1

Inversion Symmetry of the Euclidean Group:

Theory and Application in Robot Kinematics
Yuanqing Wu, Harald Löwe, Marco Carricato, Zexiang Li

Abstract—Just as the three-dimensional (3D) Euclidean space
can be inverted through any of its points, the special Euclidean
group SE(3) admits an inversion symmetry through any of its
elements, and is known to be a symmetric space. In this paper,
we show that the symmetric submanifolds of SE(3) can be
systematically exploited to study the kinematics of a variety of
kinesiological and mechanical systems, and therefore have many
potential applications in robot kinematics. Unlike Lie subgroups
of SE(3), symmetric submanifolds inherit distinct geometric
e(·) . Consider a serial manipulator comprising revolute (R),
helical (Hp ; p denotes pitch) or prismatic (P) joints, with
linearly independent joint twists {ξ̂1 , . . . , ξ̂k }, ξˆi ∈ se(3). Its
direct kinematics map is given by the product of exponentials
(POE) formula [6]. The set of end-effector motions generated
by the serial manipulator is given by:
k
Y n  o
properties from inversion symmetry. They can be generated exp{ξˆi }sp , eθ1 ξ̂1 ···eθk ξ̂k θi ∈R,i=1,...,k (1)

by kinematic chains with symmetric joint twists. The main
contribution of this paper is: (i) to give a complete classification
of symmetric submanifolds of SE(3); (ii) to investigate their
geometric properties for robotics applications; and (iii) to develop
a generic method for synthesizing their kinematic chains.
Index Terms—Euclidean group, geodesic symmetry, symmetric
space, totally geodesic submanifold, Lie triple system (LTS),
Constant-Velocity (CV) coupling, kinesiology, parallel manipu-
lator, type synthesis.
I. I NTRODUCTION
The special Euclidean group SE(3) refers to the 6D Lie
group of proper rigid displacements of 3D Euclidean space.
It is Hervé [2,3] and Brockett [4] who initiated application of
SE(3) and its Lie subgroups (e.g. the special orthogonal group
SO(3); see [5]) in robotics (kinematics and dynamics [6]–[8],
estimation and control [9]–[11], etc.).
Recent advances in type synthesis of parallel manipulators
[12]–[18] can be attributed to the successful exploitation of the
Lie algebra se(3) of SE(3). Central to the synthesis problem is
the exponential map, denoted exp, which maps se(3), locally
diffeomorphically, into SE(3). In this paper, we shall adopt
the homogeneous matrix representation for SE(3) [6], so that
the exponential map is identified with the matrix exponential
This work was supported by RGC Grant No.616509, No.615610 and
No.615212, Shenzhen Municipal Science and Technology Planning Program
i=1
which coincides with a kD submanifold of SE(3) in an open
neighborhood of the identity I ∈ SE(3). We shall refer to
(1) as a POE-submanifold, and we say it is the motion type
[18] of the serial manipulator. In particular, when the linear
span of ξˆi ’s, denoted by {ξ̂1 , . . . , ξ̂k }sp , is a Lie subalgebra g
Qk
of se(3), i=1 exp{ξˆi }sp is locally an open submanifold of
the corresponding (connected) Lie subgroup G ([19, pp. 299,
Lemma 9.2.6]). We simply say the motion type is G.
The element-wise product of two Lie subgroups [20,21] can
also be locally identified with a POE-submanifold [18], but is
in general not a Lie subgroup. Carricato et al. showed that
the tangent spaces of such submanifolds are all mutually con-
gruent, thus defining what is called a persistent screw system
of the end-effector [22,23]. When a persistent screw system
exists, the corresponding submanifold may be generated by
the envelop of a tangent space smoothly moving in SE(3)
like a rigid body [22]–[26]. In most previous studies on type
synthesis of parallel manipulators, the motion type of a parallel
manipulator can be identified with a POE-submanifold (see for
example, category I/II submanifolds [18], virtual chain [27],
displacement manifold [28]).
Since exp : se(3) → SE(3) is a local diffeomorphism of
an open neighborhood of the origin 0̂ ∈ se(3) onto an open
neighborhood of the identity I ∈ SE(3), the exponential image
of the kD vector subspace {ξ̂1 , . . . , ξ̂k }sp ⊂ se(3):
for Basic Research No. JCYJ20120618150931822, and by National Natural n  o
Science Foundation Grant No.51375413. exp{ξ̂1 ,...,ξ̂k }sp , eθ1 ξ̂1 +···+θk ξ̂k θi ∈R,i=1,...,k

This paper was partly presented at the 2013 IEEE International Conference
on Robotics and Automation (ICRA2013) [1]. The current manuscript was
originally submitted on September 21, 2012, and then, in a revised form, on
September 24, 2014.
Yuanqing Wu (corresponding author) was and Zexiang Li is with Electronic
and Computer Engineering, Hong Kong University of Science and Technology
(including Shenzhen Research Institute), Hong Kong, P.R. China (e-mail:
troy.woo@gmail.com yuanqing.wu@unibo.it, eezxli@ust.hk). Dr.Wu is cur-
rently at the Department of Industrial Engineering (DIN) of the University
of Bologna, financially supported by the PRIN grant no. 20124SMZ88 of the
Italian Ministry of Education, Universities and Research.
Harald Löwe is with Institut Computational Mathematics, Technical Uni-
versity of Braunschweig, Deutschland (e-mail: h.loewe@tu-bs.de).
Marco Carricato is with Department of Industrial Engineering (DIN),
University of Bologna, Italy (e-mail: marco.carricato@unibo.it).
is locally a kD submanifold of SE(3), which we refer to as
an Exp-submanifold (Exp for “exponential”) [1]. It is clear
from the Baker-Campbell-Hausdorff formula ([19, pp. 57,
Prop. 3.4.4]) that in general Exp-submanifolds are not POE-
submanifolds (except for example, when {ξ̂1 , . . . , ξ̂k }sp is
a Lie subalgebra g ⊂ se(3), exp{ξ̂1 , . . . , ξ̂k }sp is a subset
and contains an open neighborhood of the corresponding
connected Lie subgroup G ([19, pp. 299, Lemma 9.2.6])).
Therefore, in general, Exp-submanifolds can only be generated
by closed-loop manipulators [1,29].

(a)
(b)
Fig. 1. (a) Listing’s law of eye saccade: wp , wg and wb denote the primary
or initial direction (perpendicular to the Listing’s plane), the gaze direction
and their angle bisector (perpendicular to the velocity plane). (b) reflected
tripod [30].
Exp-submanifolds have been used in robotics and biome-
chanics without being explicitly recognized or systematically
studied. Bonev et al. [31] analyzed the rotational motion of
several constant-velocity (CV) couplings using a modified
Euler angle parametrization (tilt and torsion angles, [32]),
which is equivalent to the following parametrization for SO(3)
[29]:
(θ1 , θ2 , σ) 7→ eθ1 x̂+θ2 ŷ eσẑ ∈ SO(3) (2)
where {x, y, z} denotes the canonical basis for R3 and ŵ
defines a 3 × 3 skew-symmetric matrix such that ŵv =
w × v, ∀w, v ∈ R3 . Bonev observed that the torsion angle
σ for a CV coupling is always zero, leading to a motion type
exp{x̂, ŷ}sp , a 2D Exp-submanifold [1,29]. In kinesiology and
biomechanics, the same Exp-submanifold (under quaternion
representation) is identified to be the motion type of human
eye saccade [33,34], and is known to obey Listing’s half-angle
law: as the gaze direction rotates away from the initial primary
direction (normal of the Listing’s plane), the instantaneous
velocity plane rotates away from the Listing’s plane along the
same rotation axis but half in magnitude (see Fig. 1(a)).
Another example of Exp-submanifolds arises in three
degree-of-freedom (DoF) parallel-architecture CV couplings
for intersecting shafts [35,36]. The instantaneous velocity of
such a coupling always lies in the bisecting plane1 , demanding
the joint twists in each connecting chain to be mirror sym-
metric about the bisecting plane in all configurations. Typical
1 The
yellow plane shown in Fig. 1(b): (i) it is perpendicular to the plane of
input and output velocity win and wout ; and (ii) it bisects the complement
of the working angle formed by win and wout .
2
Fig. 2. Graphical representation of the canonical basis of se(3) and notation
for twists with different pitch value p.
CV connecting chains include RSR (S for spherical joint)
chain and RER (E for planar gliding joint) chain [35]. An
in-parallel assembly of three RSR chains that are mirror
symmetric about a common bisecting plane results in the
“reflected tripod” [30,31] (see Fig. 1(b)), which found appli-
cations in robotic wrists [37] and hyperredundant robots [38].
We showed that its motion type is the 3D Exp-submanifold
exp{ê3 , ê4 , ê5 }sp [29], where {êi }6i=1 denotes the canonical
basis of se(3) (also see Fig. 2):
     
0̂ x 0̂ y 0̂ z
ê1 , , ê2 , , ê3 , ,
0T 0 0T 0 0T 0
     
x̂ 0 ŷ 0 ẑ 0
ê4 , , ê5 , , ê6 ,
0T 0 0T 0 0T 0
If each 5-DoF RSR chain of the reflected tripod is reduced
to a 4-DoF mirror symmetric UU (U for Cardan or universal
joint) chain with the U joints of all chains on each side
of the mirror sharing the same center of rotation [36], we
have the UNITRU coupling [39] (see Fig. 9(a)), which has
the motion type of a 2D surface in exp{ê3 , ê4 , ê5 }sp [29].
This 2-DoF parallel manipulator is later used by Rosheim
in the Omni-Wrist III [40] and reportedly to mimic human
shoulder complex movement in terms of an extraordinary
orientation range [41]. So far, no further results on general
Exp-submanifolds of SE(3) are available in the literature.
It turns out that both the Listing’s law of eye saccade
and mirror symmetry of CV couplings can be attributed to
inversion (or geodesic) symmetry, a class of diffeomorphism
maps associated with symmetric spaces [42]. In fact, SE(3)
is an (affine) symmetric space with the inversion symmetry at
each point g defined by [43]:
∀g ∈ SE(3) ⇒ Sg (h) , gh−1 g ∈ SE(3), ∀h ∈ SE(3) (3)
and that both exp{x̂, ŷ}sp (or equivalently exp{ê4 , ê5 }sp ) and
exp{ê3 , ê4 , ê5 }sp can be extended via inversions to a unique
symmetric space M. Such a submanifold M is closed under
inversions:
∀g, h ∈ M ⇒ gh−1 g ∈ M (4)
and will be referred to as symmetric submanifolds of SE(3)
(they are more often referred to as symmetric subspaces in
mathematical literature [42,43]). Note that all connected Lie
 3

