Bilateral Control With Virtual Models

Hasala Ranmal Senevirathne (060452N)

Department of Electrical Engineering
University of moratuwa
Srilanka.
hasalasri@gmail.com

Abstract — Researches on Teleoperated robot have been carried 2. NEED OF VIRTUAL MODELS
out for a long time and it has been developed in several aspects by
now. Bilateral control is one of the areas which is researched under Most of early bilateral robotic systems were totally mechanical.
teleoperated robots. This paper consist of discussion on how That means the movements from the master and force feedback
bilateral control is used in teleoperated robots and a derivation a from the slave was transferred using mechanical systems.
virtual model for a simple bilateral control system with two rods Therefore it wasn’t possible to keep the slave robot at larger
connected by a torsion spring. distances. In modern days we have robots in space, under-sea
and hazardous environments. In order to improve the
Key words—Bilateral, Teleoperation, Force Feedback, Tele- performance of these robots, it is needed to equip them with
manipulation bilateral control systems. In these cases mechanical bilateral
control systems are not possible. So what is the solution? The
1. INTRODUCTION solution is to capture the movements of the master and transfer
them to the slave through a communication medium and
Tele-operation or Tele-manipulation is one of the most replicate them at the slave using relevant actuators, On the
researched areas in the field of robotics. Teleoperated robotic other hand feedback from the slave should be captured by
systems have been used in many real world applications such sensors and transferred through the communication medium
as explosive disposal, hazardous material handling. On the and replicate them at the master’s side using the actuators
other hand autonomous robots have been successfully there. To implement this type of bilateral control system it is
implemented in applications such as automotive needed to obtain a virtual model for the system. For a simple
manufacturing offering high productivity reduced cost and bilateral system mathematical model of its mechanical
counterpart can be used as the virtual model. The virtual model
higher quality. A teleoperated system or a human involved should generate the same effect as the master at slave’s side
system offers more feasible solution when considering more and generate the same feedback of the slave at the master’s
dynamic real world tasks in unpredictable environments. The side. Here a simple bilateral control system was modeled
Human interaction overcome the intelligence limitations of the virtually in order to simulate it in a computer.
robotic system by introducing desired capabilities to the
robotic system, such attributes are adaptability to a variety of 3. METHODOLOGY
environments, high level of intelligence. In tele-manipulation
human operator has to perform a given task on a remote The system have to be modeled is a vertically mounted two
environment. To capture the commands given by the operator similar rods connected by a torsion spring (Figure 3.1).
a local robotic interface is used, called the master. When
operator commands the local robot, it captures the commands
and transmits them to the remote robotic interface through a
communication medium. This remote robotic interface is
known as the slave. Slave’s task is to replicate the movements
of the master. The operator’s experience close to real if there
is a method to operator to know about the interactions of the
slave with the environment. Feedback information from the
slave can be obtained in several ways such as visual displays.
The ideal method to improve the operator’s ability is to
feedback the forces acting between the environment and the Figure 3.1
slave [5]. “When the force at the slave side is reflected back to
the human operator it is said that tele-manipulation is
controlled bilaterally” [1]. Block diagram of a bilateral control It was assumed that frictional are not present in the system and
system is shown figure 1.1 [2]. the system is always at rest before it moves. When developing
a mathematical model for virtual model we have to consider in
which manner a movement of rod1 is replicated at rod2 and on
the other hand we have to consider the feedback force exerted
on the rod 1 due to the inertia of the rod 2. Consider rod 1 is

Figure 1.1
moved by an angle of θ1 and let’s take the rod2 is rotated by (5)
an angle of θ1 in response to master’s movement (Figure 3.2).
τ ( S ) I 1 S − 2 KI 1 S
2 4 2

=
θ 2( S ) K
And also the transfer function for the reaction torqur exerted by
the slave on the master is given by

(6)
τ ' (S )
= I 1S 2
Figure 3.2 θ 2( S )
Applying = I for rod 1 and rod2
τ ..
By virtually implementing these functions the similar operation
as the mechanical system can be obtained when the spring is
θ not present in the system. The next step of modeling system is
to simulate the system in a computer environment with the help
where - Torque, I – Inertia, - angle ; of the derived mathematical model. There, it can be observed
τ the behavior of the slave according to the various force applied
to the master and forces exerted on master by the slave.
θ
III. SYSTEM ANALYSIS
(1)
When the input is measured as a torque
..
I 1θ 1 = τ − k (θ 1− θ 2) Using (5)

