See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228710121

Dimensional Synthesis of Bennett Linkages

Article in Journal of Mechanical Design · April 2001

DOI: 10.1115/1.1539507

CITATIONS READS
67 354

2 authors:

Alba Perez-Gracia J. Michael McCarthy

Universitat Politècnica de Catalunya University of California, Irvine
85 PUBLICATIONS 1,050 CITATIONS 281 PUBLICATIONS 5,184 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Alba Perez-Gracia on 04 June 2014.

The user has requested enhancement of the downloaded file.

DIMENSIONAL SYNTHESIS OF BENNETT
LINKAGES
Alba Perez, J. M. McCarthy
Robotics and Automation Laboratory
Department of Mechanical and Aerospace Engineering
University of California
Irvine, California 92697
Email: maperez@uci.edu, jmmccart@uci.edu
March 6, 2001

Abstract
This paper presents a synthesis procedure for a spatial 4R linkage,
known as Bennett’s linkage. It is known that the synthesis of the
spatial RR chain for three positions yields two solutions, which form
a Bennett linkage. While analytical solutions to these equations have
been developed previously, this paper uses the cylindroid that is known
to exist for a Bennett linkage to simplify the solution process. It is
interesting that geometric constraint associated with the spatial 4R
chain simplifies the solution of the RR chain design equations. An
example design is presented.

1 Introduction
In this paper we present a procedure to design spatial 4R chains known
as Bennett linkages, using the synthesis equations for the spatial RR
chain. Veldkamp [Veldkamp, 1967] solved the design equations of a
spatial RR chain for three instantaneous positions, and showed that
the two solutions form a Bennett linkage. Suh [Suh, 1969] solved the
problem for three finite positions, first numerically and later using

1
 W
a
α H
γ g
G

Ground link
U

Figure 1: A Bennett linkage

screw geometry, and showed that in all cases, the two solutions formed
a Bennett linkage. Tsai and Roth [Tsai and Roth, 1973] solved alge-
braically the 10 quadratic equations for three position synthesis using
the screw triangle formulation and showed that it always has two so-
lutions which form a Bennett linkage.
We use Huang’s result [Huang, 1996] that the finite displacement
screws of a Bennett linkage form a cylindroid to determine a coordi-
nate frame in which the design equations simplify. The result is three
linear equations and one cubic polynomial in four design parameters.

2 The Bennett Linkage

The Bennett linkage is formed by connecting the end-links of two
spatial RR chains to form a coupler, Figure 1. In general, the mobility
for a 4R closed spatial chain is equal to -2, applying for instance
Gruebler’s criterion,

X
m
M = 6(n − 1) − pk ck = 6.3 − 4.5 = −2. (1)
k=1
However, the Bennett linkage moves with one degree of freedom,
due to its special geometry. In order to move, the twist angles and
link lengths of the opposites sides of this linkage, (α, a) and (γ, g),
must be equal. This together with the condition
sin α sin γ
= , (2)
a g

2
 Moving Axis

Wi
α
G
a
End Effector
Fixed Axis B
Pi

Figure 2: A spatial RR robot.

ensures that the linkage moves with one degree of freedom.

To design the Bennett linkage we use the design equations for
an RR chain. The RR chain consists of a fixed revolute axis G =
(G, B × G)T = (G, R)T connected to a moving revolute axis Wi =
(Wi , Pi × Wi )T = (Wi , Vi )T by a rigid link. The vector G denotes
the direction of the fixed axis and B is a point on this line. The vector
Wi is the direction of the moving axis in the ith position and Pi a
point on this line. We choose B and Pi to be the intersection points
of these lines with the common normal N to the two axes, Figure 2.
In order to design the RR chain we need to determine the directions
G and W1 of the fixed and moving axes and the coordinates of the
points B and P1 . This is a total of 10 design parameters. The solution
of the design equations yields two RR chains which form the Bennett
linkage.

3 Geometry of the RR Chain

The workspace of the RR chain plays an important role in the simpli-
fication of the design equations. The kinematics equation for the RR
chain in matrix form represents the workspace, or the set of displace-
ments [D(θ, φ)], of the RR chain.

3
 [D] = [G][Z(θ, 0)][X(α, a)][Z(φ, 0)][H], (3)

where [G] and [H] are the initial and final transformations from the
fixed frame to the fixed axis and from the moving axis to the end
effector, respectively.
Choosing a reference configuration [D1 ], we can construct the set
of the relative displacements [D1i ] = [Di ][D1 ]−1 ,

[D1i ] = [T (∆θi , G)][T (∆φi , W)], (4)

where

[T (∆θ, G)] = [G][Z(θ, 0)][Z(θ0 , 0)]−1 [G]−1 ,

[T (∆φ, W)] = ([G][[Z(θ0 , 0)][X(α, a)])[Z(φ, 0)][Z(φ0 , 0)]−1
([G][[Z(θ0 , 0)][X(α, a)])−1 (5)