subgroups of SE(3) are automatically closed under inversions, A. SE(3) as a symmetric space
and are trivial symmetric submanifolds. We exclude Lie sub- We associate to each g ∈ SE(3) an inversion symmetry Sg
groups from our study, since their symmetric space structure as defined in (3). Sg is involutive, i.e. Sg ◦ Sg = idSE(3) , and
can be studied in a similar way to that for SE(3). reverses the exponential map exp gξ̂ , geξ̂ for any tangent
In this paper, we will show: (i) there are seven conjugacy vector gξ̂ ∈ Tg SE(3) at g ∈ SE(3):
classes of symmetric submanifolds of SE(3), all of which can
be locally represented by exp m, with m being a Lie triple Sg (exp gξ̂) = g(geξ̂ )−1 g = ge−ξ̂ = exp(−gξ̂).
(sub)system (LTS) of se(3) [42,43], i.e. a vector subspace of
se(3) that is closed under double Lie brackets: SE(3) equipped with the inversion symmetry is called a sym-
metric space. A quadratic representation Q(g) of g ∈ SE(3)
∀ξ1 , ξ2 , ξ3 ∈ m ⇒ [[ξ1 , ξ2 ], ξ3 ] ∈ m is a diffeomorphism defined by:
or simply [[m, m], m] ⊂ m, where the Lie bracket is defined Q(g) , Sg ◦ SI : SE(3) → SE(3), Q(g)(h) = ghg
by [ξ̂1 , ξ̂2 ] , ξ̂1 ξ̂2 − ξ̂2 ξ̂1 , ∀ξ̂1 , ξ̂2 ∈ se(3); (ii) these sym- The group generated by {Q(g)|g ∈ SE(3)} under composition
metric submanifolds can be systematically studied to derive of maps is called the group of displacements ([43, pp. 64]) of
several common geometric properties, which include Listing’s SE(3) and shall be denoted by G̃. Since SE(3) is connected,
law and mirror symmetry of CV couplings as special cases G̃ acts transitively on SE(3) ([43, pp. 91, Th. 3.1 a)]). See
(for exp{ê4 , ê5 }sp and exp{ê3 , ê4 , ê5 }sp respectively); and Appendix A for more details.
(iii) we can develop a generic method for synthesizing their For any ξ̂ ∈ se(3), we denote the corresponding right- and
kinematic chains.
left-invariant vector fields on SE(3) by ξ̂ r and ξ̂ l respectively,
The only revelation that comes close to our discovery i.e.: (
is Selig’s attempt to study full cycle mobility using totally ξ̂ r (g) , ξ̂g
geodesic submanifolds of SE(3) (see [44, Ch. 15.2]). Only Lie ∀g ∈ SE(3)
subgroups are considered in [44], but both Lie subgroups and ξ̂ l (g) , gξ̂
symmetric submanifolds are totally geodesic (for the concept There are two special classes of vector fields relevant to the
of totally geodesic submanifolds of a symmetric space, see inversion symmetry of SE(3): the (−)-derivations D− and
[42,45]; the totally geodesic submanifolds of a symmetric (+)-derivations D+ ([43, pp. 81]). Every twist ξ̂ ∈ se(3)
space are exactly the symmetric submanifolds [43, pp. 121, defines a (−)-derivation ξ̂− :
Coro., Lemma 1.3]). 1 r 1
Recently, during the review process of this manuscript, Selig ξ̂− (g) , (ξ̂ + ξ̂ l )(g) = (ξ̂g + gξ̂), ∀g ∈ SE(3) (5)
2 2
[46] submitted and published a paper that, among other things,
presents a classification of the LTSs of se(3). However, we and a (+)-derivation ξ̂+ :
published a systematic classification of the LTSs of se(3) in 1 r 1
an earlier conference paper [1], which this manuscript relies ξ̂+ (g) , (ξ̂ − ξ̂ l )(g) = (ξ̂g − gξ̂), ∀g ∈ SE(3) (6)
2 2
upon and extends. In comparison to our previous and current In this case, the R-vector spaces D− and D+ can be identi-
results, Selig’s classification has the following differences: it fied with se(3) respectively. The integral curves of ξ̂− and
includes several Lie subalgebras into the classification of LTSs t
ξ̂+ passing through g ∈ SE(3) are given by e 2 ξ̂ ge 2 ξ̂ =
t

(all Lie subalgebras are, in fact, trivial LTSs), but omits the t t
Q(exp 2t ξ̂)(g) and e 2 ξ̂ ge− 2 ξ̂ = C(exp 2t ξ̂)(g) respectively
5D LTS {ê1 , ê2 , ê3 , ê4 , ê5 }sp , because its does not meet the
(C(h) denotes conjugation by h ∈ SE(3)), since:
requirements of the application on which Ref. [46] focuses, 
d t

namely motion planning. 
 Q(exp ξ̂)(g)

 = ξ̂− (g)
 dt 2
This paper is organized as follows. Section II gives a 
t=0
 ∀g ∈ SE(3) (7)
brief review of the symmetric space SE(3) and a systematic  d t 
classification of its symmetric submanifolds. Section III sum-

 C(exp ξ̂)(g) = ξ̂+ (g)
dt 2 t=0
marizes a list of geometric properties common to all symmetric
submanifolds, and shows how they can be applied to the study For any ξ̂, ζ̂ ∈ se(3), from the fact that (see Appendix B
of eye saccade and CV coupling motion. Section IV proposes for proof):
a systematic approach for synthesizing kinematic chains for [ξ̂ r , ζ̂ r ] = −[ξ̂, ζ̂]r , [ξ̂ l , ζ̂ l ] = [ξ̂, ζ̂]l , [ξ̂ r , ζ̂ l ] = 0 (8)
symmetric submanifolds. Finally, Section V concludes our
work. we have:
1

 [ξ̂− , ζ̂− ] = − [ξ̂, ζ̂]+
 
2  [D− , D− ] ⊂ D+

II. I NVERSION S YMMETRY OF SE(3) AND S YMMETRIC 
 
1

S UBMANIFOLDS [ξ̂+ , ζ̂+ ] = − [ξ̂, ζ̂]+ ⇒ [D+ , D+ ] ⊂ D+ (9)
 2 
In this section, we first give a brief introduction to the [D+ , D− ] ⊂ D−
 
 [ξ̂+ , ζ̂− ] = − 1 [ξ̂, ζ̂]−



symmetric space theory of SE(3) following the elementary 2
treatment of Loos [43]. Then we give a systematic classifi-
Therefore D+ is a Lie algebra and D− is a LTS since:
cation of symmetric submanifolds or SE(3). We assume the
readers are familiar with basic Lie group theory of SE(3) [6]. [[D− , D− ], D− ] ⊂ [D+ , D− ] ⊂ D−
 4

TABLE I
C ONJUGACY CLASSES OF LTS S OF se(3) ( EXCLUDING ALL L IE SUBALGEBRAS OF se(3), WHICH ARE TRIVIAL LTS S ).

dim LTS m (normal form) screw sys. [30] hm = [m, m] gm = hm + m isotropy group
m2A , {ê3 , ê4 }sp {ê2 , ê3 , ê4 }sp
2nd special 2-sys. {ê2 }sp exp{ê1 , ê2 }sp
2 mp2A , {ê3 , ê4 + pê1 }sp {ê2 , ê3 , ê4 + pê1 }sp
m2B , {ê4 , ê5 }sp 1st special 2-sys. {ê6 }sp {ê4 , ê5 , ê6 }sp exp{ê6 }sp
m3A , {ê1 , ê3 , ê4 }sp 10th special 3-sys. {ê2 }sp {ê1 , ê2 , ê3 , ê4 }sp exp{ê1 , ê2 }sp
3
m3B , {ê3 , ê4 , ê5 }sp 4th special 3-sys. {ê1 , ê2 , ê6 }sp exp{ê1 , ê2 , ê6 }sp
4 m4 , {ê1 , ê2 , ê4 , ê5 }sp 5th special 4-sys. {ê3 , ê6 }sp se(3) exp{ê3 , ê6 }sp
5 m5 , {ê1 , ê2 , ê3 , ê4 , ê5 }sp special 5-sys. {ê1 , ê2 , ê3 , ê6 }sp exp{ê1 , ê2 , ê3 , ê6 }sp

Moreover, D+ ⊕ D− is a Lie algebra, and is identified with
the Lie algebra of G̃ ([43, pp. 91, Th. 3.1 d)]). In reference to
the identification of se(3) with D− via (5), we also say that
se(3) is a LTS.
B. Symmetric submanifolds of SE(3)
A symmetric submanifold (containing the identity I) of
SE(3) is a submanifold M which is closed under inversion
symmetry (eq. (4)). Since for ∀g0 , g, h ∈ SE(3):
C(g0 )(gh−1 g) = g0 (gh−1 g)g0−1
= g0 gg0−1 g0 h−1 g0−1 g0 gg0−1
= C(g0 )(g)(C(g0 )(h))−1 C(g0 )(g)
the conjugation C(g0 )(M) of a symmetric submanifold M
is still a symmetric submanifold. Therefore, the systematic
classification of symmetric submanifolds reduces to that of
their conjugacy classes.
Let m denote the tangent space TI M of M at I. m is
identified with the (−)-derivations D− (M) as in (5), and is a
LTS (subsystem) of se(3) ([43, pp. 121]). From the definition
of Lie group and Lie algebra ([6, Appendix A]), all Lie
subgroups of SE(3) are trivial symmetric submanifolds, with
their corresponding Lie subalgebras being trivial LTSs. Their
symmetric space structure can be understood in the same way
as that of SE(3). By the definition of LTS and Jacobi identity
([6, Appendix A]), one easily verifies that both hm , [m, m]
and gm , hm +m are Lie subalgebras of se(3). hm is identified
with D+ (M) as in (6). A relation similar to (9) holds for m
and hm :
[m, m] = hm , [hm , hm ] ⊂ hm , [hm , m] ⊂ m
The following theorem establishes the fundamental relation
between a symmetric submanifold M and its tangent space at
identity.
Theorem 1. A (connected) symmetric submanifold M ⊂
SE(3) (containing I) is generated (via inversion symmetry)
by any open neighborhood (containing I) of exp m, where the
tangent space m = TI M is a LTS of se(3):
M = {ghg ∈ SE(3)|∀g, h ∈ exp m}
Proof. See Appendix C.
According to Theorem 1, there is a one-to-one correspon-
dence between LTSs of se(3) and symmetric submanifolds of
SE(3) (containing I). Therefore, the systematic classification
of symmetric submanifolds of SE(3) up to conjugation is
equivalent to that of conjugacy classes of LTSs of se(3).
Starting from a screw system of se(3) (see for example [30,
Ch. 12] or [44, Ch. 8]), we can determine if it is a LTS by
verifying closure under double Lie brackets for an arbitrarily
chosen basis (the complete computation cannot be provided
here due to space limitations and will be presented in [47].).
A total of seven conjugacy classes of LTSs are found and
listed in Table I. The subscripts mA and nB for the LTSs in
the second column denote an mD LTS containing one rota-
tional/helical DoF and an nD LTS containing two rotational
DoFs respectively. The same subscript convention is used for
the corresponding symmetric submanifolds. Essentially the
same classification was reported in [1]. The only difference
is that the 2D LTS {ê3 , ê4 + pê1 }sp , though given here, was
omitted in [1], since it can be essentially studied in the same
way as {ê3 , ê4 }sp .
The screw systems corresponding to the LTSs (normal form)
are depicted in Fig. 3. A generic member of each conjugacy
class m is of the form Adg m, g ∈ SE(3), where Adg is the
Adjoint transformation:
Adg ξ̂ , gξ̂g−1 , ξ̂ ∈ se(3)
It is also clear from Theorem 1 that the local Exp-
submanifold exp m we discussed in Section I is actually an
open neighborhood (about I) of a unique symmetric subman-
ifold M ⊂ SE(3) when m is a LTS of se(3). We shall say
M is generated (via inversion symmetry) by exp m, which
(10) we denote by2 M = hexp mi. A generic member of each
conjugacy class M is given by:
C(g)(M) = hexp Adg mi.
It is proved that exp : g → G of a Lie subalgebra g into
its corresponding Lie subgroup G may not be surjective (sur-
jectivity fails for the conjugacy class of {ê2 , ê3 , ê4 + pê1 }sp )
[48]. The following proposition shows exp : m → M may not
be surjective either.
(11) 2 For convenience, we shall also denote a (connected) Lie subgroup G ⊂
SE(3) with lie subalgebra g ⊂ se(3) by hexp gi, since Lie subgroups are
trivial symmetric submanifolds.
 5

