You are on page 1of 14

Journal of Intelligent and Robotic Systems 32: 23–36, 2001.

23
© 2001 Kluwer Academic Publishers. Printed in the Netherlands.

3-d.o.f. Wire Driven Planar Haptic Interface

PAOLO GALLINA, GIULIO ROSATI and ALDO ROSSI


Department of Innovation in Mechanics and Management, University of Padova, via Venezia 1,
35131 Padova, Italy

(Received: 12 May 2000; in final form: 8 November 2000)


Abstract. A 4-wire driven 3-d.o.f. planar haptic device, called the Feriba-3, is described. The par-
ticular geometric configuration of the end-effector ensures a closed form kinematic pose solution
and good manipulability. Moreover, the structural arrangement adopted makes the Feriba-3 a well-
performing haptic device, whose major features are low inertia, low friction, and full dexterity in a
large workspace. The manipulability analysis has been performed by introducing a complete set of
manipulability indices.

Key words: haptic interfaces, manipulability, polytopes, wire driven systems.

1. Introduction
Haptic displays are emerging as effective interaction aids for providing the sense
of force or moment feedback to the human operator. Many devices have been built
and commercialized with different features. In most cases they are too heavy and
not manipulable enough to allow effective usage. Several desktop haptic interfaces
have been developed both for entertainment and for CAD programs. Most of them
[11] consist of haptic joysticks whose use is limited to video games applications
because of low accuracy, reduced workspace and a limited number of d.o.f. In
other cases they are planar mechanisms with complicate structures [4]. More so-
phisticated 3-D spatial desktop haptic interfaces have been developed for CAD
applications; a recent example can be found in [6]. The cost of such haptic devices
is high because of the sophisticated mechanical structure. For this reason, planar
haptic interfaces are attracting the interest of many researchers, but few examples
exist at the moment [3, 4]. The work presented here aims to overcome some prob-
lems arisen in these planar devices, such as high inertia and reduced workspace.
One way to obtain better results could be to employ a wire-direct-driven parallel
system. This is not a new concept in haptic interfaces, as some wire-based hap-
tic displays have been proposed in the literature [1, 7–9]. They all consist of a
series of wires and of a low-mass end-effector to which the human hand/finger-
tips are linked. The wires are direct-driven and their tensions provide the virtual
 E-mail: gallina@dimeg.unipd.it.
24 PAOLO GALINA ET AL.

force-feedback to the end-effector. The advantages of such a configuration are


low-inertia, low-cost, and high safety.
In this paper, we describe an electromechanical device capable of producing
a planar force and moment feedback. An attempt to produce a planar moment
feedback by direct-driven wires was made by Williams [12]. He remarked the fact
that the motors were too large and easily overpowered the human hand. Moreover,
the particular configuration of the end-effector decreased the manipulability index
[10]. The innovative design, we introduced, together with an appropriate choice of
the hardware components, led to better results in terms of manipulability, giving the
operator a fine sensitivity in a quite wide range of force (0 ÷ 5 N) and momentum
(0 ÷ 250 Nmm). The achievement of these results in a large workspace (about
300 × 300 mm2 ) makes the Feriba-3 suitable for most haptic applications and
particularly for reproducing real and virtual environments.

2. System Description
The Feriba-3 (see Figure 1) is a 4-wire planar mechanism with a circular end-
effector (named a spool). Each wire is fixed to the lateral side of the spool and
can wind around the spool. The other end of the wire is wound around a pulley,
that is directly keyed on a motor shaft. The motors, together with the encoders that
provide the angular position of the shafts, are mounted on a steel base in relation
to the vertices of a square, of 450 mm side. The spool and pulleys winding radius
is 30 mm.
Both the motors (Escap 24V DC micromotors) and the encoders (Lika 5000-
step encoders) are low rolling friction devices; furthermore, the pulley radius plays
a role in reducing the resultant effect of the shaft friction on the spool. In order

Figure 1. Feriba-3 overall view.