The set of relative screw displacements reachable by the RR chain

is achieved by a spatial rotation by ∆φ about the W1 axis followed by a
rotation by ∆θ about the G axis. Let Ĝ( ∆θ ∆θ ∆θ
2 ) = cos( 2 )+sin( 2 )G and
1 ∆φ ∆φ ∆φ 1
Ŵ ( 2 ) = cos( 2 ) + sin( 2 )W be the dual quaternions associated
with these rotations [Bottema and Roth, 1990], [McCarthy, 1990]. The
dual quaternion Ŝ1i which represents the displacement of the end-link
is obtained by the dual quaternion product Ŝ1i = ĜŴ 1 , which yields
the dual scalar

ψ̂1i ∆θ ∆φ ∆θ ∆φ
cos( ) = cos cos − sin sin G · W1 , (6)
2 2 2 2 2
and the dual vector

ψ̂1i
sin( )S1i =
2
∆θ ∆φ ∆φ ∆θ 1 ∆θ ∆φ
sin cos G + sin cos W + sin sin G × W1 . (7)
2 2 2 2 2 2

The dual angle ψ̂1i defines the relative rotation and translation of the
end-link along the screw axis S1i . Notice that the pitch corresponds
to screws representing finite displacements [Parkin, 1992].

4
 ψ̂1i ψ1i ψ1i t1i /2
sin( ) = (sin( ), sin( ) ). (8)
2 2 2 tan(ψ1i /2)

By varying the joint angles ∆θ and ∆φ in (7) we obtain a set of

screw axes that form a screw system of third order [Huang, 1992]. It
defines the set of screw axes of relative displacements of the RR chain,
and hence the set of positions reachable by the RR chain.
If, on the other hand, the values of θ and φ are coupled by the
Bennett linkage, then Hunt [Hunt, 1978] shows that

φ sin α+γ
2 θ θ
tan = α−γ tan 2 = K tan 2 , (9)
2 sin 2

where K is the constant obtained from the dimensions α and γ. Notice

that (9) gives a relation for absolute and not relative angles. The
condition translates to relative angles to obtain an expression that
depends on both the relative angle and the initial angle θ1 ,

K tan ∆θ 2 θ1
∆φ 2 (1 + tan 2 )
tan = . (10)
2 1 + (K 2 − 1) tan θ21 tan ∆θ 2
2 + K (tan
2 θ1
2)

If we apply this relation to the workspace of the RR chain, equation

(7), we obtain the set of relative displacement that the Bennett linkage
can realize,

a sin α sin ψ21i t1i

2
(1, )S1i =
a sin α cos ψ21i − t1i
2 cos α sin ψ21i tan ψ21i
∆θ ¡ ∆θ ¢
= tan G + Kθ W1 + tan Kθ G × W 1 ,
2 2
(11)

where

K(1 + K12 )
Kθ = ,
2 K1 (K − 1)
1 + K 2 K12 + tan ∆θ 2

θ1
K1 = tan . (12)
2

5
Figure 3: The cylindroid as viewed first from the top along the central axis,
then from the side and finally, from an angle.

Hence just by specifing ∆θ = θi − θ1 we obtain the set of all

posible screw displacements of the Bennett linkage. If we normalize
by dividing by their magnitude, this set of screw axes generates a
cylindroid, Figure 3.
A cylindroid is a ruled surface which appears as generated by the
real linear combination of two finite screw axes. Equation (10) defines
the relation between the coupler and input angle for a Bennett linkage,
hence the cylindroid is the locus of relative screw displacements for a
Bennett linkage.
On the following sections we will relate the cylindroid generated
by the Bennett linkage with the one that contains the relative screws
S12 and S13 defined by three positions.

4 The Cylindroid
A cylindroid is a simply-ruled surface that has a nodal line cutting all
generators at right angles. In an appropiate coordinate frame, we can
write its algebraic equation as

z(x2 + y 2 ) + (PX − PY )xy = 0,

where PX − PY is the difference of the principal screws of the cylin-

droid.
The cylindroid appears as the axes of a real linear combination of
two screws. Assume that we have designed an RR chain that passes
through three spatial positions M1 , M2 and M3 . Then the relative
screw axes S12 and S13 must lie on the cylindroid defined by (11). In
fact these two screws, as any other pair contained in the cylindroid,

6
must generate this cylindroid. This is the key to our formulation of
the RR design problem.
To describe the geometry of the cylindroid, let us consider the
screws obtained from the design positions. For this calculation it is
convenient to work with the vector part of the quaternion product (7)
and define the screws

ψ̂12 ψ̂13
Va = sin S12 , Vb = sin S13 . (13)
2 2

The dual number sin ψ̂21i = (sin ψ21i , t21i cos ψ21i ) contains the magnitude
and pitch of the screw,

ψ̂ ψ
sin = sin (1, P ), (14)
2 2
where the pitch P is given by the expression
t12 t13
Pa = ψ12
, Pb = . (15)
2 tan 2 2 tan ψ213

The two screws Va and Vb are independent and generate a cylin-

droid. To simplify the following calculations denote the screw axis S12
as I and the common normal line to S12 and S13 as K. Then we have
J = K × I to form a frame I, J, K, and

S12 = I, S13 = cos δ̂I + sin δ̂J. (16)

where δ̂ = (δ, d) is the dual angle between S12 and S13 .