(a) mp2A (m2A if p = 0) (b) m2B (c) m3A

(d) m3B (e) m4 (p arbitrary) (f) m5

Fig. 3. Screw systems of Lie triple systems. The basis screws of each LTS are denoted by green arrows, and a generic screw in each LTS is denoted by
a white arrow. Except for mp2A , the pitch value p of the generic screw takes an arbitrary finite value. The characteristic (char.) plane of mp2A (m2A ), m2B ,
m3A , m3B is defined to be the plane containing all axes of their screw systems. c(·) and s(·) denote cos(·) and sin(·) respectively.

Proposition 1. exp : m → M is surjective for all LTSs except in an open neighborhood of g ∈ GM .

mp2A = {ê3 , ê4 + pê1 }sp . Every point of Mp2A can be reached
Proof. See Appendix E.
from exp mp2A by one inversion.
Proof. See Appendix D. First, Prop.2 a) is indeed a generalization to Listing’s law of
human eye saccade for the special case of M2B = hexp m2B i,
III. G EOMETRIC P ROPERTIES OF S YMMETRIC and may therefore be referred to as the half-angle property.
S UBMANIFOLDS It is a direct consequence of the inversion symmetry of the
In Section II-B, we use a LTS m ⊂ se(3) (identified with symmetric submanifolds (see the proof of Prop.2 a)). Other
D− (M)) to recover the corresponding symmetric submanifold than correctly predicting human eye saccade behavior, the half-
M = hexp mi ⊂ SE(3). A list of useful properties common to angle property also prescribes the location of the bisecting
all symmetric submanifolds is given as follows. plane of a CV coupling, and is therefore a convenient tool for
position and constraint analysis of CV connecting chains [36].
Proposition 2. The following statements are true for all LTSs Second, Prop.2 b) can be referred to as the conjugation in-
m and symmetric submanifolds M = hexp mi: variance property of the corresponding LTS m and symmetric
a) The right translation (back to the identity I) of the tangent submanifold M = hexp mi. Physically, it corresponds to the
space at g = eξ̂ , ξ̂ ∈ m is given by: fact that the kinematics of the underlying symmetric subman-
ifold is uniform in the directions specified by the isotropy
(Tg M)g−1 = Ad 1 m = Adeξ̂/2 m
g2 group HM . In the case of a 2-DoF (M2B ) CV coupling, this
In the case of g = eξ̂1 eξ̂2 eξ̂1 (for mp2A ): uniformity ensures uniform (or constant) velocity transmission
between two intersecting shafts. The conjugation invariance
(Tg M)g−1 = Adeξ̂1 eξ̂2 /2 m of M3B , for example, can also be exploited to synthesize its
b) m is Adjoint invariant by elements of the isotropy group parallel manipulators using identical and conjugate kinematic
HM , hexp hm i with hm , [m, m] chains. From a more mathematical viewpoint, conjugation
invariance implies that symmetric submanifolds should be
∀h ∈ HM ⇒ Adh m ≡ m quotient (or homogeneous) spaces of certain transitive Lie
M is conjugation invariant by elements of HM transformation group action, which we briefly summarized in
Appendix A.
∀h ∈ HM ⇒ C(h)(M) ≡ M Third, in Prop.2 c), the completion algebra gm is very
c) gm = m + hm is the completion algebra of m (i.e. useful for type synthesis of kinematic chains for symmetric
the smallest Lie subalgebra containing m), and GM , submanifolds (see Section IV). In the case hm ∩ m = {0̂},
hexp gm i is the completion group of M (i.e. the smallest the local parametrization eg xpI for an open neighborhood of I
connected Lie subgroup containing M); all LTSs except in GM is an immediate generalization of the tilt and torsion
m5 = {ê1 , ê2 , ê3 , ê4 , ê5 }sp satisfy hm ∩m = {0̂} and admit angle parametrization for SO(3) [32].
a local parametrization for GM : We give two full computation examples with applications
for a better understanding of the aforesaid geometric properties
xpg : m × hm → GM
eg
∀g ∈ GM ⇒ of symmetric submanifolds of SE(3).
(ξ̂, ζ̂) 7→ geξ̂ eζ̂
 6

Example 1 (M2B ). It is obvious from: Then we have:

( k
[[ê4 , ê5 ], ê4 ] = [ê6 , ê4 ] = ê5 ∀g ∈ HM ⇒ M = C(g)(M) ⊂local C(g)(
Y
exp{ξ̂i }sp )
[[ê4 , ê5 ], ê5 ] = [ê6 , ê5 ] = −ê4 i=1
Qk
that m2B = {ê4 , ê5 }sp ⊂ se(3) (or {x̂, ŷ}sp ⊂ so(3), with where the conjugate submanifold C(g)( i=1 exp{ξ̂i }sp ) still
so(3) the Lie algebra of SO(3)) is a LTS. Since exp is contains M and is generated by an identical but rigidly
surjective for m2B by Prop.1, M2B = exp m2B (see also [34]): displaced chain (Adg ξ̂1 , . . . , Adg ξ̂k ). A parallel manipulator
comprising two or more such chains may generate M with
M2B = {eŵ |w ∈ {x, y}sp } appropriate velocity or force matching conditions [18, Prop.
6].
It can be verified by straightforward computation (preferably
xpI : {x̂, ŷ}sp × {ẑ}sp → SO(3) is simply a
Finally, eg
using unit quaternions) that M2B is closed under inversion
slight variation of the tilt and torsion angle parametrization
symmetry:
[32] given earlier in (2). Its parametrization singularity may
∀w1 , w2 ∈ {x, y}sp ⇒ eŵ1 e−ŵ2 eŵ1 ∈ M2B be easily investigated using half-angle property: the spatial
Jacobian of eŵ eσẑ , ŵ = θ1 x̂ + θ2 ŷ is given by:
A computational verification of the half-angle property for hR
1 R1 i
M2B = hexp m2B i can be given as follows: according to [49, ( 0 euŵ(t) du)x ( 0 euŵ(t) du)y eŵ z ∈ R3×3
eq. (20)], where the first two columns always span the half tilted xy-
 
d ŵ(t) −ŵ(t)
Z 1 plane eŵ/2 {x, y}sp . Therefore,
e e = ˙
euŵ(t) ŵ(t)e −uŵ(t)
du nR o
dt 0
1 R1
( 0 euŵ(t) du)x, ( 0 euŵ(t) du)y, eŵ z
Z 1 ∧ (12) sp
uŵ(t) ŵ/2 ŵ/2
= e ẇ(t)du =e {x, y, e z}sp
0
and parametrization singularity is reached when eŵ/2 z lies in
for ŵ ∈ m2B = {x̂, ŷ}sp (which is equivalent to {ê4 , ê5 }sp ).
the xy-plane, or when the tilt angle reaches 180 degrees.
Since both w and ẇ lie in the xy-plane, ẇ(t) can be written
as λ1 w + λ2 w⊥ for some real numbers λ1 , λ2 and w⊥ = ẑw. Example 2 (M3B ). It is obvious from:
Then (12) gives: 

 [[ê3 , ê4 ], ê3 ] = [ê2 , ê3 ] = 0̂
Z 1 ∧ 
[[ê3 , ê4 ], ê4 ] = [ê2 , ê4 ] = −ê3
 
d ŵ(t) −ŵ(t)


uŵ(t) ⊥ 
e e = e (λ1 w + λ2 w )du 

dt 0  [[ê3 , ê4 ], ê5 ] = [ê2 , ê5 ] = 0̂



Z 1 ∧ 
 [[ê3 , ê5 ], ê3 ] = [−ê1 , ê3 ] = 0̂


⊥
= (λ1 w + λ2 (cukwk w + sukwk z))du

0 (13) [[ê3 , ê5 ], ê4 ] = [−ê1 , ê4 ] = 0̂
    ∧ 
ŵ(t)/2 kwk ⊥ 
[[ê3 , ê5 ], ê5 ] = [−ê1 , ê5 ] = −ê3
= e λ1 w + λ2 sinc 2 w 



[[ê 4 , ê5 ], ê3 ] = [ê6 , ê3 ] = 0̂
 ∧ 


∈ eŵ(t)/2 {x, y}sp


[[ê4 , ê5 ], ê4 ] = [ê6 , ê4 ] = ê5





which verifies that (Teŵ M2B )e−ŵ = Adeŵ/2 m2B . [[ê4 , ê5 ], ê5 ] = [ê6 , ê5 ] = −ê4


M2B is conjugation invariant by elements of the isotropy that m3B = {ê3 , ê4 , ê5 }sp is a LTS. Since exp is surjective for
group HM2B of M2B , which is given by hexp hm2B i = m3B by Prop.1, M3B = exp m3B . A typical element g ∈ M3B
hexp{ẑ}sp i = SO(2): is given by:
   2ŵ
2λ sinc(kwk)eŵ z

