust and stable and Eq. 24 was satisfied. The

Fig. 13 Base sub-system performances with μ-controller

J Intell Robot Syst

performance of the system with controller is shown in Fig. 13 and it can be seen that
the robust performance was achieved by the system as the magnitude of the left hand
side of Eq. 25 was below the criterion (in this case is 1) over the specified frequency
The controller order at the third iteration was sixteen which was required to be
reduced while retaining the controller’s performance. In the case of the base sub-
system, the controller order could be reduced to fourth order by utilizing the Matlab
robust control toolboxes using the Hankel-norm approximation [31]. The fourth
order controller was able to give a similar performance as the sixteenth order.

3 Testing, Result and Discussion

3.1 Experimental Procedure

The existing control software system for the FUSBOT-BS system comprises:

1. Robot trajectory control

2. Image capture and display
3. HIFU control

The source code of the software system was written using Microsoft Visual C++
(MSVC). The program for robot motion comprises of few clusters.

1. Initialization where the robot axes are initialized and moved to the home
position, setting the motion parameters such as the default speed of the motor
and the PID default parameters
2. User selects treatment points through the user interface.
3. The robot guides the end-effector based on the point(s) selected on images
mapped to robotic workspace. Robot movement and treatment points are closed-
loop which cease to move under user instruction.

In the robot movement cluster, the modified control is added. The user is able to
choose whether to apply the default PID control, model-based control or the Hybrid
control (which consist of PID and model-based control as explained in the previous
section). For C++ and Matlab interfacing, appropriate files and their associated
library were included. For online computation, the time delay for the movement to
take place should be minimized.
The general block diagram of the system’s architecture showing the interaction
between the software, hardware and instrumentation is shown in Fig. 14. The user
selects the target point, treatment method, etc. from the GUI. The images shown in
the GUI are generated by the imaging system comprising of a camera and imaging
software as explained later in this section. Once the target and treatment method are
selected the control software will make necessary computation and moves the robot
manipulator to complete the specified task.
In order to analyze the performance of the hybrid supervisory controller,
experimental testing was done. A number of sets of predefined movement in all of
the four axes (VA, RA, HA and OA) were planned. These include different speeds,
directions and the range of motion. The parametric values were chosen based on the
J Intell Robot Syst


Computational Graphical User Interface Hardware

Image Initialization Imaging

Control Probe

Target Selection
Control Manipulator Move
Ref: Fig. 2

HIFU Treatment
Control Probes


Fig. 14 Block diagram of the robotic system’s architecture

kinematics structure and workspace constraints of the manipulator. In the test

different combinations of the parameters and their value were used for different
In order to measure the actual physical movement of the end-effector an imaging
system was employed. The imaging system used a camera (Intellicam [33]) to provide
real time images of the robot’s motions which were then captured by a customized
imaging software. The camera was secured to a platform at a predefined mounting
position based on a fixed set of reference points. The positions of the end-effector
before and after the movement were recorded. The measurements of the axes move-
ment were performed digitally using the images for accuracy and stability analysis.
Calibration was performed before every test to ensure its accuracy. The actual
movements of the end-effector measured from the images were then compared with
the simulation results of the FUSBOT-BS for the same input. Figure 15 shows the
schematic diagram of the experimental testing.

Fig. 15 Schematic diagram of

the of the test measurement
J Intell Robot Syst

Fig. 16 Combined VA, RA Experimental Error

and HA axes experimental 0.70

Actual Error (mm)

results 0.60
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
Test Number

3.2 Experimental Results

The outcomes of the accuracy test for each axis are shown in Figs. 16, 17, and 18. It
can be seen that in terms of accuracy, all of the axes of the manipulator are able to
give a good performance. The absolute errors are less than 0.5 mm for combined VA,
RA and HA axes, less than 0.35 degree for orientation axis and less than 0.3 mm for
imaging axis, when the model-based control was applied. All of the errors are well
within the tolerable error (which is less than 0.5 mm every axis) for the FUSBBOT-
BS system which uses 1 to 2 MHZ ultrasound transducer as its treatment probe. This
is due to the resolution limitation of the probe itself. There are some outliers e.g.
Fig. 16 point number 19 which might be due to some “hiccups” such as measuring
error or direct disturbance to the system either from internal (e.g. electrical power
inconsistency) or external (e.g. human measurement error) factors during the testing.
However, such occurrences are statistically insignificant (2 out of 90 points) and thus
the outliers can be safely neglected.
The comparison of the PID control and the model-based control in terms of the
accuracy in inner and outer region are shown in Fig. 19, 20, and 21. For the combined
VA, RA and HA axes of the manipulator (Fig. 19), model-based control proved to be
better in terms of achievable accuracy. Especially because of the vertical axis where
there is a slug when the break is released due to the weight of the manipulator, this
can be well compensated by the model-based control. Similarly for the distant points
(outer region of the workspace), model-based outperformed the non-model-based
PID control by giving about 30% more accurate results.
Figure 20 shows the comparison of control approaches for the orientation axis.
Overall both controls can be used since the errors are well within the tolerance (less
than 0.5 mm). However for the inner region, PID control performs better than the
model-based. This might be due to some assumptions in the derivation of dynamics,