3-D.O.F. WIRE DRIVEN PLANAR HAPTIC INTERFACE 25

(a) (b)
Figure 2. Spool side view (a) and bottom view (b).

to minimize the spool friction, a sheet of glass is fixed on the base and a flux of
compressed air is blown towards the glass through a series of little holes drilled in
the lower side of the spool. This way, a thin layer of air is always kept between
the spool and the sliding surface. Moreover, the handle is connected with the spool
by means of a cardan joint, so as to avoid transmission of dumping moments (see
Figure 2).
The control system is based on a VME architecture and includes a PPC6750
motherboard and some I/O modules (for motor driving, encoder acquiring and
general purpose I/O). Real-time programming ensures full control of resources
at a frequency of 5 kHz. Each control cycle comprises pulley angular position
acquisition, forward kinematic pose solution, feedback force calculation, wire ten-
sion computation, and motor driving voltage updating. The feedback general force
vector F = {Fx , Fy , Mz } is determined as a function of the spool position x =
{x, y, θ} in order to give the feeling of a virtual environment to the operator.
The wire tension computation aims at calculating the set of wire tensions f =
{f1 , f2 , f3 , f4 } that yields a resultant general force vector F to the spool; since the
system is underconstrained, we chose to stretch each wire up to a tension greater
or equal to a minimum positive value fmin , so that there will always be a wire
stretched exactly to fmin . The existence and uniqueness of such a set of tensions
are demonstrated in [5]. Finally, we calculate the torque vector τ = {τ1 , τ2 , τ3 , τ4 }
and hence the proper driving voltage for each motor.
Notice that our haptic interface is sensorless. We adopted this philosophy for
two reasons: first of all, force sensors make the system more complicated to control;
moreover, placing force sensors in the wires causes the wires to be heavier and the
workspace to be smaller. The last consideration comes up by the fact that the force
sensors could not wind around the pulleys. The sensorless strategy has turned out
26 PAOLO GALINA ET AL.

to be a good choice since we reduced all the possible friction forces by flowing
compressed air under the end-effector.
The control system is monitored via Ethernet by a PC and could be easily
interfaced with a slave robot control unit. At the moment, the Feriba-3 has been
tested in a virtual environment consisting in an elastic wall. Instability has been
reached with a virtual wall stiffness of 30000 Nm.

3. Kinematic Analysis
The kinematic analysis of the Feriba-3 is fully developed in [5]. In this paper, we
do not report all the equations but only the reverse kinematic pose solution.
The reverse kinematic pose solution is obtained in closed form. By simple
geometrical considerations (see Figure 3a) we obtain

(x − xi )2 + (y − yi )2 + (−1)i θR
ni = n0 , (1)
2π R
where P (x, y) is the spool center, θ is the counter-clockwise spool rotation,
Pi (xi , yi ) is the ith pulley center, R is the spool and pulleys radius, ni is the
angular position (measured in encoder steps) of the ith pulley, and n0 = 20000
is the number of steps per round; ni is positive when the ith wire is being unrolled,
and is null when both θ = 0 and P ≡ Pi . Note that the simplicity of Equation (1)
results from having chosen the same radius for the pulleys and the spool.
The forward kinematic pose solution is obtained in closed form as well. Re-
ferring to Figure 3a, let us consider the two ellipses whose focuses are the points
Pi−1 , Pi and Pi+1 . The first ellipse represents the trajectory that the spool center
would follow if the (i − 1)th and the ith pulleys were locked, while the second
ellipse represents the trajectory that the spool center would follow if the ith and
the (i + 1)th pulleys were locked. The ellipse parameters are easily obtained from

(a) (b)
Figure 3. Kinematic analysis (a) and manipulability analysis (b) schemes.
3-D.O.F. WIRE DRIVEN PLANAR HAPTIC INTERFACE 27

the sum of ni−1 and ni , and the sum of ni and ni+1 , respectively. Clearly, the spool
center P coincides with the only one intersection point of the two ellipses that be-
longs to the square whose vertices are the pulley centers. By writing the equations
of the ellipses in polar coordinates with the origin in Pi , after trivial manipulations
we get an explicit expression of the spool center coordinates x and y as a rational
function of ni−1 , ni and ni+1 ; finally, the value of spool rotation θ is obtained by
inverting Equation (1).
In general, two ellipses have four intersection points whose coordinates can
be found by solving a fourth-degree polynomial equation; the polar-coordinates
method we proposed is clearly much simpler, as it yields an explicit expression
for the solution we are interested in. Note that the ellipses parameters are simple
functions of wire lengths because we chose the same radius for the pulleys and the
spool; again, this choice has proved to be convenient.
The forward kinematic pose solution method presented in [5] can be performed
in four different ways by neglecting a different pulley each time. In this way, we
obtain four solutions that should be equal in value. However, in the real system we
will get four different values because of dimensional and measurement errors. As
NC machining has been performed with an overall accuracy of less than 0.01 mm,
dimensional errors can be neglected in comparison with measurement errors. The
latter can be interpreted as the difference between the acquired value ni and the
value of ni calculated by Equation (1):