The real linear combination of Va and Vb yields the screws

F = a(1, Pa )I + b(1, Pb )(cos δ̂I + sin δ̂J), (17)

where we have absorbed sin ψ212 and sin ψ213 into the constants a and
b, respectively. The axes of the screws F form a cylindroid.

5 The Principal Axes

The cylindroid consists of lines that intersect a central axis, called
nodal line, which is the common normal to all of them. It can be
proved that, at each point of the nodal line and between a minimum
and a maximum distance, there are two axes that intersect forming a

7
 K

Y
S13

X
z0
σ
S12

Figure 4: The principal axes as located from the initial screws S12 and S13

variable angle. This surface has a set of principal axes consisting of the
common normal and the only pair of lines that intersect at an angle
of 90 degrees. This occurs at the midpoint of the cylindroid along the
central axis. The expression of the cylindroid in the coordinate frame
defined by the principal axes is considerably simpler. The following
formulation combines the results of [Hunt, 1978] and [Parkin, 1997] to
define the principal axes.
We locate the axis of the general screw F of the cylindroid using
the dual angle ζ̂ = (ζ, z), that is

F = F (1, P )(cos ζ̂I + sin ζ̂J), (18)

where F is the magnitude of this screw and P is its pitch. Expand

(17) to obtain
¡ ¢
F = a + b cos δ, aPa + b(Pb cos δ − d sin δ) I
¡ ¢
+ b sin δ, b(Pb sin δ + d cos δ) J (19)

and equate to (18). We separate the real and dual parts of this result,
½ ¾ · ¸½ ¾
cos ζ 1 cos δ a
F = (20)
sin ζ 0 sin δ b

and
· ¸½ ¾ · ¸½ ¾
− sin ζ cos ζ z P Pb cos δ − d sin δ a
F = a . (21)
cos ζ sin ζ P 0 Pb sin δ + d cos δ b

8
 Now we eliminate the scalars a and b using both equations to
obtain
½ ¾ · ¸· ¸½ ¾
z(ζ) − sin ζ cos ζ Pa (Pb − Pa ) cot δ − d cos ζ
F =
P (ζ) cos ζ sin ζ 0 Pb + d cot δ sin ζ
(22)

This equation gives the pitch P and distance z of the screw for any
value of the angle ζ. We use this expression to locate the principal
axes of the cylindroid, as the screws with maximum and minimum
value of pitch. Computing the derivative of the above expression we
obtain
½ ¾ · ¸½ ¾
d z −(Pb − Pa ) cot δ + d (Pb − Pa ) + d cot δ sin 2ζ
= .
dζ P (Pb − Pa ) + d cot δ (Pb − Pa ) cot δ − d cos 2ζ
(23)

Set the second of these equations to zero to determine the angles ζ = σ

for extreme values of pitch P , given by
−(Pb − Pa ) cot δ + d
tan 2σ = . (24)
(Pb − Pa ) + d cot δ
This yields two angles separated by π/2. They define the principal
axes X and Y of the cylindroid.
The offset z(σ) at which the principal axes are located is given by
substituting the angle σ in equation (22),
1 cos δ
z0 = (d − (Pb − Pa ) ). (25)
2 sin δ
The principal axes can be expressed, using the dual angle σ̂ = (σ, z0 )
as

X = cos σ̂I + sin σ̂J, Y = − sin σ̂I + cos σ̂J. (26)

The principal axes capture the symmetry of the cylindroid. By

expressing the synthesis problem in this coordinate frame we obtain
simpler design equations.

6 The Design Equations

The design equations for the RR chain are obtained from the geo-
metrical constraints imposed by the link connecting the moving and

9
fixed axes. Let α and a denote the twist angle and distance between
these axes. Clearly they must remain constant during the movement.
Furthermore, the axes can not slide, so that the normal line to both
axes remains the same. These constraints can be found in Suh and
Radcliffe [Suh and Radcliffe, 1978], and [McCarthy, 2000]. Tsai and
Roth [Tsai and Roth, 1973] use a similar formulation based on the
equivalent screw triangle.

6.1 The matrix formulation

Let the three positions be defined by the 4×4 homogeneous transforms
[Ti ] = [Ai , di ], i = 1, 2, 3 where [Ai ] is a 3 × 3 rotation matrix and di
is a 3 × 1 translation vector. Construct the relative displacements
[T1i ] = [Ti ][T1−1 ], i = 2, 3. Associated with each of these matrices is a
6 × 6 coordinate transformation for screws,
· ¸
A1i 0
[T̂1i ] = , (27)
D1i A1i A1i

where D1i is the 3 × 3 skew-symmetric matrix obtained such that

[D1i ]y = d1i × y.
Dual vector algebra now allows to define one set of constraint equa-
tions for the RR chain such that

G · [T̂1i − I]W1 = 0, i = 2, 3, (28)

which is the same equation that appears in the design of CC chains.