∀φ ∈ R ⇒ eφẑ · hexp{x̂, ŷ}sp i · e−φẑ g = exp T
2ŵ 2λz
=
e
= hexp{cφ x̂ + sφ ŷ, −sφ x̂ + cφ ŷ}sp i 0 0 0T 1
= hexp{x̂, ŷ}sp i and
"   #
  kwk
A direct consequence of the conjugation invariance is that 1 ŵ λz eŵ λ sinc 2 eŵ/2 z
g = exp T
2 =
hexp m2B i may be generated by a parallel manipulator with 0 0 0T 1
identical and axially symmetric kinematic chains [50]. To where w ∈ {x̂, ŷ}sp and λ ∈ R.
see this, if a submanifold M with conjugation invariance It can be verified by straightforward computation that:
prescribed by a Lie Qksubgroup HM is locally contained in a 
d
 
2(λ1 ŵ + λ2 sinc (kwk) ŵ⊥ ) λ0 z − 21

POE-submanifold i=1 exp{ξ̂i }sp generated by a kinematic −1
g g =g
1
2 g
chain (ξ̂1 , . . . , ξ̂k ) (generation of symmetric submanifolds dt 0T 0
using kinematic chains will be systematically studied in Sec. for some λ0 ∈ R, where one needs to use both (13) and the
IV), following similar equation:
k  
M ⊂local
Y
exp{ξ̂i }sp d 2ŵ −2ŵ  ŵ(t)
∧
e e = 2e λ1 w + λ2 sinc (kwk) w⊥
i=1 dt
 7

and therefore (Tg M3B )g−1 = Ad 12 m3B , ∀g ∈ M3B . The if:

g (
half-angle property immediately leads to a closed-form direct ξ̂ + = ξ̂ + ζ̂,
kinematic solutions for the reflected tripod [31]. , ξ̂ ∈ m, ζ̂ ∈ hm (15)
ξ̂ − = ξ̂ − ζ̂
M3B is conjugation invariant by elements of the
isotropy group HM3B , which is given by hexp hm3B i = in which case (ξ̂ + , ξ̂ − ) is referred to as a symmetric twist
hexp{ê1 , ê2 , ê6 }sp i = SE(2): given any g ∈ SE(2) of the pair (SP). The decomposition (15) is unique except for m5
form  θẑ  where hm5 ∩ m5 6= {0̂}. A SC formed by k nesting SPs
e λ1 x + λ2 y
g= (ξ̂i+ , ξ̂i− ), ξ̂i± , ξ̂i ± ζ̂i , ξ̂i ∈ m, ζ̂i ∈ hm , i = 1, . . . , k locally
0T 1
generates M = hexp mi if:
we have:
 {ξ̂1 , . . . , ξ̂k }sp = m
 Adg ê3 = ê3

Adg ê4 = (λ1 sθ − λ2 cθ )ê3 + cθ ê4 + sθ ê5 Condition (15) is equivalent to

Adg ê5 = (λ1 cθ + λ2 sθ )ê3 − sθ ê4 + cθ ê5

{ξ̂1± , . . . , ξ̂k± }sp ⊕ hm = gm (16)
and therefore Adg m3B ≡ m3B , ∀g ∈ SE(2) and C(g)(M3B )
≡ M3B , ∀g ∈ SE(2). Using the same argument as in if hm ∩ m = {0̂}.
Example 1, we see that M3B may be generated by identical Proof. See Appendix F.
kinematic chains with arbitrary planar displacement (see the
M3B modules of The BROMMI hyperredundant robotic arm We emphasize that in the above proposition, although both
[38]). {ξ̂i+ }ki=1 and {ξ̂i− }ki=1 are linearly independent sets of twists,
xpI : {ê3 , ê4 , ê5 }sp × {ê1 , ê2 , ê6 }sp → SE(3)
Finally, eg twists of the SC (ξ̂1+ , . . . , ξ̂k+ ; ξ̂k− , . . . , ξ̂1− ) may become lin-
gives a local parametrization of SE(3). early dependent.
Corollary 1. Given a SP (ξ̂ + , ξ̂ − ) of a LTS m ⊂ se(3) and
IV. G ENERATING S YMMETRIC S UBMANIFOLDS U SING + −
a twist η̂ ∈ m, the pair (ξˆ0 , ξˆ0 ) defined by:
S YMMETRIC K INEMATIC C HAINS
( +
A. Symmetric chains of symmetric submanifolds ξˆ0 , Ade+η̂ ξ̂ +
− (17)
Consider a kD symmetric submanifold M = hexp mi ξˆ0 , Ade−η̂ ξ̂ −
and a basis of unit twists {ξ̂1 , . . . , ξ̂k } ⊂ m for the
LTS m. In general,
Qk M cannot be generated by the POE- is also a SP of m. In particular, (Ade+η̂ ξ̂, Ade−η̂ ξ̂), ξ̂ ∈ m is
submanifold i=1 exp{ ξ̂i }sp since it is not closed under a SP of m.
group product. In view of the inversion symmetry, M can
be locally generated by a (2k − 1)-DoF symmetric twist Proof. See Appendix G.
chain (SC) (ξ̂1 , . . . , ξ̂k−1 , ξ̂k , ξ̂k−1 , . . . , ξ̂1 ) with symmetric
joint variables: B. Symmetry type of symmetric chains
θ1 ξ̂1 θk−1 ξ̂k−1 θk ξ̂k θk−1 ξ̂k−1 θ1 ξ̂1
e ···e ·e ·e ···e = We shall use both the geometric condition (17) and the
Q(e θ1 ξ̂1
) ◦ · · · ◦ Q(e θk−1 ξ̂k−1 θk ξ̂k
)(e )∈M algebraic condition (15) to study the particular symmetry type
of SPs and SCs for each LTS.
θi ∈ R, i = 1, . . . , k 1) m2B : Consider the conjugacy class of 2D LTS m2B =
We emphasize that the motion generated by a SC with arbitrary {ê4 , ê5 }sp as shown in Fig. 3(b). It consists of a pencil of zero-
joint variables: pitch twists in the char. plane (Hunt’s 1st special 2-system,
+ + − − [30]). A SP (ξ̂ + ; ξ̂ − ) = (Ade+ψη̂ ξ̂; Ade−ψη̂ ξ̂), ξ̂, η̂ ∈ m2B is
eθ1 ξ̂1 · · ·eθk−1 ξ̂k−1 · eθk ξ̂k · eθk−1 ξ̂k−1 · · · eθ1 ξ̂1 generated by a pair of rotational displacements (+ψ, −ψ) of
θi± ∈ R, i = 1, . . . , k − 1 ξ̂ about the unit twist η̂. Pictorially, ξ̂ + and ξ̂ − are mirror
symmetric about the char. plane of m2B (see Fig. 4 left).
is in general not contained in M (but in GM ) unless the joint
Equivalently, the SP is given by the algebraic condition:
variables are also symmetric:
(
θi+ ≡ θi− , i = 1, . . . , k − 1. (14) ξ̂ + = cψ ξ̂ + sψ ê6
ξ̂ − = cψ ξ̂ − sψ ê6
In this case, we say the SC undergoes a symmetric movement.
The following proposition shows that the joint twists in a SC where ξ̂ ∈ m2B and ê6 ∈ hm2B = {ê6 }sp . Since gm2B =
can take general values in the completion algebra gm = hm + {ê4 , ê5 , ê6 }sp , members of the m2B -SC can be arbitrary twists
m. not equal to scalar multiples of ê6 ; a m2B -SC consists of two
Proposition 3. Given a kD LTS m of a symmetric submanifold nesting SPs:
M = hexp mi, a pair of unit twists (ξ̂ + , ξ̂ − ) of gm satisfies: ( (
ξ̂1+ = ξ̂1 + ζ̂1 ξ̂2+ = ξ̂2 + ζ̂2
+ −
∀g ∈ M, θ ∈ R ⇒ eθξ̂ geθξ̂ ∈ M ξ̂1− = ξ̂1 − ζ̂1 ξ̂2− = ξ̂2 − ζ̂2
 8

Fig. 4. Symmetric twist pairs of m2B . On the left: geometric condition, and
on the right: algebraic condition.

Fig. 5. Symmetric twist pairs of m3B . On the left: geometric condition, and
where ξ̂1 , ξ̂2 ∈ m2B = {ê4 , ê5 }sp and ζ̂1 , ζ̂2 ∈ hm2B = on the right: algebraic condition.
{ê6 }sp , and such that:
{ξ̂1 , ξ̂2 }sp = m = {ê4 , ê5 }sp . + −
b) A pair of finite-pitch twists (ξˆ0 2 , ξˆ0 2 ) with pitch (p, −p),
We shall refer to (ξ̂1+ , ξ̂2+ ; ξ̂2− , ξ̂1− ) as an even SC. When ζ̂2 = given by:
( +
0 and ξ̂2+ = ξ̂2− = ξ̂2 , we can lump the two twists together ξˆ0 2 , Ade+ψη̂2 ξ̂2+
and have an odd m2B -SC (ξ̂1+ ; ξ̂2 ; ξ̂1− ). −
ξˆ0 2 , Ade−ψη̂2 ξ̂2−
Therefore, a m2B -SC is a concentric RRRR or RRR
chain with bilateral symmetry about the char. plane. As we with η̂2 ∈ m3B having a unit direction vector of w2 , and
have pointed out earlier, without the symmetric movement (ξ̂2+ , ξ̂2− ) is another SP defined by algebraic condition:
condition (14), a m-SC may generate a general subset of the   
completion group GM = hexp gm i instead of hexp mi. Twists + 0̂ w2
 ξ̂ , ξ̂ 2 + p
 2 0T 0

of a SC need not be linearly independent (or non-redundant)  
either. Since gm2B is the three dimensional spherical Lie alge-  −
 ξ̂2 , ξ̂2 − p T
 0̂ w2
bra so(3) of SO(3), the concentric 4-R chain is necessarily 0 0
redundant. When ζ̂1 6= 0, the 3-R chain is a non-redundant + −
so(3)-chain. When ζ̂1 = ζ̂2 = 0, ξ̂i+ = ξ̂i− = ξ̂i ∈ m2B , i = (ξˆ0 2 , ξˆ0 2 ) can also be directly derived from the algebraic
1, 2 and we have a singular so(3)-chain. Both the odd m2B -SC condition:
and the even m2B -SC can be found in the design of novel CV     
0 = c ξ̂ +s pê + s ê +c p 0̂ w2
+

joints [51,52].