(x − xi )2 + (y − yi )2 + (−1)i θR
ni = ni − n0 , (2)
2π R
where x, y, and θ are supposed to be known. This difference is the result of the
encoders finite resolution, calibration errors, and wire stretching:

ni = nres + {ncal }i + {nstr }i , (3)

where ncal , nstr ∈ R4 are the calibration and stretching error vectors.
The first term on the right-hand side of Equation (3) can be set to the following
value

nres = ±0.5. (4)

With respect to calibration, let xcal = {xcal , ycal , θcal } be the accuracy of
spool positioning during the calibration. Let J(x, y, θ) be the Jacobian matrix of
system (1); we have
 ∂n1 ∂n1 ∂n1 
∂x ∂y ∂θ 
 ∂n2 ∂n2 ∂n2  xcal
 ∂x ∂θ 
ncal = Jxcal =  ∂n3 ∂n3 ∂n3  ycal .
∂y
 ∂x 
∂y ∂θ θcal
∂n4 ∂n4 ∂n4
∂x ∂y ∂θ
28 PAOLO GALINA ET AL.

We can set xcal = ycal = ±s and θcal = 0; the last condition makes the
spool rotation θ relative to the initial spool position. It follows that


∂ni ∂ni
{ncal }i = ± + s = ± n0  |x − xi | + |y − yi | s
∂x ∂y 2π R (x − xi )2 + (y − yi )2
n0
= ± (| cos ϕi | + | sin ϕi |)s,
2π R
where ϕi is the angle between the x axis and vector P − Pi ; in the worst case, we
will get
n0 √
{ncal }i = ± 2s  150 × s, (5)
2π R
where s is measured in millimeters and {ncal }i is measured in encoder steps.
The last term on the right side of Equation (3) is given by the following
n0 n 0 fi
{nstr }i = li = , (6)
2π R 2π R ki
where li , fi , and ki are the ith wire stretching, tension, and stiffness, respectively.
This term is always positive, and becomes relevant as fi becomes greater. We can
estimate the value of {nstr}i by substituting fi with τi /R in Equation (6) (the
lower the system inertia, the more this approximation works), and then subtracting
the result from the acquired value ni before performing the forward kinematic pose
solution, so as to compensate for the wire stretching error. Thus, we can neglect it
in the expression of the measurement error.
Hence, substituting Equations (4) and (5) into Equation (3), with a calibration
accuracy of s = 0.05 mm, we get the following value for measurement accuracy
n  ±(0.5 + 150 × s) = ±8. (7)
Let us now evaluate the effects of the measurement errors on the accuracy of the
forward kinematic pose solution. Let xi be the solution calculated by neglecting
the ith pulley, Ji ∈ M3×3 be the square matrix obtained from the Jacobian matrix
by neglecting the ith row, and ni be the vector obtained from the measurement
error vector n = {n1 , n2 , n3 , n4 } by neglecting the ith component. The
difference between the vector xi and the actual spool position vector xact is given
by
xi = xi − xact = J−1
i ni . (8)
A suitable value for the calculated position vector x is given by the arithmetic mean
of vectors xi ; hence, the overall error in the spool position computation will be
1 1
4 4
x = x − xact = xi − xact = (xi − xact )
4 i=1 4 i=1

1 1 −1
4 4
= xi = J ni = Hn, (9)
4 i=1 4 i=1 i
3-D.O.F. WIRE DRIVEN PLANAR HAPTIC INTERFACE 29

Figure 4. Overall x and y accuracy contour plot; contour lines are equally spaced starting
from the inner line (0.15 mm) with a step of +0.05 mm.

where the matrix H ∈ M3×4 (R) is obtained by assembling matrices J−1 i and
dividing the result by 4. The accuracy on x, y, and θ is then given by


4
x = ±n |h1j |,
j =1


4
y = ±n |h2j |,
j =1


4
θ = ±n |h3j |, (10)
j =1

where the value of n is yielded by Equation (7).