Fig. 17 Orientation axis Experimental Error

experimental results 0.40
Actual Error (deg)

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
Test Number
J Intell Robot Syst

Fig. 18 Imaging axis Experimental Error

experimental results 0.35

Actual Error (mm)

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
Test Number

Fig. 19 Combined VA, RA Error Comparison

and HA axes error 0.6
Error (mm)

Inner Outer Overall
PID Model-Based Hybrid

Fig. 20 Orientation axis error Error Comparison

Error (mm)

Inner Outer Overall

PID Model-Based Hybrid

Fig. 21 Imaging axis error Error Comparison

comparisons 0.25

Error (mm)




Inner Outer Overall

PID Model-Based Hybrid

J Intell Robot Syst

because in the inner region some dynamics parameters such as gravity is modeled
as having little or no significant effects but in real practice the assumption might
not be true. For imaging axis (Fig. 21), the dynamics has little effect in the inner
region thus, PID and model-based performed similarly well. However, for the outer
region, model-based control performs better due to its rigorously modeled dynamics.
Overall, the performance of the model-based control is better compared to the PID
“Inner” refers to the inner circular region (color green) of the manipulator
workspace. “Outer” refers to the centre and outer circular region (color yellow and
red) of the manipulator’s workspace (Fig. 3). “Overall” refers to the combined inner
and outer.
In terms of repeatability, a very good performance is achieved with absolute errors
less than 0.1 mm. From the tests it was concluded that for the outer region, model-
based is a good choice and for inner region PID should be chosen due to its good
performance and less computation. Hybrid controller is able to make use of any of
these types of controllers within the specified workspace’s region. Thus, it is able to
get the best out of the two controllers for better overall system performance.

4 Discussions

In the hybrid approach, the choice of the control scheme is based on the surgi-
cal protocol since the type, extent and sequence of movements are different for
specific application scenarios. In certain cases, the trajectory planning and boundary
conditions may involve replaceable end-effectors. For instance, in the FUSBOT-
BS system, different work-space limits (please refer to Fig. 3) cater to different
applications: inner annulus for breast surgery and outer ones for trans-abdominal
reach. For each scenario, a suitable control approach under the family of controllers
is used. Based on the results presented in the previous section (Fig. 4), the PID
approach from the hybrid controller is chosen for breast application. However, for
the abdominal configuration, the end-effector has to go to the extreme end of the
available workspace, where as indicated in the experiments, the centre of gravity
and cantilever beam effect have significant contributions to the performance of the
system. Thus, in this case the model-based approach is more suitable and would be
automatically chosen by the hybrid controller. Test Results of Fig. 22 shows that the
inaccuracy is less than the error tolerance. It also proves the earlier claim that the
hybrid control works well for robotic system with more than one system’s dynamics
(for example system with more than one configuration or working in two or more
different environments).
The controller parameters were obtained through simulation and experimentally
tested for the system. The outcome was then used to refine the control parameters
until certain performance standards (e.g. accuracy and repeatability) were achieved.
Furthermore it was necessary that the experiments were done in such a way that the
parameters to be obtained could be excited, while having the other variables kept
constant and/or zero.
The application and hence the configuration can be predefined in the planning
stage before the surgery. However, in a possible scenario of a single application
using more than one annulus as well as parametric classification of system dynamics,
J Intell Robot Syst

Fig. 22 Error analysis of Error Measurement for Application Oriented

different FUSBOT-BS
configurations 0.44



Error (mm)



Abdominal Breast

the type of controller needs to be chosen automatically in real time application.

The added value of hybrid control approach is thus switching between appropriate
controllers by exploiting their relative advantage to improve the performance.
Secondly by having more than one controller, it can be extended for wider range of
application scenarios without having to change the basic configuration and control
architecture completely.
One of the main concerns in designing a controller is the stability of the controller
or the system, in general. For this particular control strategy the concern was
emphasized on the switching event when the parameterized controller is changed
according to the supervisory control decision. The possible instability due to this
reason was unproven during the experiments and simulations. This might be due
to the simplicity of the supervisory control and small number of “plant events” and
parameterized controllers used. An increase in the number of the “plant events”
and parameterized controllers and their effect to the system instability would be
interesting to be studied further. The weighting (gain) parameters have to be chosen
carefully in order to achieve the robust performance of the controller and the system.
Tuning experiences of the control engineers and the amount of ‘trial and errors’ done
are few factors that contribute to finding a set of robust control parameters that could
give good performance.