Separating this dual equation into real and dual components we obtain
the two scalar direction constraints

G · [A1i − I]W1 = 0, i = 2, 3, (29)

and the two scalar moment constraint equations

G · [A1i − I]V1 + W1 · [A1i − I]T R+G · [D1i A1i ]W1 = 0,

i = 2, 3. (30)

These equations ensure that the dual angle α̂ = (α, a) is constant in

the three positions.
In order to ensure that the common normal passes through the
same points on the axes, B and Pi = [T1i ]P1 , we require Pi − B to

10
 S12

C12

ψ12
2 W1
G t12
2
n1 P1 n3
βG βW

B
n2

Figure 5: The equivalent screw triangle for the RR chain

be perpendicular to G and [T1i ]−1 B − P1 to be perpendicular to W1

in each position, that is, we have the six equations

G · ([T1i ]P1 − B) = 0, W1 · (P1 − [T1i ]−1 B) = 0,

i = 1, 2, 3. (31)

The 10 equations (29), (30), (31) are solved to determine G and W1 .

6.2 The Equivalent Screw Triangle

The equivalent screw triangle formulation of Tsai and Roth allows
us to write some of the constraint equations in a different way. For
each relative transformation we can construct the screw axis S1i and a
rotation angle ψ1i and a translation t1i . The equivalent screw triangle
defines the relationship between the screw axis S1i and the fixed and
moving axes G and W1 , Figure 5.
Let C1i be a reference point on the screw axis S1i . The direction
constraint in (29) can be reformulated using Cayley’s equation,

2
[A1i − I] = [B1i − B1i
T
B1i ], (32)
1 + tan2 (ψ1i /2)

11
where [B1i ] is the skew-symmetic matrix corresponding to the rotation
axis vector, B1i = tan(ψ1i /2)S1i , and [S1i ]v = S1i × v. Then we just
need to write the direction equation as

kG · ([S1i ] − k 0 [S1i
T
S1i ])W1 =
kG · (S1i × W1 ) − kk 0 ((S1i × G) · (S1i × W1 )) (33)

to obtain

ψ1i G · (S1i × W1 )
tan = . (34)
2 (S1i × G) · (S1i × W1 )

The properties of the screw triangle yield an alternative set of

equations for the moment constraints. Notice that the geometry of
the dyad triangle requires the common normal lines to G and Wi to
be separated by a distance t1i /2 along S1i . Thus the component of
B − P1 in the direction S1i is given by
t1i
(B − P1 ) · S1i − = 0, i = 2, 3. (35)
2
In this case the points on the lines are required to be on the common
normal line.
To complete the set of design equations, we transform the equa-
tions (31) to obtain

G · (P1 − B) = 0,
W1 · (P1 − B) = 0, (36)

and

G · [I − S1i ST1i ](B − C1i ) = 0,

W1 · [I − S1i ST1i ](P1 − C1i ) = 0, i = 2, 3, (37)

where C1i is any point on S1i .

This is a set of ten quadratic equations in ten unknowns, the co-
ordinates of the line G and the line W1 in its first position.

12
 b K
W 1

P1

X
c

c
2
C1
G
Y
κ
B
a

Figure 6: The tetrahedron defines the Bennett linkage.

7 Bennett Linkage Coordinates

7.1 The tetrahedron and the cylindroid
Yu [Yu, 1981] introduced a coordinate frame aimed to simplify the
expression of the Bennett linkage. The joints of the Bennett linkage
can be determined using a tetrahedron: each joint passes through one
of the four vertices and its direction is perpendicular to the adjacent
sides. The construction is showed in Figure 6.
Let B, P1 , Q, C1 be the vertices of the tetrahedron. The edges
are given by the difference of the vertices. The tetrahedron is oriented
so that the K axis of the cylindroid forms the common normal to the
lines defined by B − C1 and P1 − Q [Huang, 1996]. Let 2a = |B − C1 |
and 2b = |P1 − Q|, and let c and κ be the distance and angle between
both edges along K.
The tetrahedron completely defines the Bennett linkage. In order
to locate the tetrahedron with respect to the given relative positions,
we need the relation cylindroid and the tetrahedron, and specifically
the location of the tetrahedron with respect to the principal axes.
The tetrahedron of Figure 6 shows the linkage in position 1; that
is also our reference position, that is, the position we measure the
angles from. For the following derivation of the relation between the
tetrahedron and the cylindroid, we are going to refer to a coordinate

13
 la

α W1 α+γ
Kp
γ U1
(α+γ) H Yp
2 C1
G=Z0
P1
Y0 Q
X0 Xp
B b
lg

a =la+lg
2

b Kp
LPQ W1
NGH=X0
P1

Xp lg H
c
la θ1
c LBC
2 C1
G=Z0
Yp
κ
Y0
B
a

Figure 7: The Bennett linkage with initial angle θ = 0 and θ = θ1

frame that will not change as the Bennett linkage moves. This is the
frame {NGH , G × NGH , G}.
It is interesting because of its simplicity, to look first at the location
of the cylindroid with respect to the tetrahedron when this is in the
folded position, that is, when we are measuring the angles from the
initial angle θ = 0.
We construct the expression of the relative screw axes generated
by the linkage in this new coordinate frame, using equation (11).The
screws of the relative displacement are given by the values of the
absolute angle θ,