Figure 4 represents the contour plot of max{x, y} as a function of the spool
center coordinates x and y (the spool rotation θ is kept null). Contour lines are
equally spaced starting from the inner line (0.15 mm) with a step of 0.05 mm.
As shown, the accuracy is symmetrical with respect to the diagonals and to the
axes of the reference frame. Furthermore, the accuracy is satisfactory (lower than
0.15 mm) for most of the Feriba-3 workspace.
Figure 5 represents the contour plot of θ as a function of the spool center
coordinates x and y (the spool rotation θ is kept null). Contour lines are equally
spaced starting from the inner line (0.15◦ ) with a step of 0.01◦ . As shown, the ac-
curacy is symmetrical with respect to the diagonals and to the axes of the reference
frame. Furthermore, the accuracy is satisfactory (lower than 0.15◦ ) for most of the
Feriba-3 workspace.
30 PAOLO GALINA ET AL.

Figure 5. ϑ accuracy contour plot; contour lines are equally spaced starting from the inner
line (0.15◦ ) with a step of +0.01◦ .

4. Manipulability
Given an n d.o.f. object driven by n + m wires (with m  1), let f ∈ Rn+m be the
wire tension vector and F ∈ Rn be the feedback general force vector acting on the
object; the relationship between the vectors f and F is yielded by

Af = F, (11)

where the matrix A ∈ Mn×(n+m) (R) is a function of the object coordinates. A wire
driven system is manipulable if any feedback vector F can be generated by a proper
set of positive or null wire tensions f.
The explicit expression of system (11) for the Feriba-3 is given by
 
  f1  
v1x v2x v3x v4x    Fx
f2
v1y v2y v3y v4y = Fy , (12)

 f3 

R −R R −R Mz
f4
where vix and viy are the components of the ith wire unit norm vector vi (see
Figure 3b). It is demonstrated in [5] that the Feriba-3 is manipulable if the spool
center belongs to the square whose vertices are the pulley centers. Moreover, we
proved that, if an n d.o.f. system driven by n + 1 wires is manipulable, it can
generate each F with all wire tensions greater or equal to a minimum positive value
fmin , the minimum of fi being equal exactly to fmin . This fact is important for a
wire driven system since a minimum tension is needed in order to keep the wire
stretched [1].
The above definition of manipulability is based on geometrical considerations
only, and on the hypothesis that the value of fi is unlimited. However, in a real
system we must take into account the fact that the upper limit fmax is set to the wire
3-D.O.F. WIRE DRIVEN PLANAR HAPTIC INTERFACE 31

Figure 6. Manipulability analysis of a planar 2-d.o.f. 3-wire driven system.

tension due to motor maximum torque. This means that a physical system, even
though manipulable, will not be able to produce all feedback vectors F. Thus, in
order to evaluate the degree of manipulability of a wire-driven system, we introduce
some manipulability indices.
Let us consider a simple case of a wire-driven system generating forces but
not moments; in this case, a suitable manipulability index is the magnitude of the
maximum force F obtainable in every direction of the workspace. For example,
in Figure 6, a planar 2-d.o.f. 3-wire driven system is shown; by drawing the lines
parallel to the wires at a distance fmax from point P (measured along the directions
of the wires) we get the hexagon of all the forces the system can generate. Thus, the
manipulability index is given by the radius F of the circle of center P and tangent
to the polygon. Clearly, the value of F is a function of the coordinates of point P .
The situation is more complex in the case of the Feriba-3, since the system
produces two forces and one moment, and these quantities are heterogeneous.
Since the forces and the moment are coupled with respect to the wire tensions,
the system will not be able to produce high forces and a high moment at the same
time. So wecan set a value M of the modulus of Mz and evaluate the maximum
force F = Fx2 + Fy2 obtainable in every direction of the workspace. Thus, we
can draw the plot of F = f (M) as a function of M and define the following
manipulability indices:

iF = max{F },
iM = max{M}, (13)
 iM
1
iI = f (M) dM.
iF iM 0
As we will show further on (see Equations (14) and (15)), the maximum of F is
obtained as the moment M is kept null, and vice-versa. Moreover, the integral of
f (M) cannot exceed the value iF iM and the more iI is near to unity, the more the
32 PAOLO GALINA ET AL.

forces and the moment are uncoupled. Notice that the three indices are a function
of the position of the spool.
Let us now evaluate the manipulability indices (13) in the case of the Feriba-3.
First of all, let us define the linear transformation g associated with the matrix A:

g: R4 → R3 , g(f) = Af,

where A ∈ M3×4 (R) is the matrix of system (12). Let Hfmax be the set of all possible
wire tension vectors the motors can afford:

Hfmax = {f ∈ R4 : 0  fi  fmax ∀i}.

