Sensors and Actuators A 90 (2001) 31±37

Closed-form force sensing of a 6-axis force transducer

based on the Stewart platform
Chul-Goo Kang*
Department of Mechanical Engineering, Konkuk University, Kwangjin-gu, Seoul 143-701, South Korea
Accepted 11 December 2000

Abstract

The 6-axis force transducers based on the Stewart platform require the solution of forward kinematics of the Stewart platform in order to
obtain force and torque acting on the upper plate from given six leg forces. However, the solution of forward kinematics is not known in a
closed form. In this paper, a closed-form solution of the forward kinematics is derived by means of linearization of the inverse kinematics
equations, and the derived solution is applied to the calculation of force and torque acting on the transducer. In order to investigate the
validity of the proposed method, a prototype of the transducer based on the Stewart platform is developed and force analysis is conducted.
Moreover, the gravity compensation algorithm is proposed to reduce the weight effects of the transducer. # 2001 Elsevier Science B.V. All
rights reserved.

Keywords: Force transducer; Stewart platform; Closed-form solution; Gravity compensation

1. Introduction the transducer provides passive compliance often needed

during a robotic assembly task.
Multi-axis force transducers have been initially developed Stewart platform was originally proposed for the ¯ight
for the applications such as wind-tunnel testing and thrust simulator [7]. Gaillet and Reboulet [8] showed that the
stand testing of rocket engines. It has been recognized that Stewart platform mechanism with articulated rods, each
these transducers can be applied to robotic applications since carrying a measurement device can measure force and
the mid seventies. In order to automate dexterous and torque acting on the mechanism. Kerr [6] analyzed the
accurate tasks using robotic manipulators, we need to feed- behavior of a force transducer that has the form of a Stewart
back not only position but also force information. Thus, the platform with instrumented elastic legs. Nguyen et al. [9]
importance of sensing the force and torque with arbitrary conducted experimental studies on a Stewart platform-based
direction and magnitude is becoming crucial for the auto- force transducer, but they solved the forward kinematics
mation of these tasks, and these transducers have received problem numerically, which is required for the calculation of
considerable attention in the literature [1±4]. external force and torque from the given leg forces. Using
The 6-axis force transducers can measure three force- numerical solutions to real-time application is not recom-
components and three moment-components acting on the mended because convergence time for each iteration may be
transducers. There are some commercialized 6-axis force different, and in some instances, the convergence time may
transducers by Lord Corp., JR3 Inc., ATI, etc. and these exceed the time limit.
transducers are equipped with strain gauges stuck on a more In this paper a closed-form solution of forward kinematics
or less complex mechanical structure [5]. of the Stewart platform is derived by means of the linear-
As a different kind of 6-axis force transducer, a Stewart ization of the inverse kinematics solution, and the derived
platform-based force transducer with elastic links is pro- solution is applied to the calculation of force/torque acting
posed by Kerr [6]. The advantages of the Stewart platform- on the transducer. In order to investigate the validity of the
based force transducer are that the internal stresses of the proposed method, a prototype of the transducer based on the
transducer material do not appear in the measurement, and Stewart platform is designed, and experimental force ana-
lysis for this transducer is conducted. Gravity compensation
algorithm is included to reduce the gravity effects by the
*
Tel.: ‡82-2-447-2142; fax: ‡82-2-447-5886. weight of the upper plate, joints and other parts of the
E-mail address: [email protected] (C.-G. Kang). transducer.

0924-4247/01/$ ± see front matter # 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 4 - 4 2 4 7 ( 0 0 ) 0 0 5 6 4 - 1
32 C.-G. Kang / Sensors and Actuators A 90 (2001) 31±37