5 Conclusions

In this paper a common and optimal control hierarchy and strategy is proposed for
non-invasive medical robotic systems. In terms of safety, there are many factors
which contribute to the safety of the robot such as the mechanical design of the ma-
nipulator, the type of controller, the electronic hardware and trajectory planning for
fail-safe software design, etc. Control is one of the significant factors in determining
the safe performance of the robot. Medical applications demand complex protocols
and optimal trajectory selections, which may involve hybrid configurations. In this
paper, a hybrid supervisory control is presented. Using the FUSBOT-BS system as
the test-bed it showed that it had a better performance compared to the stand-alone
J Intell Robot Syst

controller. Hybrid controller has also been cited to be able to give better result where
dynamic of the system change considerably in its application [16, 17].
Model based control has been used in many medical robotic systems such as
in Robodoc, robots for laparoscopic surgery, etc. Meanwhile PID control is also
commonly used in wide range of applications due to its simplicity. The analysis
of the test data showed favourable result, in which the accuracy, repeatability and
stability of the robotic system were shown to be achieved within desirable limits. It
was also observed that hybrid controller worked well for switching between different
In order to establish the proposed control scheme, it would be desirable to
evaluate systems with higher complexity for statistically large numbers of tests,
especially for invasive type of medical robotic applications. It is also recommended
that other control strategies such as fuzzy logic control to be combined for example
with the model-based control or to be added into the list of the parameterized family
of controllers of the hybrid supervisory control [38]. The performance of different
control combinations can be used as measurement parameters for establishing
generic performance standards.

Acknowledgements The authors would like to thank Ministry of Education, Singapore and Agency
of Science Technology and Research, Singapore for jointly funding the project.


1. Camarillo, D.B., Krummel, T.M., Salisbury, J.K. Jr.: Robotic technology in surgery: past, present,
and future. Am. J. Surg. 188, 2–15 (2009). doi:10.1016/j.amjsurg.2004.08.025
2. Taylor, R.H., Stoianovici, D.: Medical robotics in computer-integrated surgery. IEEE Trans.
Robot. Autom. 19(5), 765–781 (2003). doi:10.1109/TRA.2003.817058
3. Cleary, K., Nguyen, C.: State of the art in surgical robotics: clinical applications and technology
challenges. Comput. Aided Surg. 6(6), 312–340 (2001)
4. Swandito, G.G.N.: Kumar, Chauhan, S.: Control hierarchy of medical robotic systems for non-
invasive for surgery. In: Proceedings of the 12th International Conference on Biomedical Engi-
neering, Singapore (2005)
5. Mishra, R.K., Chauhan, S.: Safety of surgical robots: a fundamental aspect. In: Proceedings of the
12th ISMCR—Towards Advanced Robot Systems and Virtual Reality, Bourges, France (2002)
6. Kazanzides, P., Zuhars, J., Mittelstadt, B., Taylor, R.H.: Force sensing and control for a surgical
robot. In: Proceedings of the IEEE International Conference on Robotics and Auto, Nice,
France, pp. 612–617 (1992)
7. Kennedy, C.W., Desai, J.P.: Model-based control of the Mitsubishi PA-10 robot arm: application
to robot-assisted surgery. In: Proceedings of the IEEE International Conference on Robotics
and Auto, New Orleans, LA, pp. 2523–2528 (2004)
8. Zemiti, N., Ortmaier, T., Morel, G.: A new robot for force control in minimally invasive surgery.
In: Proceedings of the IEEE International Conference on Intelligent Robots and Systems,
Sendai, Japan, pp. 3643–3648 (2004)
9. Gu, J.J., Meng, M., Cook, A., Faulkner, M.G., Liu, P.X.: Sensing and control of robotic prosthetic
eye for ocular implant. In: Proceedings of the 26th IEEE International Conference on Intelligent
Robots and Systems, Hawaii, USA, pp. 2166–2171 (2001)
10. Ginhoux, R., Ganglouf, J., Mathelin, M., Soler, L., Sanchez, M.M.A., Marescaux, J.: Active
filtering of physiological motion in robotized surgery using predictive control. IEEE Trans.
Robot. 21(1), 67–79 (2005). doi:10.1109/TRO.2004.833812
11. Zhu, W.H., Salcudean, S.E., Bachmann, S., Abolmaesumi, P.: Motion/force/image control of a
diagnostic ultrasound robot. In: Proceedings IEEE International Conference on Robotics and
Auto, San Francisco, CA, pp. 1580–1585 (2000)
12. Ang, K.H., Chong, G., Yun, L.: PID control systems analysis, design, technology. IEEE Trans.
Control Syst. Technol. 13(4), 559–576 (2005). doi:10.1109/TCST.2005.847331
J Intell Robot Syst