If we calculate the image of Hfmax through transformation g, we get the set Pg,fmax
of all the feedback vectors F that the system can generate:

Pg,fmax = {F ∈ R3 : F = g(f) ∃f ∈ Hfmax }.

It is well known that Pg,fmax is a convex polytope [2], as we obtained it by linear


transformation of a hypercube. Given a couple of values M and F , let CF,M be a
cylinder defined as follows
 
F 2 + F2  F

CF,M = {Fx , Fy , Mz } ∈ R : 3 x y .
|Mz |  M
Clearly, given a value M, Cf (M),M is the largest cylinder belonging to Pg,fmax whose
axis coincides with the z axis, whose generatrix is 2M long and whose center co-
incides with the origin of R3 . Then, in order to evaluate the manipulability indices
we introduced in (13), we must determine the base radius of the largest cylinder
CF,M contained in Pg,fmax , with M ranging from zero to the maximum obtainable
moment iM .
Let vj · F = 1 be the equation of the plane πj containing the j th side of the
polytope (vector vj = {aj , bj , cj } is orthogonal to πj and points the plane). The
locus of the contact points between πj and the largest cylinders CF,M belonging to
the semi-space vj · F  1 is a line segment (see Figure 7). The equation of this line
segment in the MF plane is given by

1 − |cj |M
F = fj (M) =  , (14)
aj2 + bj2


where M = |Mz | ∈ [0, 1/|cj |], and F = Fx2 + Fy2 varies between fj (0) =

1/ aj2 + bj2 and fj (1/|cj |) = 0. It is clear that the lower M, is the higher is F ;
therefore the maximum value of F is obtained when M is null, and vice versa. By
drawing the line segments relative to all the sides of the polytope and choosing
3-D.O.F. WIRE DRIVEN PLANAR HAPTIC INTERFACE 33

Figure 7. Contact points between the j th side of the polytope and a cylinder Cfj (M),M .

Figure 8. F = f (M) curves calculated with the spool center positioned along the x axis,
starting from the origin (curve 0) and proceeding with a step of 22.5 mm (curves 1 ÷ 9).

for every M the lowest value of fj (M), we get the function f (M) and hence the
values of iF , iM , and iI (see Figure 8); iF and iM are given by

 
1
iF = min  = f (0),
j
aj2 + bj2
  (15)
1
iM = min = f −1 (0).
j |cj |
34 PAOLO GALINA ET AL.

Figure 9. Contour plot of iF ; contour lines are equally spaced, starting from the inner line
(9 N) with a step of −1 N.

In conclusion, the evaluation of manipulability indices reduces to writing the equa-


tions of the planes that contain the sides of the polytope; such equations are easily
obtainable by transforming via g the vertices of the hypercube Hfmax .

Remark. Notice that the hexagon and the circle of Figure 6 are a two-dimension-
al polytope and a two-dimensional cylinder, respectively.

Since in the case of the Feriba-3 the matrix A is a function of x and y only,
and the system is symmetrical with respect to the axes of the reference frame, we
present the contour plots of iF , iM , and iI in the first quadrant of the xy plane. The
value of fmax is set to 10 N.
Figure 9 represents the contour plot of iF ; contour lines are equally spaced
starting from the inner line (9 N) with a step of 1 N. As shown, iF is greater or equal
to 5 N in a square whose side is 300 mm long. Figure 10 represents the contour plot
of iM ; contour lines are equally spaced starting from the inner line (550 Nmm) with
a step of 50 Nmm. As it is shown, iM is greater or equal to 250 Nmm in a square
whose side is 300 mm long. Both the contour lines of iF and the contour lines of
iM are approximately circular near the origin and square near the boundaries of the
workspace.
Figure 11 represents the contour plot of iI ; contour lines are equally spaced,
starting from the inner line (0.64) with a step of 0.01. As shown, the contour lines
of iI are different in shape from those of iF and iM . The diagonals of the xy plane
3-D.O.F. WIRE DRIVEN PLANAR HAPTIC INTERFACE 35

Figure 10. Contour plot of iM ; contour lines are equally spaced, starting from the inner line
(550 Nmm) with a step of −50 Nmm.