The paper is organized as follows. Section 2 presents the sents a rotation of a about xA-axis followed by a rotation b
inverse and forward kinematics problems, Jacobian, and about yA-axis followed by a rotation g about zA-axis. The
force relationship of the Stewart platform, and then derives equation obtaining l from the given x can be expressed as
a closed-form solution of 6-axis force transducer that solves
l ˆ g…x† (1)
the forward kinematics of the Stewart platform by means of
a linearization technique. In Section 3, the kinematic struc- and this equation is known as `inverse kinematics' of the
ture of the Stewart platform being used as a force transducer Stewart platform. Note that the bold face in this paper
is designed in terms of radii of the upper and lower plates, represents vector quantity. To express the inverse kinematics
length of six legs, and distance of joints. Section 4 presents (Eq. (1)) more specifically, we define A ai as the position
numerical error evaluation by the proposed linearization vector of Ai with respect to frame {A}, B bi as the position
technique. Section 5 shows the results of experimental vector of Bi with respect to frame {B}, A d ˆ …xc ; yc ; zc †T as
works, and Section 6 concludes. the position vector of the origin of the frame {B} with
respect to frame {A}, and AB R as the rotation matrix trans-
forming the coordinates with respect to {B} to the coordi-
2. Closed-form solution of forward kinematics of the nates with respect to {A}. Then the inverse kinematics
Stewart platform via linearization solution of the Stewart platform can be written as
q
The Stewart platform being used as the 6-axis force li ˆ lTi li ; li ˆ AB RB bi ‡ A d ÿ A ai i ˆ 1; . . .; 6
transducer is composed of an upper plate, a lower plate,
and six instrumented elastic legs connecting the two plates where li is the vector representing the length and direction of
with appropriate joints. The kinematic structure of the leg i with respect to frame {A}. In the above equation, A ai
Stewart platform is shown in Fig. 1. The coordinate frame and B bi are constant vectors, and AB R is varying with rotation
{A} and {B} in Fig. 1 are attached to the centers of the lower angles a, b, g.
and the upper plates. The `forward kinematics' of the Stewart platform is the
In order for the upper plate to have a six controllable problem obtaining the position and orientation x of the upper
degrees of freedom motion, one joint at the end of the elastic plate from the given leg length l. The closed-form solution of
leg must be a spherical joint (e.g. ball and socket joint) and the forward kinematics for the general Stewart platform is
the other joint of the leg be a universal or spherical joint [10]. not known although many researchers have tried to solve it
The Stewart platform may have some singularities related to [12,13]. Therefore, a numerical solution (e.g. Newton±
the architecture of it. To prevent these architecture singula- Raphson method) used to be applied to the place where
rities, we should design the kinematic parameters of the the solution of forward kinematics is needed [9]. In general,
Stewart platform so that the upper and the lower plates may however, numerical solution is not recommended in real-
not be regular hexagons, and the longer side of one hexagon time application because of convergence speed and timing
may be connected to the shorter side of the other hexagon limitation.
with legs if the two plates are similar [11]. In this paper, a closed-form solution for forward kine-
Let l ˆ …l1 ; . . . ; l6 †T be a vector representing lengths of matics is obtained approximately and applied to the 6-axis
six legs, and x ˆ …x; yc ; zc ; a; b; g†T be a vector representing force transducer. For this purpose, differentiating Eq. (1)
the position and the orientation of the upper plate. The with respect to time results in relationship between leg
…xc ; yc ; zc †T in x represents the coordinates of the origin of velocity and velocity of the upper plate such as
frame {B} with respect to frame {A}, and …a; b; g†T repre- l_ ˆ J…x†x_ (2)
where J…x† ˆ @g=@x, and J(x) is called the `Jacobian' of the
Stewart platform.
Note that static relationship between force/torque acting
on the upper plate and leg forces is obtained from the
principle of virtual work as follows.
F ˆ J T …x†f (3)
T
where f ˆ …f1 ; . . .; f6 † is the vector representing magnitudes
of six leg forces, and F ˆ …Fx ; Fy ; Fz ; Mx ; My ; Mz †T is the
vector representing force and the moment (torque) acting on
the center of the upper plate.
The static forces in the transducer are analyzed on the
assumption that leg de¯ections are small (this is true for
Fig. 1. Kinematic structure of the Stewart platform and coordinate frame transducers) and the effect of dynamic forces are negligible.
assignments. In Eq. (3), the Jacobian J(x) is a function of the position and
 C.-G. Kang / Sensors and Actuators A 90 (2001) 31±37 33