13. Mishra, R.K.: FUSBOT-BS: Technical Manual. Robotics Research Centre, Nanyang Technolog-
ical University, Singapore (2004)
14. Chauhan, S.: A HIFU medical robotic system for organotripsy and tissue ablation – the
FUSBOT. In: Proceedings International Conference on Computing, Communication, and Con-
trol Technologies, Austin (Texas), USA (2004)
15. Whitcomb, L.L., Arimoto, S., Naniwa, T., Ozaki, F.: Adaptive model-based hybrid control of
geometrically constrained robot arms. IEEE Trans. Robot Autom. 13, 105–116 (1997)
16. Mills, J.K.: Hybrid control: A constrained motion perspective. J. Robot. Syst. 8(2), 135–158
(1991). doi:10.1002/rob.4620080202
17. Koutsoukos, X.D., Antsaklis, P.J., Stiver, J.A., Lemmon, M.D.: Supervisory control of hybrid
systems. Proc. IEEE 88(7), 1026–1049 (2000). doi:10.1109/5.871307
18. Morse, A.S.: Control Using Logic-Based Switching Trends in Control: A European Perspective,
pp. 69–113. Springer, London (1995)
19. Enste, U., Epple, U.: Hybrid structure in process control. In: Proceedings of the American
Control Conference, San Diego, California, pp. 4482–4485 (1999)
20. Astrom, K.J., Hagglund, T., Hang, C.C., Ho, W.K.: Automatic tuning and adaptation for
PID controllers—a survey. Control Eng. Pract. 1(4), 699–714 (1993). doi:10.1016/0967–0661(93)
21. Shigemasa, T., Yukitomo, M., Kuwata, R. A model driven PID Control system and its case
studies. In: Proceedings of the IEEE International Conference on Control Application, Glasgow,
UK, pp. 571–576 (2002)
22. Garcia, C.E., Carelli, R., Postigo, J.F., Soria, C.: Supervisory control for a telerobotic sys-
tem: a hybrid control approach. Control Eng. Pract. 11, 805–817 (2003). doi:10.1016/S0967-
23. Cao, C.W.: Supervisory control of a class of hybrid dynamic systems. IEEE 22, 967–970 (1993)
24. Khalil, W., Kleinfinger, J.F.: A new geometric notation for open and closed-loop robots. In:
Proceedings IEEE Conference on Robotics and Automation, San Francisco, CA, pp. 1174–1179
25. Schilling, R.J.: Fundamentals of Robotics: Analysis and Control. Prentice-Hall, Singapore (1990)
26. Kozlowski, K.: Modelling and Identification in Robotics: Advances in Industrial Control.
Springer, Great Britain (1998)
27. Schiavicco, L., Siciliano, B.: Modelling and Control of Robot Manipulators. Prentice-Hall,
Singapore (2000)
28. Olsson, H., Astrom, K.J., De Wit, C.C., Gafvert, M., Lischinsky, P.: Friction models and friction
compensation. Eur. J. Control. 4(3), 176–195 (1998)
29. Dupont, P., Armstrong, B., De Wit, C.C.: A survey of models, analysis tools and compensation
methods for the control of machines with friction. Int. J. Autom. 30(7), 1083–1138 (1994).
30. De Wit, C.C., Lischinsky, P.: Adaptive friction compensation with partially known dynamic
friction model. Int. J. Adapt Control Signal Process. 11, 65–80 (1997). doi:10.1002/(SICI)
31. The MathWorks Inc: US. MATLAB. (2007)
32. Gu, D.W., Petkov, P.H.R., Konstantinov, M.M.: Robust Control Design With MATLAB.
Springer, London (2006)
33. Intellicam System: IntelliPIX. (2007)
34. Qu, Z.H.: Robust Control of Nonlinear Uncertain System. Wiley Series in Nonlinear Science.
Wiley-Interscience, USA (1998)
35. Cadic, M.: Strongly robust adaptive control: the strong robustness approach. Dissertation in
partial fulfillment of the requirements of the Dutch Institute of Systems and Control (DISC)
for graduate study, Twente University Press, The Netherlands (2003)
36. Shinners, S.M.: Advance Modern Control System Theory and Design. Wiley-Interscience, USA
37. Galil Motion Controller, U.S.A.: Manuals and command references.
38. Mahfouf, M., Abbod, M.F., Linkens, D.A.: A survey of fuzzy logic monitoring and control
utilisation in medicine. Artif. Intell. Med. 21, 27–42 (2001). doi:10.1016/S0933-3657(00)00072-5