By Dimiter Zlatanov, Ilian Bonev, and Clément Gosselin
If you wish to cite the current document, please cite the following paper instead: 
Zlatanov, D., Bonev, I.A., and Gosselin, C.M., "Constraint Singularities of Parallel Mechanisms," IEEE International Conference on Robotics and Automation (ICRA 2002), Washington, D.C., USA, May 11–15, 2002.

This report is a followup of the one explaining the puzzling singularity of a translational parallel manipulator built
at Seoul National University. It examines the phenomenon of constraint singularity in a more general context.
1. Introduction
Since we published on the singularity of the SNU parallel
manipulator we have received a lot of feedback. Many people found the issue fascinating and sent us comments and
questions. It became clear that the ability of 3UPU translational manipulators to allow rotations in certain
configurations has already been discovered by Di Gregorio and ParentiCastelli [1,2]. The occurrence of the same
phenomenon for a 3PUU translational robot was studied by Matteo Zoppi in his Master's thesis in
PMAR lab at the University of Genoa, completed in 2000. In the present paper, we will briefly comment on the
results of [1] and [2]. Then, we will look in more depth into the nature of what we called
constraint singularities and show that they represent a more general phenomenon which is not limited to
the "rotational singularities" [1] of translational parallel robots.
In [1] the authors observe that there exist configurations where a general 3RRPRR translational robot may
allow a fourth instantaneous DOF of the platform. They also list a few cases when this can occur. In their fine
paper [2] Di Gregorio and ParentiCastelli analyse the singularities of a general 3UPU parallel mechanism
assembled for translation, i.e., satisfying the conditions given by Tsai [3]. To do that, they start with the most
general velocity equations of the mechanism, i.e., the expressions of the (angular and linear) velocity of the
platform as linear functions of all joint velocities, active and passive and find the conditions for a
nonzero endeffector instantaneous motion to be possible when the actuator velocities are all zero. After studying
the linear system of equations, the authors observe that all conditions for a nonzero platform motion can be separated
into conditions for either a redundant translation or an uncontrolled rotation of the end effector. They call the two
groups of configurations, translational and rotational singularities, respectively.
2. The Importance of the Passive Joint Velocities
The first important observation that should be made is that the reason why Di Gregorio and ParentiCastelli,
unlike others, did not miss the rotational singularities was their use of the full velocity equations rather than only
the inputoutput velocity equations, where the passive velocities are eliminated.
The fact that the singularities of a mechanism with closed kinematic loops cannot be properly described and understood
if only inputoutput equations are used was emphasised in [4], where singularities were defined and classified using the
general velocity equation of an arbitrary mechanism. For some parallel mechanisms, a careful selection of the inputoutput
equations allows comprehensive singularity analysis [5]. These are mechanisms where the
mobility of the serial chain of each leg is the same as that of the mechanism as a whole (e.g., 6DOF platforms, spherical
and planar mechanisms, positioning mechanisms with only prismatic joints). However, for many parallel mechanisms with fewer
than six DOFs this condition does not hold since each leg has more freedoms and fewer constraints than the platform –
only the combination of the legs' constraints results in the desired mobility of the platform.
It should be noted that, even for the mechanisms discussed in [5], there is one singularity type that cannot be described
with the inputoutput equation alone. This is the singularity of type IIM (increased instantaneous mobility) also referred
to as an uncertainty configuration by Hunt (for singleloop chains) [6]. In such configurations the mechanism as a whole
gains an additional degree of freedom. Note that this phenomenon is obviously unrelated to the choice of the actuated joints
or the endeffector in a kinematic chain.
We will return to IIM singularities and their relationship with constraint singularities later.
3. The Importance of Geometrical Insight
Di Gregorio and ParentiCastelli correctly derive the condition for a rotational singularity of a 3UPU mechanism, namely,
the linear dependence of the three normal vectors to the Ujoint planes in the three legs.
To reach this geometrical result they apparently needed to calculate symbolically the determinant of the relevant Jacobian, an
expression that takes half a page in their paper's Appendix. This amazing formula is, in a way, a perfect advertisement for the
application of geometrical methods, such as screw theory, in robotics. Indeed, the result becomes almost obvious when reciprocal
screws are used. The constraint wrench of each leg is a pure moment normal to the Ujoint plane; the platform can rotate
only when the constraint system of the platform does not contain all couples, i.e., when the three moments are linearly dependent.