 2 ξˆ ψ 2 ψ 3 ψ 6 ψ
0T

0
2) m3B : Consider the conjugacy class of 3D LTS    (19)
ξˆ0 2 = cψ ξ̂2 +sψ pê3 − sψ ê6 +cψ p 0̂T w2
−
 
m3B = {ê3 , ê4 , ê5 }sp as shown in Fig. 3(d), with hm3B =


{ê1 , ê2 , ê6 }sp . It consists of a field of zero-pitch twists in, 0 0
and an infinite-pitch twist perpendicular to, the char. plane The SP corresponds to a pair of helical joints with equal
(Hunt’s 4th special 3-system, [30]). Its SP can be one of the and opposite pitches p and −p (or a pair of revolute joints if
following cases (see Fig. 5): p = 0), and is mirror symmetric about the char. plane. This
a) A pair of infinite-pitch twists (ξ̂1+ , ξ̂1− ) given by: gives an explicit proof of Hunt’s observation that mirror
   symmetric helical joints in a CV kinematic chain should
+ 0̂ ŵ1 z
 ξ̂1 , Ade+φη̂1 ê3 = cφ ê3 + sφ 0T 0 have equal and opposite pitches [35].


  (18) Note that when η̂2 ∈ m3B is a zero-pitch twist, the axes
 −
 ξ̂1 , Ade−φη̂1 ê3 = cφ ê3 − sφ
 0̂ ŵ 1 z of the two helical joints intersects at a point on the char.
0T 0 plane; when η̂2 is chosen to be the infinite-pitch twist ê3 ,
where η̂1 ∈ m3B , with unit direction vector w1 . The second the axes of the two helical joints become parallel.
half of (18) gives the algebraic condition for the same SP, A m3B -SC consists of three nesting SPs {ξ̂i+ ; ξ̂i− }3i=1 , each
where being one of the aforementioned cases. According to (16), the

0̂ ŵ1 z
 three twists ξ̂1+ , ξ̂2+ , ξ̂3+ ∈ gm3B = se(3) must be chosen such
∈ hm3B = {ê1 , ê2 , ê6 }sp that:
0T 0
{ξ̂1+ , ξ̂2+ , ξ̂3+ }sp ⊕ {ê1 , ê2 , ê6 }sp = se(3)
The SP corresponds to a pair of prismatic joints that are
symmetric about the z-axis in a plane containing them, and The enumeration of eligible candidates is studied in our earlier
is the same as a mirror symmetry about the char. plane if work [53]. The two most commonly seen m3B -SCs [35] are
we flip the direction of ξ̂1− . This will make no difference the mirror symmetric RER chain, which is equivalent to a
in type synthesis but will reverse the joint variable for ξ̂1− mirror symmetric RRPRR chain, and the mirror symmetric
in the symmetric movement condition (14). RSR chain, which is equivalent to a mirror symmetric 5-R

Fig. 6. Symmetric twist pairs of mp2A (or m2A by letting p = 0). On the
left: geometric condition, and on the right: algebraic condition.
chain. When (14) is not enforced, these chains generate 5D
submanifolds of SE(3).
The above results corroborate Hunt’s exhaustive classifica-
tion of 3D CV chains in [35]. It also confirms our earlier
conclusion that M3B , hexp m3B i is indeed the motion type
of 3-DoF CV couplings. Moreover, the mirror symmetry is
a manifestation of the inversion symmetry of the underlying
symmetric submanifold.
3) m2A and mp2A : Consider the conjugacy class of 2D LTS
p axis in the char. plane (xy-plane), and a twist of infinite-pitch
m2A = {ê3 , ê4 +pê1 }sp (and also m2A = {ê3 , ê4 }sp by letting
p = 0) as shown in Fig. 3(a), with hmp2A = {ê2 }sp . It consists
of all twists with pitch p parallel to the x-axis in the char.
plane, and also a twist of infinite pitch perpendicular to the
char. plane (Hunt’s 2nd special 2-system, [30]). Note that m2A
is a LTS (subsystem) of m3B ; its SPs can be synthesized in a
similar manner and therefore have the same type of symmetry:
as shown in Fig. 6, m2A -SPs are mirror symmetric about the
char. plane, with typical SPs given by:
( where ξ̂1 = λ1 ê1 + λ2 ê3 for some real constants λ1 , λ2 (see
ξ̂1+ , Ade+φη̂ ê3 = cφ ê3 + sφ ê2
ξ̂1− , Ade−φη̂ ê3 = cφ ê3 − sφ ê2
(
ξ̂2+ , Ade+µê3 ξ̂2 = ξ̂2 + µê2
ξ̂2− , Ade−µê3 ξ̂2 = ξ̂2 − µê2
mp2A -SPs admit exactly the same symmetry type, with the zero-
pitch SP replaced by a SP of pitch p. Unlike the case of m3B ,
the two finite-pitch twists in (ξ̂2+ ; ξ̂2− ) in Fig. 6 have equal but
not opposite pitches for the obvious reason that the LTS mp2A
itself admits finite-pitch twists.
Since gm2A = {ê2 , ê3 , ê4 }sp is the Lie algebra of the 3D
planar Euclidean group, a m2A -SC (ξ̂1+ , ξ̂2+ ; ξ̂2− , ξ̂1− ) should
satisfy:
{ξ̂1+ , ξ̂2+ }sp ⊕ {ê2 }sp = {ê2 , ê3 , ê4 }sp
A m2A -SC can be one of the following:
a) A mirror symmetric RRRR or RRR chain with parallel
axes;
b) A mirror symmetric PRRP or PRP chain with R
perpendicular to the two P’s;
c) A mirror symmetric RPPR or RPR with parallel R’s
both perpendicular to the P’s.
9
Fig. 7. Symmetric twist pairs of m3A . On the left: geometric condition, and
on the right: algebraic condition.
These are all planar motion generators if the symmetric motion
condition (14) is not enforced. Synthesis of mp2A follows
exactly the same approach and have exactly the same result
with all revolute joints replaced by helical joints with pitch p.
Therefore mp2A -SCs are planar helical motion generators when
(14) is not enforced.
4) m3A : Consider the conjugacy class of 3D LTS m3A =
{ê1 , ê3 , ê4 }sp as shown in Fig. 3(c), with hm3A = {ê2 }sp . It
consists of twists of all pitches on all lines parallel to the x-
perpendicular to the xy-plane (Hunt’s 10th special 3-system,
(p)
[30]). From the fact that m2A ⊂ m3A , we see that any m2A -
p
SPs and m2A -SPs are also m3A -SPs (compare Fig. 7 with Fig.
6). Besides, m3A admits the following SP by the algebraic
condition:
(
ξ̂1+ , Ade+φê4 ξ̂1 = λ1 ê1 + cφ λ2 ê3 − λ2 sφ ê2
ξ̂1− , Ade−φê4 ξ̂1 = λ1 ê1 + cφ λ2 ê3 + λ2 sφ ê2
Fig. 7). The prismatic SP is no longer mirror symmetric about
the xy-plane, but instead becomes mirror symmetric about the
xz-plane.
Since gm3A = {ê1 , ê2 , ê3 , ê4 }sp is the Lie algebra of the 4D
Schönflies group, a m3A -SC (ξ̂1+ , ξ̂2+ , ξ̂3+ ; ξ̂3− , ξ̂2− , ξ̂1− ) should
satisfy:
{ξ̂1+ , ξ̂2+ , ξ̂3+ }sp ⊕ {ê2 }sp = {ê1 , ê2 , ê3 , ê4 }sp
The enumeration of eligible candidates of (ξ̂1+ , ξ̂2+ , ξ̂3+ ) can
be found in [53]. Since m3A -SCs are 5 or 6-DoF chains (in
comparison to the dimension of the Schönflies group being 4),
they are redundant Schönflies motion generators in the absence
of the symmetric motion condition (14).
5) m4 : Consider the conjugacy class of 4D LTS m4 =
{ê1 , ê2 , ê4 , ê5 }sp as shown in Fig. 3(e), with hm4 =
{ê3 , ê6 }sp . It consists of twists of all pitches along the lines
of pencils in each plane normal to the z-axis, and that the
centers of the pencils all lie on the z-axis (Hunt’s 5th special
4-system, [30]). Its SP can be one of the following cases (see
Fig. 8):
a) A pair of finite-pitch twists (ξ̂1+ , ξ̂1− ), both with pitch p,
given by: ( +
ξ̂1 , Ade+φη̂1 e+λη̂10 ξ̂1
ξ̂1− , Ade−φη̂1 e−λη̂10 ξ̂1
 10

Fig. 8. Symmetric twist pairs of m4 . On the left: geometric condition, and

on the right: algebraic condition.

Fig. 9. (a) a UNITRU coupling with four U U -SCs; (b) twists of a U U chain
where η̂1 , η̂10 ∈ m4 have pitches equal to 0 and ∞ as an incomplete m3B -SC.
respectively, and ξ̂1 ∈ m4 has unit direction vector w1 and
pitch p. It can also be given by the algebraic condition:
Example 3 (UU SC). A UU SC (as shown in Fig. 9(b)) is
   briefly mentioned in [35] and studied in [36]. Its SPs are given
+ 0̂ w1 by:
 ξ̂ =c ξ̂
φ 1 +λs φ +(pcφ −λcφ )ê3 +sφ ê6
 1 0T