   
 K sin α tan2 2θ  θ
 Kla cos α tan2 2 
S0i = −K sin α tan 2θ + ² −Kla cos α tan 2θ . (38)
 θ  
(1 + K cos α) tan 2 −Kla sin α tan 2θ

14
 We can locate the principal axes, as before, as the screw axes with
maximum and minimum pitch. The pitch of the normalized screw
axes is given by

Kla sin α(K cos α sin2 θ

2 − cos2 2θ )
P = θ
(39)
(1 + 2K cos α + K 2 cos2 α) cos2 2 + K 2 sin2 α

The maximum pitch correspond to the values of θ = π, 2π; these

angles correspond to the principal axes Xp and Yp . The angle θ = π
corresponds to the Xp -axis and we can see that it is located along the
X0 -axis. The value of θi = 2π corresponds to the Yp -axis, as seen in
Figure 7; it lies also on the plane perpendicular to the X0 -axis and the
intersection point with the Xp -axis is

 la +lg 
 2 
z0 = 0 , (40)
 
0

which correspond to the mid point of the tetrahedron when in the

folded position.
To find the nodal line, the Zp axis, we use two arbitrary screw axes
of the cylindroid. We obtain

   
 0   0 
Kp = cos α+γ + ² − 1
(l a + l g ) sin α+γ
, (41)
 2   12 α+γ 
2
sin α+γ
2 (l
2 a + l g ) cos 2

which again intersects the X0 line in the midpoint of the tetrahedron,

and is oriented with an angle (α + γ)/2.
For the case of an initial angle θ1 different from zero, consider the
same frame defined by {NGH , G×NGH , G}, which remains fixed during
the movement of the linkage. Given the expression of the moving axis
in this frame,

   
 sin α sin θ1   la cos α sin θ1 
W1 = − cos θ1 sin α + ² −la cos α cos θ1 , (42)
   
cos α −la sin α

15
and using eq.(11), we can construct the expression of the screws of the
relative displacement, which we are going to call S1i . The expression
for the direction part is

 θ−θ1 θ−θ 


K sin α sin (sin θ1 +cos θ1 tan 2 1 ) 

 
2

 cos
θ1 θ θ θ
cos 2 +K 2 sin 21 sin 2 



2 


 

 θ−θ1 θ−θ 
K sin α sin (sin θ1 tan 2 1 −cos θ1 )
S1i = 2
, (43)

 cos
θ1 θ
cos θ2 +K 2 sin 21 sin θ2 



2



 


 

 
θ−θ1
K cos α sin
tan θ−θ
2 +
1
θ1 θ
2
θ θ

cos 2
cos 2
+K 2 sin 21 sin 2

and for the moment vector,

 θ−θ θ−θ 


Kla cos α sin 2 1 (sin θ1 +cos θ1 tan 2 1 ) 


 θ θ 



cos 21 cos θ2 +K 2 sin 21 sin θ2 


 


 θ−θ1 θ−θ1


Kla cos α sin (sin θ1 tan −cos θ1 )
S01i = 2 2
. (44)

 cos
θ1
cos θ2 +K 2 sin
θ1
sin θ




2 2 2



 


 

 
θ−θ1
Kla sin α sin
 θ1 θ
2
θ θ

cos 2
cos 2
+K 2 sin 21 sin 2

We construct the nodal line K in this frame from the geometry of

the tetrahedron, Figure 6,

   
2 la (cos α − cos γ) sin θ1
1
 sin α sin θ1   
K = sin γ − sin α cos θ1 + ² − 1
(cos α − cos γ)(l a cos θ 1 + l g ) ,
   2 1 (cos2 α−cos2 γ) 
cos α − cos γ l
2 a sin α
(45)

and apply the dot product with K = K + ²K0 , K being the direction
vector of the axis K, and S1i = Si1 + ²S01i . We obtain that

K · S1i = (K · S1i , K0 · S1i + K · S01i ) = (0, 0) (46)

for all values of the angle θ.

The first part shows that all the screws of the relative displace-
ments are orthogonal to the line K, and the second part shows that all

16
screws intersect the line K; hence this is the nodal line of the cylindroid
and forms its principal axes. To locate the Xp , Yp principal axes we
find the distance from a given axis of the cylindroid to the principal
axes, using equation (25). We obtain that the principal axes intersect
at the point

1 la sin α sin γ sin2 θ1



2 (l g + l a cos θ 1 − 1
2 1−cos α cos γ−sin α sin γ cos θ1 
)
1 1 sin γ(sin α cos θ1 −sin γ)
z0 = l a sin θ 1 (1 + , (47)


2 2 1−cos α cos γ−sin α sin γ cos θ1
(cos γ−cos α) sin γ sin θ1


1
4 la 1−cos α cos γ−sin α sin γ cos θ1

which corresponds to the midpoint of the tetrahedron along the line

Kp .
We also use two arbitrary axes of the cylindroid to locate the orien-
tation of the principal axes by using equation (24). Then notice that,
as we have defined κ as the angle between LP QˆLBC = Π − LCBˆLP Q ,
then by subtracting the angle between the reference axis and LCB
from the angle σ we obtain the angle κ/2, with tangent

√
2 sin α sin γ sin θ1
tan κ = √ (48)
(cos α + cos γ) 1 − cos α cos γ − sin α sin γ cos θ1

Hence the principal axes {Xp , Yp , Zp } are located with respect to

the tetrahedron at a half distance between its extreme points and Xp
bisecting the angle κ.