orientation of the upper plate. Therefore, we have to know To investigate the isotropy of the forces F acting on the
the position and orientation of the upper plate to obtain the upper plate, let us consider the case that the upper plate can
force/torque F acting on the upper plate from the given leg exert forces F without generating any rotational motion from
forces f. In other words, we have to solve the forward the leg forces satisfying
kinematics problem.
jf j  1 (5)
In this section, the solution of forward kinematics of the
Stewart platform is obtained from the solution of the inverse The evaluation of the forces F for this case (i.e. the
kinematics using ®rst-order approximation (i.e. lineariza- mapping of Eq. (5) with F ˆ JfT f satisfying JmT f ˆ 0) results
tion), and is used for the 6-axis force transducer on the in the following inequality.
assumption that there are no singularities within the work-
space of the transducer. For the linearization of the inverse F T QF  1 (6)
kinematics solution (1), let a normal operating condition x0, 33
where Q…2 R † is a symmetric matrix derived from Jf and
l0 to be Jm. This inequality forms an ellipsoid. Now we calculate the
x0 ˆ …0; 0; z0 ; 0; 0; 0†; l0 ˆ g…x0 † ratio of the minor axis to the major axis of the ellipsoid
(Eq. (6)), i.e.
where z0 is the height of the upper plate when the spring is
the minor axis
neither stretched nor compressed. The linearized inverse sf ˆ
kinematics is given by the major axis

l ÿ l0 ˆ J…x0 †…x ÿ x0 † where sf ˆ 1 implies complete isotropy of the force F in all

directions of the Stewart platform.
From this linearized solution of inverse kinematics, the Similarly, the analysis of isotropy of moment acting on the
solution of forward kinematics of the Stewart platform can upper plate can be carried out for the case the upper plate can
be expressed as follows. exert moment M without generating any translational
motion. The evaluation of the moments M for this case
x ˆ x0 ‡ J ÿ1 …x0 †…l ÿ l0 † (4)
results in an ellipsoid such that
Substituting this solution (4) into (3) gives the relationship
between leg forces f and force/torque F acting on the upper M T RM  1 (7)
plate. and the ratio sm of the minor axis to the major axis of the
Externally applied forces and moments on the upper plate ellipsoid (Eq. (7)) can be calculated. Based on these values
then is obtained as follows. One directional force on each leg is sf and sm, the radii of the lower and upper plates, leg lengths,
converted to voltage and is read through an A/D converter, and and distances between joints of the Stewart platform are
the deviation of leg length is calculated using the spring designed.
constant k. Then the position and orientation of the upper Considering manufacturing problems of the transducer,
plate is directly determined from the Eq. (4) and the externally the distances between joints (A1A2, A3A4, B6B1, B2B3, etc.)
applied forces and moments is obtained using Eq. (3). of the Stewart platform used in this paper are all set to d, and
the angles between A1 and A3, A3 and A5, B1 and B3, B3 and
B5 are set to 1208. Furthermore, for manufacturing purpose,
3. Kinematic structure design and error analysis
the following constraints for the kinematic parameters of the
Stewart platform are imposed.
There could be several performance indexes for optimum
design of the kinematic structure of the Stewart platform 60 mm  rA  170 mm
[14,15]. Kosuge et al. [14] evaluated the kinematic structure 60 mm  rB  170 mm
of the Stewart platform in terms of isotropy of measurable 30 mm  d  50 mm
forces and moments via force analysis dealing with force 110 mm  li  170 mm; i ˆ 1; . . .; 6
and moment independently.
Under these constraints, we ®rst select li and d using the
In this section, we determine kinematic design parameters
data of the existing components, and calculate the values sf
of the Stewart platform using the method proposed by
and sm for the parameters rA and rB (see Fig. 2). After several
Kosuge et al. [14]. First, Eq. (3) is divided into two sub-
trials for rA and rB, we determine rA and rB values. The
equations as follows.
kinematic parameters of the Stewart platform being used as
   T
F J the 6-axis force transducer are ®nally selected from the
ˆ fT f above analysis as follows.
M Jm
where F…2 R3 † and M…2 R3 † represent three force compo- li ˆ 133:5 mm; i ˆ 1; 2; . . .; 6
nents and three moment components, respectively, and
d ˆ 30 mm
Jf …2 R63 † and Jm …2 R63 † represent sub-matrices related
to force and torque, respectively. rA ˆ 134 mm; rB ˆ 90 mm
34 C.-G. Kang / Sensors and Actuators A 90 (2001) 31±37