0 (
  (20) ξ̂1+ = Ade+λê3 ◦ Ade+φη̂1 ξ̂1 , Ade+λê3 ζ̂1+
ξ̂1− =cφ ξ̂1 +λsφ 0̂T w1

 −(pcφ −λcφ )ê3 −sφ ê6 ξ̂1− = Ade−λê3 ◦ Ade−φη̂1 ξ̂1 , Ade−λê3 ζ̂1−
0 0
(
ξ̂2+ = Ade+λê3 ◦ Ade+ψη̂2 ξ̂2 , Ade+λê3 ζ̂2+
b) A pair of infinite-pitch twists (ξ̂2+ , ξ̂2− ), given by: ξ̂2− = Ade−λê3 ◦ Ade−ψη̂2 ξ̂2 , Ade−λê3 ζ̂2−
(
ξ̂2+ , Ade+ψη̂2 ξ̂2 = cψ ξ̂2 + sψ ê3 where λ ∈ R, ξ̂1 , ξ̂2 , η̂1 , η̂2 ∈ m2B . In other words, the UU
SC is generated from a m2B SC (ζ̂1+ , ζ̂2+ ; ζ̂2− , ζ̂1− ) using a
ξ̂2− , Ade−ψη̂2 ξ̂2 = cψ ξ̂2 − sψ ê3 geometric condition (17) for m3B .
The SC generates a 2D submanifold of M3B under symmet-
where η̂2 ∈ m4 has pitch 0, and ξ̂2 ∈ m4 has infinite-pitch. ric motion condition (14). In comparison to a m3B -SC, it has
not a free but a dependent translational DoF [29]. Rosheim
If we flip the direction of ξ̂i− ’s, (ξ̂i+ , ξ̂i− ), i = 1, 2 both used this phenomenon to characterize the human shoulder
admit a 2-fold rotational symmetry about the z-axis, with complex movement [41].
the two twists (ξ̂1+ , ξ̂1− ) having equal finite pitch of p (or in The UNITRU coupling can be synthesized using three
particular 0). Therefore, a m4 -SP may be a pair of prismatic, or more identical UU SCs with axial symmetry, since the
helical or revolute joints. generated 2D submanifold is conjugation invariant by elements
Since gm4 = se(3), a m4 -SC (ξ̂1+ , . . . , ξ̂4+ ; ξ̂4− , . . . , ξ̂1− ) of exp{ê6 }sp [29].
should satisfy:
V. C ONCLUSION
{ξ̂1+ , . . . , ξ̂4+ }sp ⊕ {ê3 , ê6 }sp = se(3) In this paper, we have systematically studied inversion
symmetry of the special Euclidean group SE(3) and its
Eligible candidates of (ξ̂1+ , . . . , ξ̂4+ ) can be found in [53]. symmetric submanifolds that arise from kinesiology and robot
Since m4 -SCs are 7 or 8-DoF chains, they are redundant (and mechanical systems. Besides sharing many similarities with
possibly, singular) SE(3) motion generators in the absence of Lie subgroups of SE(3), symmetric submanifolds expand
the symmetric motion condition (14). our general knowledge of motion types for the analysis and
We remark that since m5 contains all other LTSs, its SPs synthesis of many kinesiological joints or robot mechanical
can be any of the SPs of the aforementioned LTSs and do not generators that defy a Lie group explanation.
possess a particular symmetry type. The derivation of its SCs The main contribution of our work is as follows. First, we
can be conducted in a similar manner to the previous cases. have identified, for the first time, seven conjugacy classes
Due to space limitations, a complete treatment of m5 -SPs and of symmetric submanifolds of SE(3) (Table I), which are
SCs will not be given here. generated via inversion symmetry by the exponential image
Finally, when a SC of a kD LTS m has less than k nesting of Lie triple subsystems of se(3). So far as the authors are
SPs, its symmetric movement generates a submanifold of the aware of, this is also the first time the inversion symmetry
symmetric submanifold M = hexp mi. Such incomplete SCs of SE(3) is studied. Second, we found that these symmetric
are also prevalent in practice. submanifolds share both a list of geometric properties and also
a universal type synthesis method for their symmetric twist
 11

chains, which are kinematic chains comprising nesting sym-
metric twist pairs. Third, the symmetry type of the symmetric
pairs and symmetric chains for each LTS is systematically
studied, thereby systematically extending (if not completing)
Hunt’s observation made more than forty years ago.
Our ongoing work comprises: systematic type synthesis
of parallel manipulators generating symmetric submanifolds
and systematic kinematics and singularity analysis of such
parallel manipulators (some preliminary case study is available
in [50]); application of symmetric submanifolds of SE(3) in
robot dynamics, planning and control; study on general Exp-
submanifolds exp n with n a general vector subspace of se(3)
that is neither a Lie subalgebra nor a Lie triple subsystem.
Therefore, G̃ is generated by elements of the form
(eξ̂+ζ̂ , e−ξ̂+ζ̂ ) and a generic element of M is given by
τ ((eξ̂+ζ̂ , e−ξ̂+ζ̂ ), g) = eξ̂+ζ̂ geξ̂−ζ̂ , ∀g ∈ M. The vector sub-
spaces m, hm and gm in Prop.2 b), c) are in fact the projection
of m̃, h̃ and g̃ into the first component of se(3) × se(3).
A PPENDIX B
P ROOF OF EQ . (8)
Please refer to [54] for basic concepts in differential geom-
etry and Lie groups used in this proof.
Given any smooth function f : SE(3) → R and two right-
invariant vector fields ξ̂ r , ζ̂ r defined by ξ̂, ζ̂ ∈ se(3), we have
for any g ∈ SE(3):

ACKNOWLEDGEMENTS [ξ̂ r , ζ̂ r ]f (g) = (ξ̂ r ζ̂ r − ζ̂ r ξ̂ r )f (g)

d r d r
The first author thanks Dr.Jon Selig and Dr.Liu Guanfeng = ζ̂ f (euξ̂ g)|0 − ξ̂ f (euζ̂ g)|0
for some fruitful discussions. du du
d r uξ̂ d r uζ̂
= ζ̂ (e g)f |0 − ξ̂ (e g)f |0
du du
A PPENDIX A d d
S YMMETRIC S PACE AS H OMOGENEOUS S PACE [42,43] = (ζ̂euξ̂ g)f |0 − (ξ̂euζ̂ g)f |0
du du
The group of displacements G̃ generated by Q(SE(3)) is = (ζ̂ ξ̂g − ξ̂ ζ̂g)f = −[ξ̂, ζ̂]r f (g)
the Cartesian product SE(3)×SE(3) which acts on SE(3) via:
and therefore [ξ̂ r , ζ̂ r ] = −[ξ̂, ζ̂]r .
τ : G̃ × SE(3) → SE(3), ((g1 , g2 ), h) 7→ g1 hg2−1 (21) Similarly,
where the quadratic representation Q(g) of g ∈ SE(3) is given [ξ̂ l , ζ̂ l ]f (g) = (ξ̂ l ζ̂ l − ζ̂ l ξ̂ l )f (g)
by (g, g−1 ) ∈ G̃. The associated involution σ : G̃ → G̃ is d l d l
given by: = ζ̂ f (geuξ̂ )|0 − ξ̂ f (geuζ̂ )|0
du du
d l uξ̂ d l uζ̂
σ(g1 , g2 ) = SI ◦ (g1 , g2 ) ◦ SI = (g2 , g1 ) = ζ̂ (ge )f |0 − ξ̂ (ge )f |0
du du
the stabilizer Hσ of σ equals {(g, g)|g ∈ SE(3)}, and is the d d
= (geuξ̂ ζ̂)f |0 − (geuζ̂ ξ̂)f |0
same as the isotropy group H̃ of I ∈ SE(3). du du
The homogeneous space G̃/H̃ is diffeomorphic to SE(3) = (gξ̂ ζ̂ − gζ̂ ξ̂)f = [ξ̂, ζ̂]l f (g)
via:
and therefore [ξ̂ l , ζ̂ l ] = [ξ̂, ζ̂]l .
π : G̃/H̃ → SE(3), (g1 , g2 )H̃ 7→ g1 g2−1
Finally,
and dπH̃ (ξ̂1 , ξ̂2 ) = ξ̂1 − ξ̂2 . [ξ̂ r , ζ̂ l ]f (g) = (ξ̂ r ζ̂ l − ζ̂ l ξ̂ r )f (g)
σ induces an involution dσ(I,I) on se(3) × se(3):
d l d r
= ζ̂ f (euξ̂ g)|0 − ξ̂ f (geuζ̂ )|0
dσ(I,I) : (ξ̂1 , ξ̂2 ) 7→ (ξ̂2 , ξ̂1 ) du du
d l uξ̂ d r
Its two eigenspaces h̃ (of eigenvalue 1) and m̃ (of eigenvalue = ζ̂ (e g)f |0 − ξ̂ (geuζ̂ )f |0
du du
−1) are given by: d uξ̂ d
= (e gζ̂)f |0 − (ξ̂geuζ̂ )f |0
du du
h̃ , {(ξ̂, ξ̂)|ξ̂ ∈ se(3)}, m̃ , {(ξ̂, −ξ̂)|ξ̂ ∈ se(3)}
= (ξ̂gζ̂ − ξ̂gζ̂)f = 0
and satisfy:
and therefore [ξ̂ r , ζ̂ l ] = 0.
[h̃, h̃] ⊂ h̃, [h̃, m̃] ⊂ m̃, [m̃, m̃] ⊂ h̃
A PPENDIX C
dπH̃ maps the Lie subalgebra h̃ and LTS m̃ to {0̂} and se(3) P ROOF OF T HEOREM 1
respectively, thus identifying m̃ with se(3).
Given a symmetric submanifold M with TI M = m and In reference to (7), the group of displacements G̃(M) of M
hm , [m, m], define m̃ and h̃ by: is generated by Q(exp m) , {Q(eξ̂ )|ξ̂ ∈ m}. Since G̃(M)
acts transitively on M, M is recovered by letting G̃(M) act on
m̃ , {(ξ̂, −ξ̂)|ξ̂ ∈ m}, h̃ , {(ζ̂, ζ̂)|ζ̂ ∈ hm } the identity I ∈ M:
The group of displacements of M is generated by the Lie Q(eξ̂ )(I) = eξ̂ Ieξ̂ = e2ξ̂ ∈ M, ξ̂ ∈ m
algebra g̃ given by:
Therefore exp m ⊂ M and also contains an open neighborhood
g̃ , h̃ ⊕ m̃ = {(ξ̂ + ζ̂, −ξ̂ + ζ̂)|ξ̂ ∈ m, ζ̂ ∈ hm } of M, which generates M via inversion symmetry, i.e. M =
 12

{gh−1 g ∈ SE(3)|∀g, h ∈ exp m}. For more details, see [43, A PPENDIX E
pp. 95, Prop. 3.2]. P ROOF OF P ROPOSITION 2
To see that {ghg ∈ SE(3)|∀g, h ∈ exp m} is the same Proof of a): Since Q(eξ̂/2 ) is a diffeomorphism sending I
as {gh−1 g ∈ SE(3)|∀g, h ∈ exp m}, note that if h = eξ̂ ∈ to g = eξ̂ , it induces an isomorphism from m onto Teξ̂ M.
exp m, so is h−1 = e−ξ̂ ∈ exp m. Therefore:
(Tg M)g−1 = (eξ̂/2 meξ̂/2 )e−ξ̂
A PPENDIX D = eξ̂/2 me−ξ̂/2 = Adeξ̂/2 m
P ROOF OF P ROPOSITION 1
Similarly, Q(eξ̂1 ) ◦ Q(eξ̂2 /2 ) sends I to g = eξ̂1 eξ̂2 eξ̂1
The proposition is proved by straightforward computation, diffeomorphically, and
as follows. Consider first the 2D LTS m2A = {ê3 , ê4 }sp .
exp m2A comprises homogeneous matrices of the form: (Tg M)g−1 = (eξ̂1 eξ̂2 /2 meξ̂2 /2 eξ̂1 )e−ξ̂1 e−ξ̂2 e−ξ̂1
 θ1  = (eξ̂1 eξ̂2 /2 )m(eξ̂1 eξ̂2 /2 )−1
exp(θ1 ê4 + θ2 ê3 ) = e θ1 x̂
θ20 e 2 x̂
z = Adeξ̂1 eξ̂2 /2 m
0T 1
Proof of b): For any [ξ̂1 , ξ̂2 ] ∈ hm and ξ̂ ∈ m:
where θ1 , θ2 ∈ R and θ20 = 2sθ1 /2 θ2 , and also ∞ adi
[ξ̂1 ,ξ̂2 ]
ad[ξ̂
X
  Ade[ξ̂1 ,ξ̂2 ] ξ̂ = e 1 ,ξ̂2 ] ξ̂ = ξ̂ ∈ m. (22)
I θ2 z i!
exp θ2 ê3 = i=0
0T 1
since ad[ξ̂1 ,ξ̂2 ] ξ̂ , [[ξ̂1 , ξ̂2 ], ξ̂] ∈ m (since m is a LTS). Here
Combining the two cases, exp m2A comprises elements of the we have used the adjoint map ad(·) defined by:
form:
 θ1  adξ̂1 ξ̂2 , [ξ̂1 , ξ̂2 ], ∀ξ̂1 , ξ̂2 ∈ se(3)
eθ1 x̂ θ2 e 2 x̂ z , θ , θ ∈ R
1 2
0T 1 and the fact that ([19, pp. 54, Lemma 3.4.1]):