(2)
.. θ 2( S ) K / I12 (7)
= 2 2
I 2 θ 2 = k (θ 1− θ 2) τ ( S ) S {S − (2 K / I1 )}

Taking the laplacian of (1) and (2) we get, There are poles at S = 0 and
(3)
S = ± ( 2 K / I1 )
..
I 1[ S θ 1(s ) − Sθ 1(0) − θ 1(0)] + kθ 1( S ) − kθ 2( S ) = τ ( S )
2

This can be graphically interpreted on a S plane as follows

(4)
.. jω

I 2[ S 2θ 2( s ) − Sθ 2(0) − θ 2(0)] − kθ 1( S ) + kθ 2( S ) = 0

Since a system which is initially at rest,

σ
− (2 K / I 1 ) 0 + (2 K / I1 )
.. ..
θ 1(0),θ 2(0) = 0
Figure 3.3
Since two rods are equal. Examining the locations of poles of S-plane it can be noted
that there is a exponentially increasing term and exponentially
decreasing term. But, since magnitudes of locations where
I1 = I 2 poles located, are equal these exponential terms are cancelled
0ut. So it’s difficult to comment on the stability of the system
By (3) and (4) transfer function of the master’s force to slave’s using those poles. Therefore we have to consider the S=0 pole
position can be derived, which governs the impulse response of the system. This can be
seen by deriving the impulse response of the system.
Using partial fraction for (7)
(8)
H(S)=kI2-1/2kIS2-1/(2k/I)32S+2k/I+1/(2k/I)32S-2k/I rod to a some angular position and then moving it back to the
original position in very small time period.

Then taking inverse Laplace transform function gives the 3. DIGITAL IMPLEMENTATION
impulse response of the system.
ht=L-1[Hs] For bilateral tele-manipulaton we have to use digital systems
in order to capture and process data and regenerate the forces
ht=-kI212kIt-1/(2k/I)32e-2k/It++1/(2k/I)32e2k/It (9) and feedbacks. The features of the digital system decide how
well the bilateral system works. That is how closer the system
work to the actual scenario. The systems quality heavily
Since the exponential terms are cancelled out. -12I3t term depends on the processing speed of the digital system. It
dominates the impulse response of the system. It can be noted
that is a constantly increasing term (not considering the decides the continuity of the system. If we are using a system
magnitude). Practically, this scenario can be noted. Due to with low processing speed we can’t expect a continuous
impulse torque slave tend to move at a constant angular speed. operation from the system. If the processing speed is too low
That angular speed depends only on the inertia of the system. compared to the speed of the movements of the system even
stability problems can arise. So when designing a suitable
When the input measured as an angle digital system for bilateral control systems we have to use a
consider it’s processing speed by analyzing the features of the
control system. In addition to the digital system ,the system’s
θs( s ) k/I performance depend on the features of the sensors used.
= 2 (10)
Generally sensors such as position sensors (Angular, Linear)
θm( s ) S + k / I and force, torque sensors used in less complicated bilateral
systems. The responsiveness and the resolution of the sensors
There are poles at and are very critical for a system. In order to transmit the master’s
date to slave and feedback of slave’s to the master, the sensors
S = ± (K / I ) should sense the data without any delays. A little delay could
make the system unstable. And also the resolutions of the
sensors should be at an acceptable level which helps to
Graphically interpreting this on S- plane reproduce the action of the master at slave’s side without any
discontinuities. If a sensors with low resolutions used small
jω deflection or small forces on the master’s side are not
transferred to the slave which reduces the performance of the
system.

In this paper two algorithms have been introduced. One for

S =− K/I S =+ K/I
σ using a torque sensor and other using angular position sensor.
System has an actuator (this actuator take inputs of angular
positions) at the slave’s side to alter the position according to
Figure 3.4 the signals from the master. Moreover there is also an actuator
This also gives rise a impulse response of with two to generate feedback torque from the slave at the master side.
exponential terms, decreasing one and an increasing one. But
there is not a pole at S = 0 as the previous case .Therefore the
exponential terms cancelled out and nothing left that is h. This 3. 1 IMPLEMENTATION USING A POSITION SENSOR
scenario can be cleary seen by examining the impulse response.