Moreover, with a knowledge of the basic screw systems it is easy to see which exactly are the acquired freedoms when the
constraints do become singular. This is why one of the authors of the present paper was able to confidently describe the redundant
instantaneous freedoms of the SNU robot at the breakfast table in Seoul on the following morning after the lab visit, and not even
a paper napkin was needed for the proof.
4. Constraint Singularity is More Than Just Rotation Singularity
It is important to realize that the phenomenon of constraint singularity is more general and more fundamental than the ability of
certain translational robots to rotate. To make this completely obvious, let us look at an example of a constraint singularity that
can also be described as a translation singularity. Consider another 3UPU robot but this time it is assembled not as a
translational device but rather as a spherical wrist (Fig. 1). The design is such that all three R joint axes in the base
have a common intersection point; the same is true for the three platform R joints; and, finally, these two points are made to
coincide. This mechanism was first described in [7].
Figure 1: A 3UPU orientational parallel manipulator in (a) a nonsingular configuration and (b) a constrain singularity
The constraint wrench for each leg is a pure force through the rotation centre parallel to the two R joints fixed in the leg.
Normally, these three forces are linearly independent. The constraint system is then composed of all forces through the centre and
hence the system of platform freedoms is the reciprocal screw system, namely, all rotations about the point O. Which explains
why the mechanism is indeed rotational.
However, if the three leg planes can intersect in a common line, as is the case in Fig. 1(b), the three forces become linearly
dependent. Now the constraint system is only twodimensional and its reciprocal system, that of the platform freedoms, becomes
fourdimensional. And since the constraint system is a planar pencil of rotations, the platform freedom system is spanned by a
translation normal to the plane of the pencil in addition to all the rotations through O. Thus, the end effector gains
a translational degree of freedom.
For this mechanism, the rotation singularities are the usual parallel singularities (Type 2) where no extra freedom is gained
but one of the existing platform freedoms becomes uncontrollable. These singularities can also be found, without wasting napkins,
with some screwsystem observations, which we will leave as a nontrivial exercise for the reader. But before the reader can get
on with the homework, we need to give an extra piece of data. That is, we "forgot" to mention which were the actuated joints.
For our purposes, we will assume that the actuated joints are the revolute joints at the base. (This is feasible. Moreover it
is not a very strange choice, especially if the prismatic joint is replaced with a (fifth) revolute parallel to the two
revolutes fixed in the leg in Fig. 1. We will look at what happens when the prismatic joints are actuated later).
The answer is that, for an RO (i.e., a parallel type, or Type 2) singularity to occur, the three planes of the upper U
joints must intersect in a common line, i.e., they must have linearly dependent normal vectors. It is no accident that this is
similar to the condition for the rotational singularities of the 3UPU mechanism studied in [2]. The condition for the
existence of an instantaneous self motion when the actuators of the robot in Fig. 1 are locked decomposes in two just as was
the case in the Di Gregorio study. Indeed, the constraint system with the actuators locked is decomposed into two systems:
a system of forces through the rotation centre; and a system of moments normal to the upper U joint planes. The two
systems cannot intersect, so singularity occurs when one of them loses rank. In one case the platform has an uncontrollable
rotation, in the other there is a translation unstoppable with the actuators. It is the translational singularity that is a
constraint singularity, while the rotational one is of the plain parallel type.
The fundamental difference between the two groups of singularities is not that the "gained" freedom is a rotation in one case
and a translation in the other. Rather, the uncontrollable freedom is a truly new one at a constraint singularity, and one of
the already present platform motions at a usual parallel singularity.
To see more clearly that the rotationtranslation division is beside the point, let us look at another mechanism, the
3RPS parallel manipulator and two of its singular configurations.
Figure 2: A 3RPS parallel manipulator in (a) a constraint singularity and (b) a Type 2 singularity
The configuration in Fig. 2(b) is an ordinary parallel singularity. The platform has one uncontrollable freedom, namely a rotation
about the line BC, when the prismatic actuators are locked. In Fig. 2(a) we see another singular configuration. Here, the
uncontrollable motion is again a rotation, with an axis passing through both point B and the intersection point of legs A
and C. The second configuration is, in fact, a constraint singularity.