To see these are the only elements of M2A , compute ghg for Adeξ̂ = eadξ̂ , ∀ξ̂ ∈ se(3)
any g, h ∈ exp m2A (see eq. (11)): Now we have Ade[ξ̂1 ,ξ̂2 ] m ⊆ m. The equality holds because
θ1 " 0 0
# θ1
Ade[ξ̂1 ,ξ̂2 ] is a linear isomorphism. Since HM is generated by
 θ1 
eθ1 x̂ θ2 e 2 x̂ z eθ1 x̂ θ20 e 2 x̂ z eθ1 x̂ θ2 e 2 x̂ z exp hm and Adgh = Adg ◦ Adh , ∀g, h ∈ SE(3), we have
0T 1 0T 1 0T 1
| {z }| {z }| {z } ∀h ∈ HM ⇒ Adh m ≡ m (23)
g h g
ξ̂
"
0
0
2θ1 +θ1
# Next, in light of Prop.1, for any ∀g ∈ M, either g = e , ξ̂ ∈
(2θ1 +θ1 )x̂
= eT (θ20 +2c(θ1 +θ10 )/2 )e 2 x̂
z m or g = eξ̂ eη̂ eξ̂ , η̂, ξ̂ ∈ m. In the former case,
0 1
C(h)(eξ̂ ) = heξ̂ h−1 = eAdh ξ̂ ∈ exp m ⊆ M
∈ exp m2A
by the previous equation (23). The latter case:
This verifies that M2A = exp m2A , or exp : m2A → M2A is
surjective. In a similar way, we can verify that exp is also C(h)(eξ̂ eη̂ eξ̂ ) = heξ̂ eη̂ eξ̂ h−1
surjective for m2B , m3A , m3B , m4 and m5 . = heξ̂ h−1 heη̂ h−1 heξ̂ h−1
For mp2A = {ê3 , ê4 + pê1 }sp , it can be verified that the
only elements in Mp2A that do not correspond to elements in = C(h)(eξ̂ ) · C(h)(eη̂ ) · C(h)(eξ̂ )
exp mp2A are of the form (c.f. [48, Prop. 2.2]): is reduced to the former. Therefore we have C(h)(M) ⊆ M.
  The equality holds since C(h) is a diffeomorphism for any
I 2ipπx + θz h ∈ SE(3), i.e.
, i ∈ Z − {0}, θ ∈ R − {0}
0T 1
∀h ∈ HM ⇒ C(h)(M) ≡ M
which can be generated by exp mp2A via only one inversion: Proof of c): gm , hm + m is a Lie subalgebra and therefore
  contains the completion algebra of m. On the other hand, the
I 2ipπx + θz
= completion algebra should contain both m and hm , [m, m].
0T 1
Therefore, gm is the completion algebra of m.
I θ2 z I θ2 z
    
I 2ipπx Next, since m ⊂ gm , we have
0T 1 0T 1 0T 1
θ θ M = hexp mi ⊂ hexp gm i = GM
= e 2 ê3 e2iπ(ê4 +pê1 ) e 2 ê3
On the other hand, The Lie algebra of the (connected) comple-
In particular exp is not surjective for mp2A . tion group of M will be a Lie subalgebra of se(3) containing
 13

m and therefore gm . Therefore, GM is indeed the completion with  ∞ ∞

group of M. X ad2i
η̂
X ad2i+1
η̂
ζ̂ ∈ m


 ξ̂ +
If hm ∩ m = {0̂}, we have gm = hm ⊕ m and 
 (2i)! (2i + 1)!
i=0 i=0
∞ ∞
xpg : m × hm → GM
eg  X ad2i
η̂
X ad2i+1
η̂
ξ̂ ∈ hm


 ζ̂ +
defines a local diffeomorphism by the inverse function theorem
 (2i)!
i=0
(2i + 1)!
i=0
[54].
by (10). Thus (Ade+η̂ ξ̂ , Ade−η̂ ξ̂ − ) defines a SP by the
+

A PPENDIX F algebraic condition (15).

P ROOF OF P ROPOSITION 3 R EFERENCES
The first claim is immediate, since according to Appendix [1] Y. Wu, G. Liu, H. Löwe, and Z. Li, “Exponential submanifolds: A new
+ −
A, eξ̂ geξ̂ with kinematic model for mechanism analysis and synthesis,” in Robotics
( and Automation (ICRA), 2013 IEEE International Conference on, May
ξ̂ + = ξ̂ + ζ̂ 2013, pp. 4177–4182.
, ξ̂ ∈ m, ζ̂ ∈ hm , g ∈ M [2] J. M. Hervé, “Analyse structurelle des mécanismes par groupe des
ξ̂ − = ξ̂ − ζ̂ déplacements,” Mechanism and Machine Theory, vol. 13, no. 4, pp.
437–450, 1978.
is a generic element of M. [3] ——, “Lie group of rigid body displacements, a fundamental tool for
The second claim is the same as saying that when mechanism design,” Mechanism and Machine Theory, vol. 34, no. 5, pp.
{ξ̂1 , . . . , ξ̂k }sp = m, 719–730, 1999.
[4] R. W. Brockett, “Robotic manipulators and the product of exponentials
+ + − −
{eθ1 ξ̂1 · · · eθk ξ̂k eθk ξ̂k · · · eθ1 ξ̂1 |θ1 , . . . , θk ∈ R} formula,” in In P.A. Fuhrman, editor, Mathematical Theory of Networks
and Systems. Springer-Verlag, 1984, pp. 120–129.
with ξ̂i± = ξ̂i ± ζ̂i , ξ̂i ∈ m, ζ̂i ∈ hm is locally identical [5] J. Beckers, J. Patera, M. Perroud, and P. Winternitz, “Subgroups of
the Euclidean group and symmetry breaking in nonrelativistic quantum
to M = hexp mi. The Jacobian of the map (θ1 , . . . , θk ) 7→ mechanics,” Journal of mathematical physics, vol. 18, no. 1, pp. 72–83,
+ + − −
eθ1 ξ̂1 · · · eθk ξ̂k eθk ξ̂k · · · eθ1 ξ̂1 at (0, . . . , 0) is given by: 1977.
[6] R. M. Murray, Z. Li, and S. S. Sastry, A mathematical introduction to
(θ̇1 , . . . , θ̇k ) 7→θ̇1 (ξ̂1+ + ξ̂1− ) + · · · + θ̇k (ξ̂k+ + ξ̂k− ) robotic manipulation. CRC Press, 1994.
[7] F. C. Park, J. E. Bobrow, and S. R. Ploen, “A Lie group formulation of
= 2θ̇1 ξ̂1 + · · · + 2θ̇k ξ̂k robot dynamics,” The International journal of robotics research, vol. 14,
no. 6, pp. 609–618, 1995.
The claim follows immediately from the inverse function [8] A. Müller and P. Maisser, “A Lie-group formulation of kinematics
theorem [54]. and dynamics of constrained mbs and its application to analytical
mechanics,” Multibody System Dynamics, vol. 9, no. 4, pp. 311–352,
Finally, if hm ∩ m = {0̂} and {ξ̂1 , . . . , ξ̂k }sp = m, we have: 2003.
[9] J. Kwon, M. Choi, F. C. Park, and C. Chun, “Particle filtering on the
{ξ̂1 ± ζ̂1 , . . . , ξ̂k ± ζ̂k }sp + hm = gm Euclidean group: framework and applications,” Robotica, vol. 25, no. 06,
pp. 725–737, 2007.
The sum being direct follows from a dimension argument. On [10] R. Mahony, T. Hamel, and J.-M. Pflimlin, “Nonlinear complementary
the other hand, if filters on the special orthogonal group,” Automatic Control, IEEE
Transactions on, vol. 53, no. 5, pp. 1203–1218, 2008.
{ξ̂1 ± ζ̂1 , . . . , ξ̂k ± ζ̂k }sp ⊕ hm = gm [11] F. Bullo and R. M. Murray, “Proportional derivative (PD) control on the
Euclidean group,” in European Control Conference, vol. 2, 1995, pp.
A change of basis leads to 1091–1097.
[12] J. M. Hervé and F. Sparacino, “Structural synthesis of parallel robots
{ξ̂1 , . . . , ξ̂k }sp ⊕ hm = gm generating spatial translation,” in Proc. 5th Int. Conf. Advanced Robotics,
vol. 1, 1991, pp. 808–813.
The claim follows from a dimension argument. [13] A. Frisoli, D. Checcacci, F. Salsedo, and M. Bergamasco, “Synthesis
by screw algebra of translating in-parallel actuated mechanisms,” in
Advances in Robot Kinematics. Springer, 2000, pp. 433–440.
A PPENDIX G [14] M. Carricato and V. Parenti-Castelli, “Singularity-free fully-isotropic
P ROOF OF C OROLLARY 1 translational parallel mechanisms,” The International Journal of
Robotics Research, vol. 21, no. 2, pp. 161–174, 2002.
Given a SP (ξ̂ + , ξ̂ − ): [15] ——, “A family of 3-dof translational parallel manipulators,” Journal of
mechanical design, vol. 125, no. 2, pp. 302–307, 2003.
(
ξ̂ + = ξ̂ + ζ̂ [16] Z. Huang and Q. Li, “Type synthesis of symmetrical lower-mobility par-
ξ̂ ∈ m, ζ̂ ∈ hm
ξ̂ − = ξ̂ − ζ̂ allel mechanisms using the constraint-synthesis method,” International
Journal of Robotics Research, vol. 22, no. 1, pp. 59–79, 2003.
and some η̂ ∈ m, we have: [17] X. Kong and C. Gosselin, “Type synthesis of 3-dof spherical parallel
manipulators based on screw theory,” Journal of Mechanical Design,
Ade±η̂ ξ̂ ± = e±adη̂ (ξ̂ ± ζ̂) vol. 126, no. 1, pp. 101–108, 2004.
[18] J. Meng, G. Liu, and Z. Li, “A geometric theory for analysis and syn-
∞ ∞
ad2i+1
!
X ad2iη̂
X η̂ thesis of sub-6 dof parallel manipulators,” Robotics, IEEE Transactions
= ± (ξ̂ ± ζ̂) on, vol. 23, no. 4, pp. 625 –649, aug. 2007.
i=0
(2i)! i=0 (2i + 1)! [19] J. Hilgert and K.-H. Neeb, Structure and geometry of Lie groups.
∞ ∞ Springer Science & Business Media, 2011.
ad2i+1
!
X ad2iη̂
X η̂ [20] J. M. Hervé. (2003) The planar spherical kinematic bond:
= ξ̂ + ζ̂ ± Implementation in parallel mechanisms. [Online]. Available:
i=0
(2i)! i=0
(2i + 1)!
http://www.parallemic.org/Reviews/Review013.html
∞ ∞
ad2i+1
!
X ad2iη̂
X η̂
[21] C. C. Lee and J. M. Herve, “Translational parallel manipulator with
ζ̂ + ξ̂ doubly planar limbs,” Mechanism and Machine Theory, vol. 41, no. 4,
i=0
(2i)! i=0
(2i + 1)! pp. 359–486, 2006.