Figure 11. Contour plot of iI ; contour lines are equally spaced, starting from the inner line
(0.64) with a step of −0.01.
36 PAOLO GALINA ET AL.

become critical regions near the vertices of the workspace. However, the value of iI
ranges from 0.5 to 0.65 in the most part of the workspace; these values correspond
to nearly linear F = f (M) curves.

5. Conclusions
In this paper, we presented a 4-wire driven 3-d.o.f. planar haptic device. The kine-
matic pose solution was obtained in closed form, due to the particular geometrical
configuration of the end-effector. Moreover, a close examination of the forward
kinematic pose solution accuracy yielded good results. The manipulability was
investigated using polytopes, and a complete set of manipulability indices has been
introduced. As a result, our device demonstrated to be a well-performing haptic
device, ensuring a good manipulability in a large workspace.

References
1. Bonivento, C., Eusebi, A., Melchiorri, C., Montanari, M., and Vassura, G.: WireMan: A
portable wire manipulator for touch-rendering of bas-relief virtual surfaces, in: 8th Inter-
nat. Conf. on Advanced Robotics, Proceedings, ICAR’97 IEEE, New York, NY, USA, 1997,
pp. 13–18.
2. Chiacchio, P., Bouffard-Vercelli, Y., and Pierrot, F.: Force polytope and force ellipsoid for
redundant manipulators, J. Robotic Systems 14(8) (1997), 613–620.
3. Ellis, R. E., Ismaeil, O. M., and Lipsett, M. G.: Design and evaluation of a high-performance
haptic interface, Robotica 14 (1996), 321–327.
4. Frisoli, A., Prisco, G. M., Salsedo, F., and Bergamasco, M.: H I 2 : A two-degrees-of-freedom
planar, highly isotropic haptic interface for the desktop, in: Proc. of SPIE, Vol. 3840, Boston,
MA, USA, 1999, pp. 65–75.
5. Gallina, P. and Rosati, G.: Manipulability of a planar wire driven haptic device, under review,
Mechanism and Machine Theory, submitted in Jan. 2000.
6. Gutierrez, T. et al.: Virtual assembly and disassembly simulation, in: Intelligent Assembly and
Disassembly (IAD’98), Proceedings of the 1st IFAC Intelligent Assembly and Disassembly
Workshop, 1998, pp. 35–40.
7. Ishii, M. and Sato, M.: A 3D spatial interface device using tensed strings, Presence-
Teleoperators and Virtual Environments 4(4) (1994), 81–86.
8. Ishii, M. and Sato, M.: Six degree-of-freedom master using eight tensed strings, in: ISMCR ’98,
Proc. of the Eighth Internat. Symp. on Measurement and Control in Robotics, Vol. 455, Czech
Tech. Univ. Prague, Prague, Czech Republic, 1998, pp. 251–255.
9. Liendmann, R. and Tesarhave, D.: Construction and demonstration of a 9-string 6-d.o.f. force
reflecting joystick for telerobotics, in: NASA International Conference on Space Telerobotics,
Vol. 4, 1989, pp. 55–63.
10. Shen, Y., Osumi, H., and Arai, T.: Manipulability indices in multi-wire driven mechanisms,
Distributed Autonomous Robotic Systems 394 (1994), 359–370.
11. Weisenberger, J. M., Krier, M. J., Rinker, M. A., and Furness, R. J.: Resolution of virtual grating
orientation with 2-DOF and 3-DOF force feedback systems, in: Proc. of the 1998 ASME In-
ternat. Mechanical Engineering Congress and Exposition, Vol. 64, Anaheim, CA, USA, 1998,
pp. 295–301.
12. Williams, R. L. II: Cable-suspended haptic interface, Internat. J. Virtual Reality 3(3) (1998),
13–21.

You might also like