Using partial fraction for (10)

H(S)=KI1/(-2k/I)S+k/I)+ 1/(2k/I)S-k/I)
(11)
Then taking the inverse laplace of (11)
ht=KI2K/Iek/I) t- e-k/I) t (12)
since exponential terms are cancelled out the (12) becomes
ht=0

That means there is no movement of the slave in response to a

impulse angle change of the slave. In practice this scenario can
be experienced when impulse movement is applied to the
master. Impulse input of angle can be considered as moving the
 Then the angle angular deflection of slave, δθS t is given
by

δθS t= 2Tt-δt(δt)2+K(δt)2δθMt+ 2IδθS t-δt2I+K(δt)2

T
δƟs
Using the above equation an algorithm can be implemented
T for this bilateral system.
δƟm
Algorithm for this is as follows,
Ʈ If the system is initially at rest.

The starting torque T = 0;

Figure 3.5 And θs t-δt= 0;

Let’s take a force is applied to the master rod for a small time We have to sample the acquired data from the sensors from
regular intervals, δt. The sampling interval depends on
period(δt) and it is rotated by an angle of δƟM due to that frequency of outputs of the sensors or the processing speed of
force, and let’s consider the slave rod is deflected by an angle the digital system used. For the algorithm inertia, spring
of δƟS due to the force transferred through the torsion spring. constant, sampling time should be initialized first.
Let’s take the tension of the torsion spring before applying the
force is T(t-δt), Initialize I,K, δt
Then tension of the spring after a time period of δt
θMt-δt = 0;
Tt-δt = 0;
Tt-δt+K{δθMt– δθS t} (13) θs t-δt= 0;
Then the average tension during the time period δt while(true)
Tt-δt+ K{δθMt– δθS t}2 (14)
{
Considering Slave’s acceleration and applying the Newton’s
δθMt= θMt-θMt-δt
central difference formula for second derivate
// angular deflection of master during a sampling time period

θs= θs t+δt - 2θs t + θs t-δt(δt)2 (15) // θMt and θMt-δt obtained from position sensor of the master

θS={θS t+δt- θS t} –{θS t - θS t-δt} δθS t= 2Tt-δt(δt)2+K(δt)2δθMt+ 2IδθS t-δt2I+K(δt)2

But, // angular deflection of the slave

Tt-δt = Tt-δt+K{δθMt- δθMt}

θS t+δt- θS t= δθS (t) (16)
// This is the feedback torque and this is generated at the master using
(17) an actuator
θS t - θS t-δt= δθS (t-δt)
//starting torque for the next time period
Then applying (16) and (17) in (15) we get
θSt= θSt+ δθS
θs= δθS t- δθS (t-δt) (δt)2 (18) // current angular position of slave
Applying = I for slave, and substituting from (18)
τ //This value is given to the slave’s actuator
..
θ δθS t-δt= δθS t

}
I δθS t- δθS (t-δt) (δt)2= Tt-δt+ K{δθMt– δθS t}2
when the torsion spring is initially stressed
(19)
The same algorithm can be used when the spring is stressed
due to a force. But the initial values are different . θMt-δt and

(20)
θs t-δt should be known and then the Tt-δt can be found using Pδt2I=δθS t2Ikδt2+ 2+δθS t-δt-4Ikδt2-2+δθS t-2δt2Ikδt2+Tt-
the relationship δt-2k +Tt-2δt{2k }
Tt-δt=θMt-δt-θs t-δt
Then we can get δθS t as follows (27)

3. 1 IMPLEMENTATION USING A TORQUE SENSOR δθS t=Pδt2I-δθS t-δt-4Ikδt2-2-δθS t-2δt2Ikδt2-Tt-δt-2k -Tt-