torque F, consider the case that the upper plate is maximally

displaced only in the yA direction (yc ˆ 0:80 mm). In this
case, the forces acting on the six legs are as follows.
f1 ˆ 49:62 N; f2 ˆ 20:36 N; f3 ˆ ÿ29:17 N;
f4 ˆ 29:52 N; f5 ˆ ÿ19:97 N; f6 ˆ ÿ49:41 N
When these forces are acting on the legs, force/torque F
acting on the center of the upper plate is calculated by using
Fig. 2. sf, sm for the radii of the lower and upper plates (d ˆ 30 mm,
the proposed method using linearization technique. For
li ˆ 133:5 mm). comparison purpose, force/torque F is also calculated by
neglecting the movement of the upper plate. The results are
compared to the exact values in Table 1. The forces and
4. Force error evaluation of the proposed method torques in Table 1 are values expressed with respect to frame
{A}.
The 6-axis force transducer based on the Stewart platform Similarly, forces/torques acting on the upper plate can be
has linear voltage differential transformers (LVDT) mounted obtained for the cases the upper plate has maximum linear
along the legs to measure the de¯ection of springs, which displacement only in xA direction and only in zA direction,
will be used to calculate leg forces along leg directions. We maximum angular displacement only in xA direction, only in
assume that the maximum displacement of springs (i.e. the yA direction, and only in zA direction. The force error
change of leg lengths) is small compared to the leg length. evaluation in these cases are summarized in Table 2. In
The force/torque F acting on the center of the upper plate order to make the table simple, we do not present the errors
is obtained from the Eq. (3) in which forces f acting on the of small torques existing in the cases of linear displace-
legs are calculated from the data of LVDT de¯ections and ments, and the errors of small forces existing in the cases of
spring constants. In Eq. (3), the Jacobian matrix J is a angular displacements.
function of the position and orientation of the upper plate, Tables 1 and 2 show that the errors of the calculated forces
x, which is the solution of forward kinematics of the Stewart and torques of the given 6-axis force transducer with the
platform. Instead of using the exact solution x which is not designed parameters are within 0.005% (in case leg devia-
known, an approximate solution is used in this paper. tions are 0.43% of leg lengths) when the proposed method
For the purpose of examining accuracy of the proposed using the linearization technique is used. In other words, the
method, the force F obtained from the proposed method is result by the proposed linearization technique is exactly
compared to the exact value, and also compared to the value same until two digits below decimal point with the true
calculated from the case in which the movement of the upper values for the given transducer, but the resulting error
plate due to small deviation of leg lengths is neglected, i.e. becomes much bigger when the motion of the upper
J…x† ˆ J…x0 † is assumed. plate is neglected. Further calculation shows that the
Let the spring constant of springs mounted on the legs to errors in F are within 0.03% when leg deviations are
be 87.5 N/mm, and the maximum deviation of the legs to be 2% of leg lengths and the proposed linearization technique
0.57 mm (0.43% of the leg length). The design para- is used.
meters of the Stewart platform were selected in Section 3. In In other words, the proposed closed-form solution has
order to investigate the accuracy of the calculated force/ 0.005% maximum error due to linearization in force and

Table 1
Obtained force/torque F when the upper plate is maximally displaced only in the yA direction

Fx (N) Fy (N) Fz (N) Mx (N m) My (N m) Mz (N m)

Exact values ÿ0.11 105.68 0.22 ÿ0.56 0.03 0.01

Using linearization ÿ0.11 105.68 0.22 ÿ0.56 0.03 0.01
Neglecting movement ÿ0.04 105.78 0.68 ÿ0.56 0.01 0.01

Table 2
Maximum errors in F when the upper plate has maximum linear and angular displacements in xA, yA, zA directions (0.43% leg deviations)

Linear displacement Angular displacement

xA (N) yA (N) zA (N) xA (N m) yA (N m) zA (N m)