[22] M. Carricato and J. M. R. Martı́nez, “Persistent screw systems,” in
Advances in Robot Kinematics: Motion in Man and Machine. Springer,
2010, pp. 185–194.
[23] M. Carricato and D. Zlatanov, “Persistent screw systems,” Mechanism
and Machine Theory, vol. 73, pp. 296–313, 2014.
[24] M. Carricato and J. M. R. Martınez, “Persistent screw systems of
dimension three,” in Proc. of 13th World Congress in Mechanism and
Machine Science, Guanajuato, Mexico, 2011, pp. 1–12.
[25] M. Carricato, “Persistent screw systems of dimension four,” in Latest
Advances in Robot Kinematics. Springer, 2012, pp. 147–156.
[26] ——, “Four-dimensional persistent screw systems of the general type,”
in Computational Kinematics, ser. Mechanisms and Machine Science.
Springer Netherlands, 2014, vol. 15, pp. 299–306.
[27] X. Kong and C. Gosselin, Type synthesis of parallel mechanisms.
Springer-Verlag, Berlin, Heidelberg, 2007.
[28] Q. Li, Z. Huang, and J. M. Herve, “Type synthesis of 3R2T 5-dof parallel
mechanisms using the Lie group of displacements,” IEEE Transactions
on Robotics, vol. 20, no. 2, pp. 173–180, 2004.
[29] Y. Wu, Z. Li, and J. Shi, “Geometric properties of zero-torsion parallel
kinematics machines,” in Intelligent Robots and Systems (IROS), 2010
IEEE/RSJ International Conference on. IEEE, 2010, pp. 2307–2312.
[30] K. H. Hunt, Kinematic geometry of mechanisms. Oxford University
Press, 1978.
[31] I. A. Bonev, D. Zlatanov, and C. M. Gosselin, “Advantages of the
modified euler angles in the design and control of pkms,” 2002 Parallel
Kinematic Machines International Conference, pp. 171–188, 2002.
[32] I. A. Bonev and J. Ryu, “New approach to orientation workspace anal-
ysis of 6-dof parallel manipulators,” Mechanism and Machine Theory,
vol. 36, no. 1, pp. 15–28, 2001.
[33] D. Tweed and T. Vilis, “Geometric relations of eye position and velocity
vectors during saccades,” Vision Research, vol. 30, no. 1, pp. 111–127,
1990.
[34] A. D. Polpitiya, W. P. Dayawansa, C. F. Martin, and B. K. Ghosh,
“Geometry and control of human eye movements,” Automatic Control,
IEEE Transactions on, vol. 52, no. 2, pp. 170–180, 2007.
[35] K. H. Hunt, “Constant-velocity shaft couplings: a general theory,” ASME
J. Eng. Ind, vol. 95, no. 2, pp. 455–464, 1973.
[36] M. Carricato, “Decoupled and homokinetic transmission of rotational
motion via constant-velocity joints in closed-chain orientational manip-
ulators,” Journal of Mechanisms and Robotics, vol. 1, no. 4, pp. 1–14,
2009.
[37] S. L. Canfield, A. J. Ganino, C. F. Reinholtz, and R. J. Salerno,
“Spatial, parallel-architecture robotic carpal wrist,” U.S. Patent 5 699
695, Dec. 23, 1997.
[38] R. Behrens, M. Poggendorf, E. Schulenburg, and N. Elkmann, “An
elephant’s trunk-inspired robotic arm: Trajectory determination and
control,” in Robotics; Proceedings of ROBOTIK 2012; 7th German
Conference on. VDE, 2012, pp. 1–5.
[39] I. H. Culver, “Constant velocity universal joint,” U.S. Patent 3 477 249,
Nov. 13, 1969.
[40] M. E. Rosheim and G. F. Sauter, “New high-angulation omni-directional
sensor mount,” in Proceedings of SPIE - The International Society for
Optical Engineering, vol. 4821, 2002, pp. 163–174.
[41] M. E. Rosheim, Leonardo’s lost robots. Springer, 2006.
[42] S. Kobayashi and K. Nomizu, Foundations of differential geometry.
New York, 1963, vol. 2.
[43] O. Loos, Symmetric spaces. WA Benjamin, 1969, vol. 1.
[44] J. M. Selig, Geometrical Fundamental of Robotics, ser. Monographs in
Computer Science. New York ; Hong Kong: Springer, 2005.
[45] S. Kobayashi and K. Nomizu, Foundations of differential geometry.
New York, 1963, vol. 1, no. 2.
[46] J. M. Selig, “A class of explicitly solvable vehicle motion problems,”
Robotics, IEEE Transactions on, vol. 31, no. 3, pp. 766–777, June 2015.
[47] Y. Wu and M. Carricato, “Identification and geometric characterization
of Lie triple screw systems,” in 14th IFTOMM World Congress, Taipei,
Taiwan, October 25–30, 2015.
[48] R. Chhabra and M. R. Emami, “A generalized exponential formula for
forward and differential kinematics of open-chain multi-body systems,”
Mechanism and Machine Theory, vol. 73, pp. 61–75, 2014.
[49] F. C. Park and K. Okamura, “Kinematic calibration and the product of
exponential formula,” Advances in Robot Kinematics and Computational
Geometry, pp. 119–128, 1994.
[50] Y. Wu and M. Carricato, “Design of a novel 3-dof serial-parallel robotic
wrist: A symmetric space approach,” in Proceedings of International
Symposium on Robotics Research, September 12 – 15, Sestri Levante,
Italy, 2015.
14
[51] G. A. Thompson, “Constant velocity coupling and control system
therefor,” U.S. Patent 7 144 326, Dec. 5, 2006.
[52] H. Kocabas, “Design and analysis of a spherical constant velocity
coupling mechanism,” Journal of Mechanical Design, vol. 129, no. 9,
pp. 991–998, 2007.
[53] Y. Wu, H. Wang, and Z. Li, “Quotient kinematics machines: Concept,
analysis, and synthesis,” ASME Journal of Mechanisms and Robotics,
vol. 3, no. 4, pp. 041 004–1 – 041 004–1, Nov 2011.
[54] W. M. Boothby, An introduction to differentiable manifolds and Rie-
mannian geometry, 2nd ed. Academic Press, 2003.
Yuanqing Wu received his B.Sc. degree in me-
chanical engineering in 2001 and Ph.D. degree in
mechatronics engineering in 2009, both from Shang-
hai Jiaotong University, China.
He was first a research assistant and later a
research associate at the Electronic and Computer
Engineering Department, Hong Kong University of
Science and Technology, China. He is currently a
contract researcher with the Department of Industrial
Engineering, University of Bologna. His research
interests includes kinematic geometry, mechanism
theory, parallel robots, geometric control and geometric algebra computing.
Harald Löwe received his Diploma (with honors)
in mathematics in 1990 and the degree of a Dr.
rer. nat. (summa cum laude) in 1994, both from
the Technical University of Braunschweig, Germany.
Since his Habilitation in 2001, he is working as a
professor at the Department of Mathematics of the
Technical University of Braunschweig.
He was visiting professor at the University of La
Laguna (Spain), at the University Henri Poincaré
(France). Recently, he is guest professor at the Bei-
jing Union University (China). He is chief editor of
the journal “Mathematikinformation”. He received the Heinrich-Büssing Price
for outstanding research in 2002. Currently, he is concerned with projects in
mathematics with high schools. His research interests are geometry, geometric
methods in robotics, and the design of spiral bevel gears.
Marco Carricato (M’10) received the M.Sc. degree
(with honors) in mechanical engineering in 1998 and
the Ph.D. degree in mechanics of machines in 2002.
He has been with the University of Bologna since
2004. He is currently an associate professor.
He was visiting researcher at the University of
Florida (USA), at Laval University (Canada), at
the University of Guanajuato (Mexico), at INRIA–
Sophia Antipolis (France), at the Ecole Centrale of
Nantes (France), and at the Hong Kong University
of Science and Technology. He is an associate editor
of the journal Mechanism and Machine Theory. He was awarded the AIMETA
Junior Prize 2011 by the Italian Association of Theoretical and Applied
Mechanics for outstanding results in the field of mechanics of machines. His
research interests include robotic systems, servo-actuated automatic machin-
ery and the theory of mechanisms, with a particular emphasis on parallel
manipulators, efficiency and optimization of servomechanisms, the theory of
screws, and the theory of homokinetic couplings.
 15

Zexiang Li (M’83–F’08) received the B.S. degree

in electrical engineering and economics (with hons.)
from Carnegie Mellon University (CMU), Pitts-
burgh, PA, in 1983, the M.A. degree in mathematics
and the Ph.D. degree in electrical engineering and
computer science from the University of California,
Berkeley, in 1985 and 1989, respectively.
He worked at ALCOA, the Robotics Institute of
CMU and the AI Lab of MIT (89–90). He was
an assistant professor at the Courant Institute of
New York University (90–92). In 1992, he joined
the Department of Electronic and Computer Engineering of the Hong Kong
University of Science and Technology (HKUST) and is currently a professor
of the department. He founded the Automation Technology Center (ATC) and
co-founded the Robotics Institute (RI) at HKUST. He received the ALCOA
Foundation Fellowship in 1979, and the E. Anthony Fellowship in 1983. He
was a recipient of the University Scholar award from CMU in 1983, the
E.I. Jury award from UC Berkeley in 1989, the Research Initiation award
from NSF (US) in 1990, the Outstanding Young Researcher award (Class
B) from NSF China in 2000, the LEAD award from AMI, USA in 2001,
and the Natural Science award (3rd class) from China in 1997. His research
interests include multi-fingered robotic hand, parallel manipulators, workpiece
localization and inspection, motion control, precision assembly, and unmanned
aerial vehicles (UAVs). He is a co-founder of Googol Technology and DJI.