2δt{2k }2Ikδt2+ 2
When a torque sensor is used to measure the applied torque to
the master, instead of a angular position sensor, slave’s Algorithm when a torque sensor is used,
behavior is determined according to the input torque pattern to (28)
the master. For accuracy it’s essential to have a torque sensor If the system is initially at rest
with lowest minimum measurable force. And also the Tt-δt=0,Tt-2δt=0,θS t-δt=0, θS t-2δt=0
frequency of feeding data to the digital systems also influences
the accuracy of the system. Initialize I,k, δt
Considering Figure 3.5, if a torque of P applied to master While(true)
during an interval δt, and it can be assumed that the applied
torque p is constant during this time interval. And let’s take {
the tension of the torsion spring is T before torque P applied.
Tension of the spring when the master and slave is deflected δθS t=Pδt2I-δθS t-δt-4Ikδt2-2-δθS t-2δt2Ikδt2-Tt-δt-2k -Tt-
by angles of θM and θs respectively, 2δt{2k }2Ikδt2+ 2
Tt-δt+K{δθMt– δθS t} δθS t-2δt=δθS t-δt
δθS t-δt=δθS δt
average torque during the time period δt,
Tt-2δt=Tt-δt
// setting Previous T to present T
Tt-δt+ K{δθMt– δθS t}2
Tt-δt = Tt-δt+IδθS t- δθS t-δtδt2-kδθS t2
Applying =I for slave, and substituting from 1 and 2
τ // New present T
.. // This is the feedback torque and this is generated at the master
θ using an actuator
θSt= θSt+ δθS
I δθS t- δθS (t-δt) (δt)2= Tt-δt+ K{δθMt– δθS t}2 // current angular position of slave
//This value is given to the slave’s actuator
(21)
Using equation (21) we can find δθMt }

δθMt= 2I{δθS t- δθS (t-δt)}k(δt)2- 2T(t-δt)k+δθS t when the torsion spring is initially stressed t,he same algorithm
can be used but the initial values are different . θMt-δt and θs t-
(22) δt should be known and then the Tt-δt can be found using the
Applying τ=Iθ for slave relationship Tt-δt=θMt-δt-θs t-δt

T=IθS (23)
Applying τ=Iθ for master 4. SIMULATION AND RESULTS
P-T=IθM (24)
Using 1 and 2 The algorithm using a position sensor was implemented in
matlab and the output values were graphically plotted altering
P-IθS=IθM (25) the parameters of the system and noticeable results could be
Using Newton’s central difference formula for second observed. Graphs were obtained for the two behaviors when
derivative master rod moved at a constant speed and sinusoidally moved

P(δt)2I= δθM t- δθM t-δt+ δθS t- δθS t-δt (26)

Applying (26) in (25) When the master rod is moving at a constant angular speed
Input
 actuators and sensors and also the data processing and
communication delays. At slower speeds nearly similar
behavior to the mechanical system could be observed. But at
higher speeds, faster data acquisition and processing methods
actuators and sensors with low response time should be used
in order to obtain that behavior. Communication delay is the
one of the significant issue in tele-operated bilateral control
systems. When the master and slave are located far away from
each other delays can arise when transmitting data through
communication medium. As an example when the tele-
Figure 4.1
operation uses the a packet switching network such as internet
as the communication medium there can be considerable
The response of the slave for a constant speed of a master is as amount of delay when transferring data from master to slave
follows and slave to master. When this delay can’t be neglected the
Output stability of the system can’t be guaranteed. [4]

There are two well known bilateral control architectures used

at present, there are position-position architecture and
position-force architecture.

Position to Position Architecture [3]

In this architecture both robots are programmed to track each

other’s position. The slave imitates the movements of the
master and also slave transfer the forces acting on it back to
the master. If master want to feel force acting on slave’s side
that force should be large enough to move the slave
Figure 4.2 actuators of the slave. If it doesn’t there is no method to
When the master is moved sinusoidally master to sense the forces acting on the slave. So this doesn’t
agree with the concept of force feedback all the time.
Input Position-Force Architecture [3]
In this architecture master can sense all forces acting on the
slave whether they move the slave’s actuators or not. But this
architecture is less stable than the previous one.

The system modeled above is a system with a position-

position architecture because a torque should move the
system if the force should be sensed.