Using linearization 0.002 0.001 0.002 0.000 0.000 0.000

Neglecting movement 0.200 0.107 0.667 0.012 0.030 0.021
 C.-G. Kang / Sensors and Actuators A 90 (2001) 31±37 35

moment measuremnets, if 0.43% leg deviation is allowed, moment (i.e. couples) on the transducer is generated by
and 0.03% maximum error, if 2% leg deviation is allowed using a pulley and string mechanism.
with the selected design parameters. Actually, these errors We de®ne the clockwise rotation of the transducer in a
are within noise levels come from sensor electronics and horizontal plane of the experimental set-up to be a yaw
environmental changes. This fact shows the validity of the motion, and the rotation of the transducer about the axis of
proposed method using linear transformation matrix instead the encoder to be a pitch motion. In order to evaluate the
of using numerical solutions of nonlinear equations. effect of the weight of the transducer on measurement,
forces and moments in case that any external forces and
torques are not applied, are measured at every 458 in yaw
5. Experimental results direction and at every 68 in pitch direction, and then a look-
up table is constructed from these measurement data. Fig. 5
In order to investigate the validity of the proposed method, shows one example of the measurement results.
a prototype of the transducer based on the Stewart platform Since the gravity effect of the transducer weight on force
is developed, which is shown in Fig. 3. Each leg is composed and torque measurements is expectable if the position and
of two springs (one for compression, and one for extension), orientation of the transducer is known in advance, it can be
LVDT and core. It can measure up to 5 kgf (1 VDC/1 kgf) compensated by using the look-up table generated above and
and can deviate 0.57 mm maximally. The six legs and linearly interpolating it. Fig. 6 shows the measurement result
upper and lower plates are connected using ball and socket after gravity compensation in the pitch direction.
joints and joint adapters. The upper and lower plates and When 40 N force is applied to the center of the upper plate
joint adapters are made from Al 7075 and Al 2024, respec- of the transducer to evaluate the gravity compensation for
tively, in order to reduce weights. transducer weight, the difference of the true and measured
Fig. 4 shows the schematics of the experimental set- data is within 2.6% in force measurements and within 0.54%
up. The encoder and counter board in the ®gure are for in torque measurements. Thus, the gravity effect due to the
the measurement of the angle of the transducer. If net transducer weight is compensated effectively.
external force and torque acting on the upper plate is Next, the experiments on force measurements are con-
required, the gravity effect of the upper plate should be ducted for arbitrary forces and moments with arbitrary
compensated from the known position and orientation of the magnitudes and directions. Linearly measurable ranges of
lower plate and the known weight of the upper plate. For this forces and moments of the transducer is determined by
purpose, gravity effect is measured at each rotated angle of measuring forces and moments with increasing external
the transducer and compensated by using the measurement forces and moments in each direction until saturations
data and an interpolation algorithm. Externally applied pure occur.

Fig. 3. Photograph of the 6-axis force transducer.

Fig. 5. Force and torque generated by the weight of the transducer (in case
Fig. 4. The schematic diagram of the experimental set-up. yaw ˆ 0 ).
36 C.-G. Kang / Sensors and Actuators A 90 (2001) 31±37

Table 3
The ranges of force and torque measurements

Fx (N) Fy (N) Fz (N) Mx (N m) My (N m) Mz (N m)

78 80 185 9.5 8.9 13

ÿ78 ÿ80 ÿ180 ÿ9.5 ÿ8.9 ÿ13

Fig. 7 shows one of the measurement results, and Table 3

summarizes the linearly measurable range of force and
torque of the transducer in each direction. Experimental
results show that the linearity is within 1.58% in force
measurements and within 0.83% in torque measurements,
and the accuracy is within 3.5% in force measurements and
within 7.5% in torque measurements.

6. Conclusion

Stewart platform-based force transducers require the

solutions of forward kinematics of the Stewart platform.
Since the solution of forward kinematics of the general
Stewart platform is not known in a closed-form, an approx-
Fig. 6. Gravity-compensated results by a linear interpolation of the look- imate closed-form solution has been derived via lineariza-
up table. tion technique and has been applied to a 6-axis force
transducer in this paper.
Numerical calculations have shown that errors in force
measurements generated by using the approximate closed-
form solution are within 0.005% of measured forces (under
given design parameters), and these errors are actually less
than noise levels. Experimental study on the 6-axis force
transducer using the proposed method has shown that the
linearity is within 1.58% in force measurements and within
0.83% in torque measurements, and the accuracy is within
3.5% in force measurements and within 7.5% in torque
measurements under given experimental conditions.
Therefore, using the proposed closed-form solution of
forward kinematics is valid for Stewart platform based-force
transducers instead of using numerical iterative solutions or
other solutions obtained from indirect complex ways, if
structural deformation of the transducer is small (0.005%
force error due to 0.43% leg deviation, and 0.03% force
error due to 2% leg deviation with design parameters in
this paper).
Furthermore, it has been shown that the proposed gravity
compensation algorithm is effective for reducing the effect
of transducer weight on measurements, which is based on a
look-up table and linear interpolation.