Despite the apparent similarity, the two singularities are quite different. In Fig. 2(b) the constraint wrenches (forces through
the S joints and parallel to the R joints) are linearly independent and the platform has three DOFs. In Fig. 2(a),
however, the three constraint wrenches all pass through point B and lie in the platform plane, and hence form only a
twosystem. The platform gains a freedom and can instantaneously move with 4 DOFs. Indeed, it can rotate arbitrarily about point
B and translate vertically.
From a kinematic point of view these are completely different phenomena. In one case, the loss of control is due to the (so to
speak) unfortunate choice of the actuators. If, in Fig. 2(b), you can actuate the base R joint rather than the P
joint of leg B, you would have full control of the platform. In Fig. 2(a), there is no "fortunate" choice of three actuators
that will prevent the platform from moving instantaneously.
The distinction between the two types is important also because a constraint singularity, whether it is rotational or not, will
generally fail to show up if you look only at the inputoutput equations of the manipulator.
Finally, let us give another example of a constraint singularity of a 3RPS robot (Fig. 3). This time, the platform is
in a much more unperturbed pose than in Fig. 2(a), however, the three base R joints need to intersect in a point.
Figure 3: A 3RPS parallel manipulator in a constraint singularity
5. Constraint Singularities are a Subset of IIM Singularities
It is now clear from the discussion in Section 4 that constraint singularities are in fact IIM singularities. Indeed assuming that
the mechanism is nonredundant, its fullcycle mobility is equal to the mobility of the endeffector. If, at a certain configuration,
the platform gains a degree of freedom the instantaneous mobility of the whole mechanism must have instantaneously increased as well.
As we mentioned in the Introduction, it was remarked in [1] that the translational parallel robot gains a fourth DOF in a rotational
singularity.
Yet, not every IIM configuration is a constraint singularity since it is possible for the mechanism to gain a freedom while
the DOF of the endeffector remains unchanged.
To illustrate this and simultaneously provide what must be the simplest example of a constraint singularity, let us consider the
fourbar linkage in Fig. 4. Assume that the link lengths are such that the mechanism can become flattened, Fig. 4(b). In the shown
configuration, the kinematic chain as a whole acquires an instantaneous mobility of two. Usually the output link of a four bar is
one of the rotors and, therefore, the endeffector will always remain with at most one DOF, including in Fig. 4(b).
Figure 4: A fourbar mechanism in (a) a nonsingular configuration and (b) a constraint singularity
However, let us assume that the output link is the coupler, a kind of "platform" of a 1DOF planar "parallel
manipulator" with two RR legs, one of which is passive. Now, in the singularity of Fig. 4(b) the endeffector gains a freedom
and has two instantaneous DOFs. Indeed, it can now rotate about any point on the line containing all joint centres and, of course,
translate in the normal direction. (Note that this singularity adds both new rotations and a new translation to the platform freedoms.
Is it a rotational or a translational singularity?)
To see that this is a constraint singularity like the ones already described, observe that each RR leg imposes one planar constraint on
the platform: a force through the two joint centres of the leg. Normally, these two lines are distinct and span a 2system. The planar
component of the reciprocal system is the 1DOF system of the coupler freedoms. When the lines of the two constraint forces intersect,
the freedoms are the rotations about the intersection point, Fig 4(a). When the two lines are parallel the coupler must instantaneously translate
in the normal direction.
However, in Fig. 4(b) the constraints become one, more precisely they are linearly dependent and span only a one system. The system
of output motions then becomes the 2system which we already described.
Extra Freedoms and Uncontrollable Freedoms
We saw that at a constraint singularity the platform acquires additional DOFs. Then, obviously, not all of the output freedoms can be
controlled. However this does not mean that it is always a new degree of freedom (i.e., one that exists due to the constraint
singularity) that will be uncontrollable. More precisely, the
redundant endeffector motion present when the actuators are locked may be inside the screw system of platform freedoms which exist
at a nonsingular configuration.
For example, let us look again at the mechanism in Fig. 1(a), but this time, let us assume that the prismatic joints are actuated.
In the configuration shown in Fig. 1(b), as we pointed out, there is a constraint singularity and the platform gains a translational
DOF. Yet, when the prismatic actuators are locked the endeffector cannot translate. The only instantaneously possible motion is a
rotation about a vertical axis through the centre. The translation of the platform can be achieved but only if the actuators can move.