REFERENCES

[1] Cristian Secchi,Stefano Stramigioli,Cesare Fantuzzi,“Control of

Interactive Robotic Interfaces”,Springer,2007
Figure 4.3
[2] Horan, Ben, Creighton, Douglas, Nahavandi, Saeid and Jamshidi, Mo
2007, Bilateral haptic teleoperation of an articulated track mobile robot,
The response of the slave for a sinusoidal variation of a master in IEEE International Conference on System of Systems Engineering,
‘s position is as follows 2007. SoSE '07., IEEE Xplore, Piscataway, N.J, pp. 1-8.
[3] Bruno Siciliano, Oussama Khatib ,” Handbook of Robotics”,
Springer,2008
Output William R. Feme11,“Remote Manipulation with Transmission Delay,”
National Aeronautics and Space Administration.Washington, D. C.,1965
[4] Thorsten A. Kern,”Engineering Haptic Devices”,Springer,2009

5. CONCLUSION

When implementing the above system physically using the

virtual model. The same behavior as the mechanical system.
can’t be expectedFigure
exactly
4.4 This happens due to delays of
 For bilateral tele-manipulaton we have to use digital systems
in order to capture and process data and regenerate the forces
and feedbacks. The features of the digital system decide how
well the bilateral system works. That is how closer the system
work to the actual scenario. The systems quality heavily
depends on the processing speed of the digital system. It
decides the continuity of the system. If we are using a system
with low processing speed we can’t expect a continuous
operation from the system. If the processing speed is too low
compared to the speed of the movements of the system even
stability problems can arise. So when designing a suitable
digital system for bilateral control systems we have to use a
consider the it’s processing speed by analyzing the features of
the control system. In addition to the system the system’s
performance depend on the features of the sensors used, most
commonly sensors such as position sensors(Angular, Linear)
and force, torque sensors used in less complicated bilateral
systems. The responsiveness and the resolution of the sensors
are very critical for a system. In order transmit the master’s
date to slave and feedback of slave’s to the master, the sensors
should sense the data without any delays. A little delay could
make the system unstable. And also the resolutions of the
sensors should be at a acceptable level which helps to
reproduce the action of the master at slave’s side without any
discontinuities. If a sensors with low resolutions used small
defelction or small forces on the master’s side arent’s
transferred to the slave which reduces the performance of the
system.

In this paper the system has a position sensor at the master’s

side and and actuator (this actuator take inputs of angular
positions) at the slave’s side to alter the position according to
the signals from the master. Moreover there is also an actuator
to generate feedback torque from the slave at the master side.

T
δƟs

T
δƟm

Let’s take a force is applied to the master rod for a small time
period(δt) and it is rotated by an angle of δƟM due to that
force, and let’s consider the slave rod is deflected by an angle
Software Implementation of the system of δƟS due to the force transferred through the torsion spring.
Let’s take the tension of the torsion spring before applying the
force is T(t-δt),
Then tension of the spring after a time period of δt Initialize I,K, δt

Tt-δt+K{δθMt– δθS t} θMt-δt = 0;

Tt-δt = 0;
Then the average tension during the time period δt θs t-δt= 0;

while(true)
Tt-δt+ K{δθMt– δθS t}2 (2)
{
Considering Slave’s acceleration and applying the Newton’s
central difference formula for second derivate δθMt= θMt-θMt-δt
// angular deflection of master during a sampling time period

θs= θs t+δt - 2θs t + θs t-δt(δt)2 (4) // θMt and θMt-δt obtained from position sensor of the master

θS={θS t+δt- θS t} –{θS t - θS t-δt}

δθS t= 2Tt-δt(δt)2+K(δt)2δθMt+ 2IδθS t-δt2I+K(δt)2
But, // angular deflection of the slave

θS t+δt- θS t= δθS (t) Tt-δt = Tt-δt+K{δθMt- δθMt}

// This is the feedback torque and this is generated at the master using
θS t - θS t-δt= δθS (t-δt) an actuator

//starting torque for the next time period

Then applying 1 and 2 in 3 we get
θSt= θSt+ δθS
θs= δθS t- δθS (t-δt) (δt)2
Applying = I for slave, and substituting from 1 and 2 // current angular position of slave
τ .. //This value is given to the slave’s actuator
θ δθS t-δt= δθS t