7.2 The Bennett linkage coordinates

Given three specified positions, we can determine the two relative
screws of the displacements, with screw axes S12 and S13 . The princi-
pal axes of the cylindroid generated by these two screws provides an
efficient form for the design equations.
We can completely describe the Bennett linkage in the principal
axes frame using the four parameters of the tetrahedron a, b, c, κ.

17
The coordinates of the vertices are given by
   
a cos κ2   b cos κ2 
B = a sin κ2 , P1 = −b sin κ2 ,
 −c   c 
2 2
   
−b cos κ2  −a cos κ2 
Q= b sin κ2 , C1 = −a sin κ2 . (49)
 c   −c 
2 2

To find the direction of the joint axes G and W1 we compute the

cross product of the edges
 
 2bc sin κ2 
G = Kg (Q − B) × (P1 − B) = Kg 2bc cos κ2 (50)
 κ κ
4ab cos 2 sin 2

and
 
 −2ac sin κ2 
W1 = Kw (B − P1 ) × (C1 − P1 ) = Kw 2ac cos κ2 , (51)
 
4ab cos κ2 sin κ2

where the constants Kg and Kw normalize the vectors. The expres-

sions of the screws G = (G, B × G)T and W1 = (W1 , P1 × W1 )T in
the principal axis coordinates are
   
 2bc sin κ2   b cos κ2 (4a2 sin2 κ2 + c2 ) 
G = Kg 2bc cos κ2 + ²Kg −b sin κ2 (4a2 cos2 κ2 + c2 ) (52)
 κ  
κ
4ab cos 2 sin 2 2abc(cos2 κ2 − sin2 κ2 )

and
   
 −2ac sin κ2  −a cos κ2 (4b2 sin2 κ2 + c2 )
W1 = Kw 2ac cos κ2 + ²Kw −a sin κ2 (4b2 cos2 κ2 + c2 ) .
   
4ab cos κ2 sin κ2 2abc(cos2 κ2 − sin2 κ2 )
(53)

Similarly, the coordinates of the second RR dyad, H and U1 , are given

by the expressions
¡ ¢
H = Kh (C1 − Q) × (B − Q) + ²Kh Q × (C1 − Q) × (B − Q)
¡ ¢
U1 = Ku (P1 − C1 ) × (Q − C1 ) + ²Ku C1 × (P1 − C1 ) × (Q − C1 ) .
(54)

18
 By using the Bennett linkage coordinates to define G and W1 , we
reduce the number of design parameters from ten to four. Further-
more, the six equations (36) and (37) are identically satisfied. The
result is four equations (34) and (35) in the unknowns a, b, c and κ.

8 Solving the Design Equations

We now transform the task positions [Ti ] to the principal axis frame
and determine the relative screw axes
ψ̂12
S12 = sin (cos δ̂1 X + sin δ̂1 Y)
2
ψ̂13
S13 = sin (cos δ̂2 X + sin δ̂2 Y), (55)
2
where ψ̂1i = (ψ1i , t1i ) are the angle and translation of the screw dis-
placements. The dual angles δ̂i are defined relative to the principal
axes frame.
Substitute (55) into (35) to obtain two linear equations in a and b.

t12 κ κ
+ (a − b) cos δ1 cos + (a + b) sin δ1 sin = 0
2 2 2
t13 κ κ
+ (a − b) cos δ2 cos + (a + b) sin δ2 sin = 0
2 2 2
They can be solved to obtain
Ks Kd
a= κ + ,
2 sin 2 2 cos κ2
Ks Kd
b= − . (56)
2 sin κ2 2 cos κ2

The constants Ks and Kd combine information from the relative screws

and are listed in Table 1.
Next we substitute (56) into the set of equations (34) and introduce
the usual definition of y = tan κ2 to eliminate the sine and cosine
functions of κ. The result is two rational equations in c and y,

2 ψ12
ψ12 Ks tan cy Ks sin δ1
tan ( − y 2 ) + c2 2 (y 2 (cos 2δ1 − 1) + cos 2δ1 + 1) − 2 K (cos δ1 + )
2 K2 2K 2 d Kd
d d
= 0,
K2 c2
( s − y2 ) + (y 2 (cos 2δ1 − 1) + cos 2δ1 + 1)
K2 2K 2
d d
(57)

19
 Constant Expression
t12 cos δ2 −t13 cos δ1
Ks 2 sin(δ1 −δ2 )
t13 sin δ1 −t12 sin δ2
Kd 2 sin(δ1 −δ2 )
³ ´
t12 /2 1
K12 ψ sin2 δ1 −sin2 δ2
tan 212
³ ´
t13 /2 1
K13 ψ sin2 δ1 −sin2 δ2
tan 213

Table 1: Constants computed from the specified positions.