Acknowledgements

The author acknowledges the ®nancial support of Konkuk

University for this work in 1999. Aslo, he wish to thank
Mr. J.-H. Han for experimental help and Mr. J.-B. Jeong and
Fig. 7. Results of force and torque measurements. J.-H. Kim for preparing ®gures.
 C.-G. Kang / Sensors and Actuators A 90 (2001) 31±37 37

References [12] P. Nanua, K.J. Waldron, V. Murthy, Direct kinematic solution of

Stewart platform, IEEE Trans. Robotics Automat. 6 (1990) 438±443.
[13] H.-Y. Lee, B. Roth, A closed-form solution of the forward
[1] D.M. Perry, Has force/torque sensing gained factory acceptance?, in:
displacement analysis of a class of in-parallel mechanisms, in:
Proceedings of the IEEE/RSJ International Conference on Intelligent
Proceedings of the IEEE International Conference on Robotics and
Robots and Systems, 1992, pp. 44±46.
Automation, 1993, pp. 720±724.
[2] A.K. Bejczy, Smart Sensors for Smart Hands, in R. Brechen Rielge
[14] K. Kosuge, M. Okuda, H. Kawamata, T. Fukuda, Input/output force
(Ed.) Prog. Astronaut. Aeronaut., D. Durand, C. Diaz., 67 AIAA
analysis of parallel link manipulators, IEEE J. Robotics Automat.
1979 pp. 275±304.
(1993) 714±719.
[3] A. Ch'Hayder et al., US Patent No. 5,063,788, 12 November
[15] C. Gosselin, J. Angeles, The optimum kinematic design of a planar
1991.
three-degree-of-freedom parallel manipulator, J. Mech. Trans.
[4] M.M. Svinin, M. Uchiyama, Optimal geometric structures of force/
Automat. Design 110 (1988) 35±41.
torque sensors, Int. J. Robotics Res. 14 (1995) 560±573.
[5] JR3 Inc., Product Catalogue, Users Manual, 1997.
[6] D.R. Kerr, Analysis, properties, and design of Stewart-platform
transducer, J. Mech. Trans. Automat. Design 111 (1989) 25±28. Biography
[7] D. Stewart, A platform with six degree of freedom, Proc. Inst. Mech.
Eng. 180 (1965) 371±386. Chul-Goo Kang received his BS and MS degrees in Mechanical Design
[8] A. Gaillet, C. Reboulet, A isostatic six component force and torque and Production Engineering from Seoul National University (Korea) in
sensors, in: Proceedings of the 13th International Symposium on 1981 and 1985, respectively, and his PhD degree in Mechanical
Industrial Robots, 1983, pp. 102±111. Engineering from the University of California, Berkeley (USA) in 1989.
[9] C.C. Nguyen, S.S. Antrazi, Z.L. Zhou, C.E. Campbell Jr., Analysis He joined Konkuk University (Seoul, Korea) in 1990 as a faculty member,
and experimentation of a Stewart platform-based force/torque sensor, and currently is Professor of Mechanical Engineering. His current research
Int. J. Robotics Automat. 7 (1992) 133±140. interests include force/torque sensing, beacon guidance, brake systems of
[10] D.C.H. Yang, T.W. Lee, Feasibility study of a platform type of high-speed trains, and intelligent control logic. He is a member of the
robotic manipulators from a kinematic viewpoint, J. Mech. Trans. American Society of Mechanical Engineers, the Korean Society of
Automat. Design 106 (1984) 191±198. Mechanical Engineers, the Korea Fuzzy Logic and Intelligent Systems
[11] O. Ma, J. Angles, Architecture singularities of platform manipulators, Society, Korean Society of Precision Engineering, and the Institute of
in: Proceedings of the IEEE International Conference on Robotics Control, Automation and Systems Engineers. He is currently an Editor of
and Automation, 1991, pp. 1542±1547. the Journal of Control, Automation and Systems Engineering.