I δθS t- δθS (t-δt) (δt)2= Tt-δt+ K{δθMt– δθS t}2 }

when the torsion spring is stressed

Then the angle angular deflection of slave, δθS t is given The same algorithm can be used when the spring is stressed
due to a force. But the initial values are different . θMt-δt and
by θs t-δt should be known and then the Tt-δt can be found using
the relationship
δθS t= 2Tt-δt(δt)2+K(δt)2δθMt+ 2IδθS t-δt2I+K(δt)2 Tt-δt=θMt-δt-θs t-δt

Simulation Results
Using the above equation an algorithm can be implemented The algorithm was implemented in matlab and the output
for this bilateral system. values were graphically plotted altering the parameters of the
Algorithm system and noticeable results could be observed. Graphs were
obtained for the two behaviors when master rod moved at a
constant speed and sinusoidally moved.
If the system is initially at rest.
When the master rod is moving at a constant angular speed
The starting torque T = 0;
And θs t-δt= 0;

We have to sample the acquired data from the sensors from

regular intervals, δt. The sampling interval depends on
frequency of outputs of the sensors or the processing speed of Input
the digital system used. For the algorithm inertia, spring
constant, sampling time should be initialized first.
 average torque during the time period δt,

The response of the slave for a constant speed of a master is as Tt-δt+ K{δθMt– δθS t}2
follows Applying =I for slave, and substituting from 1 and 2
τ ..
θ

I δθS t- δθS (t-δt) (δt)2= Tt-δt+ K{δθMt– δθS t}2

Using equation 1 we can find δθMt

δθMt= 2I{δθS t- δθS (t-δt)}k(δt)2- 2T(t-δt)k+δθS t

When the master is move sinusoidally

Applying τ=Iθ for slave
T=IθS
Applying τ=Iθ for master
P-T=IθM
Using 1 and 2
P-IθS=IθM
Using Newton’s central difference formula for second
derivative
P(δt)2I= δθM t- δθM t-δt+ δθS t- δθS t-δt

Using 1 and 2 we get

Applying 1 in 2
Pδt2I=δθS t2Ikδt2+ 2+δθS t-δt-4Ikδt2-2+δθS t-2δt2Ikδt2+Tt-
δt-2k +Tt-2δt{2k }
Then we can get δθS t as follows

When the master has a Torque Sensor δθS t=Pδt2I-δθS t-δt-4Ikδt2-2-δθS t-2δt2Ikδt2-Tt-δt-2k -Tt-
2δt{2k }2Ikδt2+ 2
Algorithm
When there is a sensor to measure the applied torque to the
master, instead of a angular position sensor, slave behavior is If the system is initially at rest
determined according to the input torque pattern to the master. Tt-δt=0,Tt-2δt=0,θS t-δt=0, θS t-2δt=0
For accuracy its essential to have a torque sensor with lowest
minimum measurable force. And also the frequency of feeding Initialize I,k, δt
data to the digital systems also influence the accuracy of the
system. While(true)
Considering figure one, if a torque of P applied to master {
during an interval δt, and it can be assumed that the applied
torque p is constant during this time interval. And let’s take
the tension of the torsion spring is T before torque P applied. δθS t=Pδt2I-δθS t-δt-4Ikδt2-2-δθS t-2δt2Ikδt2-Tt-δt-2k -Tt-
Tension of the spring when the master and slave is deflected 2δt{2k }2Ikδt2+ 2
by angles of θM and θs respectively δθS t-2δt=δθS t-δt
Tt-δt+K{δθMt– δθS t} δθS t-δt=δθS δt
Tt-2δt=Tt-δt Applying =I for rod 1 and rod2
// setting Previous T to present T τ ..
Tt-δt = Tt-δt+IδθS t- δθS t-δtδt2-kδθS t2 θ
// New present T
// This is the feedback torque and this is generated at the master where - Torque, I – Inertia, - angle ;
using an actuator τ
θSt= θSt+ δθS θ
// current angular position of slave (1)
//This value is given to the slave’s actuator
..
} I 1θ 1 = τ − k (θ 1− θ 2)

when the torsion spring is initially stressed t,he same algorithm (2)
can be used but the initial values are different . θMt-δt and θs t- ..
δt should be known and then the Tt-δt can be found using the
relationship Tt-δt=θMt-δt-θs t-δt I 2 θ 2 = k (θ 1− θ 2)