2 ψ13
ψ13 Ks tan cy Ks sin δ2
tan ( − y 2 ) + c2 2 (y 2 (cos 2δ2 − 1) + cos 2δ2 + 1) − 2 K (cos δ2 + )
2 K2 2K 2 d Kd
d d
= 0,
K2 c2
( s − y2 ) + (y 2 (cos 2δ2 − 1) + cos 2δ2 + 1)
K2 2K 2
d d
(58)

We do not clear fractions and solve because the numerator and

denominator share two roots associated with c = 0, that is
Ks
c = 0, y = ± . (59)
Kd
We eliminate them from the numerator forcing the linear system
to have more solutions than the trivial,
 ψ12 
ψ12 tan Ks sin δ1
 tan 2
2 (y 2 c(cos 2δ1 − 1) + cos 2δ1 + 1) − 2 Ky (cos δ1 + ) ½ ¾
 2K 2 d  (Ks2 /Kd
Kd 2
− y2 )
 d
ψ13  = 0.
 ψ13 tan K sin δ2  c
tan 2 2 (y 2 c(cos 2δ2 − 1) + cos 2δ2 + 1) − 2 Ky (cos δ2 + sK )
2K 2 d d
d

In order for this set of equation to have roots other than those
given by (59), the determinant of the coefficient matrix must be zero.
This yields an equation that is linear in c. Solve this to obtain

c = (K13 − K12 ) sin κ (60)

where the constants K12 and K13 are shown in Table 1.

In order to determine κ, we substitute the expressions for a, b and
c into one of the one of the direction equations (34). We obtain a
cubic polynomial in y 2 ,

P : C3 y 6 + C2 y 4 + C1 y 2 + C0 = 0 (61)

20
with the coefficients given by
C3 = −Kd2 ,
C2 = Ks2 − 2Kd2 + 4(K12 − K13 )(K13 sin2 δ1 − K12 sin2 δ2 ),
C1 = 2Ks2 − Kd2 − 4(K12 − K13 )(K13 cos2 δ1 − K12 cos2 δ2 ),
C0 = Ks2 . (62)
Substitute z = y 2 , and solve the cubic polynomial to determine its
three roots.
This cubic polynomial has only one real positive root for z. To see
this we note that the value of the polynomial at z = 0 is positive; for
large positive values of z, the polynomial becomes negative because,
hence we have at least one positive root. For large negative values of
z, the polynomial remains positive. Now for z = −1, the value of the
polynomial is negative for all values of the coefficients, as we can see
in eq. (63). We conclude that there must be two negative roots.

P (0) = Ks2
P (∞) = −Kd2
P (−1) = −4(K12 − K13 )2 (63)
The square root of the positive root gives the two solutions for κ.

κ √
tan = ± z0 (64)
2
The result is two sets of solutions (a, b, c, κ) and (−b, −a, −c, −κ).
The design procedure yields two unique algebraic solutions for
the three position synthesis of the RR chain. Due to the symme-
try of the principal axes, the second solution given by the values
(−b, −a, −c, −κ) corresponds to the coordinates of the RR chain as-
sociated with the other two axes of the Bennett linkage, as defined in
Equation (54) and can be seen in Table 8. Hence this procedure yields
two RR chains, which combine to form a Bennett linkage.
Solution 1 Solution 2
G( a, b, c, κ) = H(−b, −a, −c, −κ)
W( a, b, c, κ) = U(−b, −a, −c, −κ)
H( a, b, c, κ) = G(−b, −a, −c, −κ)
U( a, b, c, κ) = W(−b, −a, −c, −κ)

21
 x y z θ φ ψ
M1 0.0 0.0 0.0 0◦ 0◦ 0◦
M2 0.0 0.0 0.8 0◦ 0◦ 40◦
M3 1.11 0.66 0.05 18.8◦ -28.0◦ 67.2◦

Table 2: Three specified positions

Axis Line coordinates

G (0.36, 0.45, 0.81) , (0.26, 1.05, −0.70)

W1 (0.60, 0.36, 0.72) , (0.87, 0.83, −1.14)

H (0.60, −0.36, 0.72) , (0.87, −0.83, −1.14)

U1 (0.36, −0.45, 0.81) , (0.26, −1.05, −0.70)

Table 3: The joint axes of the Bennett linkage

9 Example
In Table 2 and Figure 8 we present three specified positions, given by
the distance (x, y, z) to the origin and the longitude, latitude and roll
orientation angles (θ, φ, ψ) of the moving frame.
We obtain the fixed joint lines G and H and moving joint lines
W and U1 . Table 3 presents their coordinates in the original frame.
1

Figure 9 shows the obtained Bennett linkage while passing by the three
specified positions.