It should be noted that while the system of uncontrollable freedoms or redundant endeffector motions is well defined, a system of
"controllable freedoms" cannot be specified in a unique way. Indeed the redundant freedoms of the platform are the motions possible
when the actuators are locked. It is useful to define a screw system as controllable if it has a zero intersection with the
uncontrollable system and together with it spans the whole system of possible platform motions. Note that we cannot ensure that the
output twist is restricted to such a system, due to the possible presence of transversal redundant motion (so, practically speaking,
we do not have full control).
For example, assuming prismatic actuators in Fig. 1(b), the system spanned by the horizontal rotations through the centre and the
vertical translation is controllable, in the defined sense. Yet we cannot prescribe a motion in that system and enforce it since we
cannot prevent the platform to also rotate about the vertical axis.
In a 3UPU translational mechanism at a rotational constraint singularity we will say that the system of platform translations
is controllable, although we cannot actually control the linear velocity of all points of the body.
In the two examples the mentioned controllable systems are not unique. In fact, any motion that is not a RO motion can be part of a
controllable system.
This understanding on the meaning of the terms controllable and uncontrollable, when referring to the possible
platform motions will allow us to make a clear distinctions between the two constraint singularities illustrated by Fig 1(b) (for
the two different choices of the actuated joints).
In both cases we have a redundant motion of the platform, but only in one case we can
say that we have a redundant output motion. That is so since for a rotational device, which the mechanism is intended
to be, the output motion is not the motion of the platform in general, but rather the rotation of the platform. Hence, In one case,
when the actuators are prismatic, we have a constraint singularity that leads to redundant output, i.e., one of the desired rotational
freedoms is uncontrollable. In the other case, when the first revolute joints are the active ones, the output rotations are a
controllable system. Therefore, rigorously speaking, there is no redundant output motion, and the uncontrollable translation of
the platform can be thought of as a redundant passive motion. Similarly, in the rotational singularities of translational robots
we should not speak of the uncontrollable rotation as a redundant output motion, since rotations were never intended as output,
but rather as an instantaneous passive motion in the mechanism.
Conclusion
A constraint singularity is a configuration of a parallel mechanism with a DOF of the platform, n, lower than the number of
joints in any leg. At such a singularity, the system of the constraint wrenches degenerates and becomes a screw system with a
dimension lower than 6n. As a result, the system of output freedoms instantaneously increases its dimension. At a such a
configuration, both the mechanism as a whole and the platform have at least n + 1 DOFs. The extra degree(s) of freedom of
the platform may or may not be controllable by the actuators.
Bibliography
[1] 
Di Gregorio, R., and ParentiCastelli, V., 1998, "A Translational 3DOF Parallel
Manipulator," Advances in Robot Kinematics: Analysis and Control, J. Lenarcic and M. L. Husty (eds.),
Kluwer Academic Publishers, pp. 4958. 
[2] 
Di Gregorio, R., and ParentiCastelli, V., "Mobility Analysis of the 3UPU
Parallel Mechanism Assembled for a Pure Translational Motion," Proc. 1999 IEEE/ASME International
Conference on Advanced Intelligent Mechatronics, Atlanta, GA, pp. 520525, September 1923, 1999. 
[3] 
Tsai, L.W., 1996, "Kinematics of a ThreeDOF Platform With Three Extensible Limbs,"
Recent Advances in Robot Kinematics, J. Lenarcic and V. ParentiCastelli (eds.), Kluwer Academic Publishers,
pp. 401410. 
[4] 
Zlatanov, D., Fenton, R. G., and Benhabib, B., 1994, "Singularity Analysis of
Mechanism and Robots Via a VelocityEquation Model of the Instantaneous Kinematics," Proc. IEEE International
Conference on Robotics and Automation, San Diego, CA, pp. 986991. 
[5] 
Zlatanov, D., Fenton, R. G., and Benhabib, B., 1994, "Analysis of the Instantaneous
Kinematics and Singular Configurations of HybridChain Manipulators," Proc. ASME 23rd Biennial Mechanisms
Conference, DEVol. 72, Minneapolis, MN, September 1114, pp. 467476. 
[6] 
Hunt, K. H., 1978, Kinematic Geometry of Mechanisms, Oxford University Press. 
[7] 
Karouia, M., and Hervé, J. M., 2000, "A ThreeDOF Tripod For Generating
Spherical Rotation," Advances in Robot Kinematics, J. Lenarcic and M. M. Stanisic (eds.), Kluwer Academic Publishers,
pp. 3954020. 