Taking the laplacian of (1) and (2) we get,

(3)
..
III. METHODOLOGY I 1[ S 2θ 1( s ) − Sθ 1(0) − θ 1(0)] + kθ 1( S ) − kθ 2( S ) = τ ( S )
(4)
The system have to be modeled is a vertically mounted two
similar rods connected by a torsion spring (Figure 3.1). ..
I 2[S 2θ 2( s ) − Sθ 2(0) − θ 2(0)] − kθ 1( S ) + kθ 2( S ) = 0

Since a system which is initially at rest,

.. ..
θ 1(0),θ 2(0) = 0
Since two rods are equal.
Figure 3.1

I1 = I 2
It was assumed that frictional are not present in the system and
the system is always at rest before it moves. When developing By (3) and (4) transfer function of the master’s force to slave’s
a mathematical model for virtual model we have to consider in position can be derived,
which manner a movement of rod1 is replicated at rod2 and on
the other hand we have to consider the feedback force exerted (5)
on the rod 1 due to the inertia of the rod 2. Consider rod 1 is θ 2( S ) K
moved by an angle of θ1 and let’s take the rod2 is rotated by = 2 4
an angle of θ1 in response to master’s movement (Figure 3.2). τ ( S ) I1 S − 2 KI1 S 2
And also the transfer function for the reaction force exerted by
the slave on the master is given by

(6)
τ ' (S )
= I 1S 2
θ 2( S )

Figure 3.2 In equation (5) its been considered that input to the system is
the torque applied to the master. That means there is a torque
sensors at the master’s side.If the input needed is the angular jω
deflection of the master ,it should be equipped with an angular
position sensor . The transfer function can be derived as
follows when angular position is used as the input.

Taking the laplacian of (2) and considering system is initially

at rest i.e ( we get
.. − (2 K / I 1 ) 0 + (2 K / I1 )
σ
Sθ1(0) = 0, θ1(0) = 0

I 1[ S 2θ 1( s )] + kθ 1( S ) − kθ 2( S ) = τ ( S ) Examining the locations of poles of S-plane it can be noted

that there is a exponentially increasing term and exponentially
decreasing term. But, since magnitudes of locations where
θs ( s ) k/I poles located, are equal these exponential terms are cancelled
= 2 0ut. So it’s difficult to comment on the stability of the system
θm( s ) S + k / I using those poles. Therefore we have to consider the S=0 pole
which governs the impulse response of the system. This can be
seen by deriving the impulse response of the system.
Transfer function of feedback torque can be derived by
multiplying 3 and 4 Using partial fraction for (3)
H(S)=kI2-1/2kIS2-1/(2k/I)32S+2k/I+1/(2k/I)32S-2k/I
τ ' (S ) kI 2 S 2
= 2
θm( s ) S + k / I Then taking inverse Laplace transform function gives the
impulse response of the system.
ht=L-1[Hs]
By virtually implementing these functions the similar operation
as the mechanical system can be obtained when the spring is ht=-kI212kIt-1/(2k/I)32e-2k/It++1/(2k/I)32e2k/It
not present in the system. The next step of modeling system is
to simulate the system in a computer environment with the help Since the exponential terms are cancelled out. -12I3t term
of the derived mathematical model. There, it can be observed dominates the impulse response of the system. It can be noted
the behavior of the slave according to the various force applied that is a constantly increasing term(not considering the
to the master and forces exerted on master by the slave. magnitude). Practically this scenario can be noted. Due to
impulse torque slave tend to move at a constant angular speed.
Frequency Domain Analysis That angular speed depend only on the inertia of the system.

When the input is a torque When the input is angle

Using (5) θs( s ) k/I

= 2
θm( s ) S + k / I
θ 2( S ) K / I12
= 2 2
τ ( S ) S {S − (2 K / I1 )} There are poles at and
S = ± (K / I )
There are poles at S = 0 and

S = ± ( 2 K / I1 ) Graphically interpertting this on S- plane

This can be graphically interpreted on a S plane as follows jω

S =− K/I S =+ K/I
σ
This also gives rise a impulse response of with two
exponential terms, decreasing one and an increasing one.But
there is not a pole at S = 0 as the previous case .Therefore the
exponential terms cancelled out and nothing left that is h. This
scenario can be cleary seen by examining the impulse response.