10 Conclusions
This paper presents a new formulation of the solution for the three
position synthesis of a Bennett linkage. The procedure combines the
results of Tsai and Roth (1973) for the spatial RR chain with the
geometric properties of the cylindroid studied by Huang (1996). Using
the properties of the cylindroid, we were able to use the principal
axis frame of the cylindroid and a new set of design parameters to
simplify the original design problem from ten quadratic equations in
ten unknowns to four equations in four unknowns. We find algebraic

22
 Figure 8: The specified positions

expressions for the unknowns and were able to prove that there are
two unique solutions that combine to form a Bennett linkage.

11 Acknowledgments
Shawn, Haijun, Balsells fellowship, bla, bla, bla....

References
[Alhers and McCarthy,2000] Ahlers, S. G., and McCarthy, J. M.,
“The Clifford Algebra of Double Quaternions and the Optimization
of TS Robot Design”. Applications of Clifford Algebras in Com-
puter Science and Engineering, E. Bayro and G. Sobczyk, editor,
(in press).

[Baker,1998] Baker, J.E., “On the Motion Geometry of the Ben-

nett Linkage.” Proc. 8th Internat. Conf. on Engineering Computer
Graphics and Descriptive Geometry, Austin, Texas, USA, 1998, vol.
2, pp. 433-437.

23
Figure 9: The Bennett linkage

24
[Bennett, 1903] Bennett, G.T., A New Mechanism. Engineering,
1903, vol. 76, pp. 777-778.
[Bottema and Roth, 1990] Bottema, O., and Roth, B., Theoretical
Kinematics, North-Holland, 1979, (reprinted by Dover Publica-
tions, 1990).
[Etzel and McCarthy,1999] Etzel, K.R., and McCarthy, J.M., “Inter-
polation of Spatial Displacements Using the Clifford Algebra of E4”.
ASME Journal of Mechanical Design, vol. 121(1), pp. 39-44.
[Ge,1995] Ge, Q.J., and Kang, D., “Motion Interpolation with G2
Composite Bezier Motions”. ASME Journal of Mechanical Design,
vol. 116, pp. 749-755.
[Huang, 1992] Huang, C., “On the Finite Screw System of the Third
Order Associated with a Revolute-Revolute Chain” Journal of Me-
chanical Design, 1994, vol. 116, pp. 875-883.
[Huang, 1996] Huang,C., “The Cylindroid Associated with Finite Mo-
tions of the Bennett Mechanism,” Proceedings of the ASME Design
Engineering Technical Conferences, Irvine, CA, 1996.
[Hunt, 1978] Hunt, K.H., Kinematic Geometry of Mechanisms.
Clarendon Press, 1978.
[McCarthy, 1990] McCarthy, J. M., Introduction to Theoretical Kine-
matics. The MIT Press, 1990.
[McCarthy, 2000] McCarthy, J. M., Geometric Design of Linkages.
Springer-Verlag, 2000.
[McCarthy and Alhers,1999] McCarthy, J.M., and Ahlers, S., “Di-
mensional Synthesis of Robots Using a Double Quaternion Formula-
tion of the Workspace”. 9th. International Symposium of Robotics
Research ISRR’99, pp. 1-6.
[Park,1995] Park, F. C, “Distance Metrics on Rigid Body Motions
with Applications to Mechanism Design”. ASME J. of Mechanical
Design, vol. 117(1), pp. 48-54.
[Parkin, 1992] Parkin, I.A., “A Third Conformation with the Screw
Systems: Finite Twist Displacements fo a Directed Line and
Point”.Mechanisms and Machine Theory, 1992, vol. 27, pp. 177-
188.

25
 [Parkin, 1997] Parkin, I.A., “Finding the Principal Axes of Screw
Systems.” Proceedings of the ASME Design Engineering Technical
Conferences, Sacramento, CA, 1997.

[Suh, 1969] Suh, C.H., “On the Duality in the Existence of R-R Links
for Three Positions.” J. Eng. Ind. Trans. ASME, 1969, vol. 91B,
pp. 129-134.

[Suh and Radcliffe, 1978] Suh, C.H. and Radcliffe, C.W., Kinematics
and mechanisms design. John Wiley & Sons, 1978.

[Tsai and Roth, 1973] Tsai, L.W. and Roth, B., “A Note on the De-
sign of Revolute-Revolute Cranks”. Mechanisms and Machine The-
ory, 1973, Vol.8, pp 23-31.

[Tsai, 1972] Tsai, L.W., Design of open-loop chains for rigid body
guidance. Stanford University Ph.D.dissertation, 1972.

[Veldkamp, 1967] Veldkamp, G.R., “Canonical Systems and Instan-

taneous Invariants in Spatial Kinematics.” Journal of Mechanisms,
1967, Vol.3., pp. 329-388.

[Yu, 1981] Yu, H.C., “The Bennett Linkage, its Associated Tetrahe-
dron and the Hyperboloid of its Axes”. Mechanism and Machine
Theory, Vol. 16, pp. 105-114, 1981.

26

View publication stats