Using partial fraction for (3)

H(S)=KI1/(-2k/I)S+k/I)+ 1/(2k/I)S-k/I)

Then taking the inverse laplace of (4)

ht=KI2K/Iek/I) t- e-k/I) t
since exponential terms are cancelled out the 3 becomses
ht=0

That means there is no movement of the slave in response to a

impulse angle change of the slave.In practice this scenario can
be experienced when impulse movement is applied to the
master. Impulse input of angle can be considered as moving the
rod to a some angular position and then moving it back to the
original position in very little time period.
Applying =I for slave
τ ..
θ

(1)
(2)
..
Applying =I for slave I 2 θ 2 = k (θ 1− θ 2)
τ ..
θ Taking the laplacian of (1) and (2) we get,

P-T=IθM
..
I 1[ S 2θ 1( s ) − Sθ 1(0) − θ 1(0)] + kθ 1( S ) − kθ 2( S ) = τ ( S )
(3)
(4)
..
I 2[ S 2θ 2( s) − Sθ 2(0) − θ 2(0)] − kθ 1( S ) + kθ 2( S ) = 0

Since a system which is initially at rest,

.. ..
θ 1(0),θ 2(0) = 0
Since two rods are equal.

I1 = I 2
By (3) and (4) transfer function of the master’s force to slave’s
position can be derived,
 (5) There are two well known bilateral control architectures used
at present, there are position-position architecture and
τ ( S ) I1 S − 2 KI1 S
2 4 2
position-force architecture.
=
θ 2( S ) K
Position to Position Architecture [3]
And also the transfer function for the reaction force exerted by
the slave on the master is given by In this architecture both robots are programmed to track each
other’s position. The slave imitates the movements of the
(6) master and also slave transfer the forces acting on it back to
the master. If master want to feel force acting on slave’s side
τ ' (S ) that force should be large enough to move the slave
= I 1S 2 actuators of the slave. If it doesn’t there is no method to
θ 2( S )
master to sense the forces acting on the slave. So this doesn’t
agree with the concept of force feedback all the time.
By virtually implementing these functions the similar operation
as the mechanical system can be obtained when the spring is Position-Force Architecture [3]
not present in the system. The next step of modeling system is
to simulate the system in a computer environment with the help In this architecture master can sense all forces acting on the
of the derived mathematical model. There, it can be observed
the behavior of the slave according to the various force applied slave whether they move the slave’s actuators or not. But this
to the master and forces exerted on master by the slave. architecture is less stable than the previous one.
IV. CONCLUSION The system modeled above is a system with a position-
position architecture because a torque should move the
When implementing the above system physically using the system if the force should be sensed.
virtual model, we can’t expect the exactly the same behavior
as the mechanical system. This happens due to delays of REFERENCES
actuators and sensors and also the data processing and
communication delays. At slower speeds nearly similar [5] Cristian Secchi,Stefano Stramigioli,Cesare Fantuzzi,“Control of
behavior to the mechanical system could be observed. But at Interactive Robotic Interfaces”,Springer,2007
higher speeds, faster data acquisition and processing methods [6] Horan, Ben, Creighton, Douglas, Nahavandi, Saeid and Jamshidi, Mo
actuators and sensors with low response time should be used 2007, Bilateral haptic teleoperation of an articulated track mobile robot,
in IEEE International Conference on System of Systems Engineering,
in order to obtain that behavior. Communication delay is the 2007. SoSE '07., IEEE Xplore, Piscataway, N.J, pp. 1-8.
one of the significant issue in tele-operated bilateral control [7] Bruno Siciliano, Oussama Khatib ,” Handbook of Robotics”,
systems. When the master and slave are located far away from Springer,2008
each other delays can arise when transmitting data through [8] William R. Feme11,“Remote Manipulation with Transmission Delay,”
communication medium. As an example when the tele- National Aeronautics and Space Administration.Washington, D. C.,1965
operation uses the a packet switching network such as internet [9] Thorsten A. Kern,”Engineering Haptic Devices”,Springer,2009
as the communication medium there can be considerable
amount of delay when transferring data from master to slave
and slave to master. When this delay can’t be neglected the
stability of the system can’t be guaranteed. [4]
σω