Proceedings of
A Symposium Commemorating the
Legacy, Works, and Life of
Sir Robert Stawell Ball
Upon the 100th Anniversary of
A Treatise on the Theory of Screws
July 9-11, 2000
University of Cambridge, Trinity College
© 2000
Design of high performance parallel arm robots for industrial
applications.
Torgny Brogårdh,
ABB Robotics
SE-721 68
Västerås
Sweden
Abstract
A method for finding parallel arm robot structures with the same performance as the
FlexPicker robot from ABB is presented. The method is very simple and based on
systematic clustering of the links connected to the actuated platform of the robot. In
the case of 3 DOF actuation, 3 new useful clustering schemes have been found. With
these clustering schemes as a starting point, new parallel arm robot concepts have been
identified. Besides fulfilling the basic properties of the FlexPicker robot, these
concepts will make it possible to design more compact robots, to design robots with
selective tilting movements and to design robots with SCARA type kinematics. These
new design possibilities may be useful for several potential applications of parallel arm
robots in the future.
Key words: Robot, kinematics, parallel, design, assembly, SCARA, DELTA,
FlexPicker.
Introduction.
In 1998 ABB released its first parallel arm robot, the FlexPicker, see figure 1. This
robot is designed for applications needing fast pick and place operations and it can
perform a complete pick and place cycle in 0.4 seconds. The working range is a
cylinder of 1.1 meter in diameter and with a height of 0.25 meters. Maximum payload
is 1 kg, max speed 10 m/s and max acceleration 100 m/s2.
Figure 1: The FlexPicker robot from ABB.
The FlexPicker robot makes use of the kinematics described by Hunt 1973 (1) and the
Delta design invented by Clavel 1985 (2). Since Hunt identified the kinematics by
using screw theory, one could say that the FlexPicker robot has its origin in the
theories made by Sir Ball. The main advantages with the Flex Picker, compared with
for example the competing SCARA robots, are lower weight and higher stiffness of
the arm system. This will make it possible to achieve shorter cycle times at lower
actuator power together with more accurate movements. The price for these
advantages is a relative small working range in relation to the volume of the arm
system. However, there are many applications where cycle time is so important that it
is not a problem to accept the space needed for the arm system, especially since the
robot is mounted above the working envelope, where there is usually unused space.
The most important kinematic features to achieve the FlexPicker performance are:
1. All the actuators are mounted on a fixed platform, which minimises the weight of
the moving arm system.
2. The links connected to the actuated platform transmit only compression and
expansion forces and do not need to be stiff with respect to bending and twisting,
which makes it easy to achieve a stiff and simultaneously low weight arm
arrangement.
3. The joints can be implemented as ball and socket bearings, which makes it
possible to obtain high stiffness and high precision simultaneously with low
weight for the joint arrangement.
4. The actuated platform is positioned in 3 degrees of freedom (x,y,z).
5. The actuated platform is positioned in a parallel fashion without changing its tilt
angles.
From an industrial design point of view it is interesting to know if there are more
structures with the kinematic features of the FlexPicker robot. If that is the case, then it
could perhaps be possible to find useful parallel arm structures with bigger working
range in relation to the arm volume and with more actuated degrees of freedom than
the FlexPicker robot. Thus, a design approach based on systematic clustering of the
links connected to the actuated robot platform has been studied. Based on this design
approach new parallel arm structures have been identified and some new robot
concepts have been found.
The link clustering approach.
Figure 2 shows schematically the basic components needed to achieve the FlexPicker
parallel arm robot with the kinematic features listed above. The actuated platform is
connected to 6 links of type A by means of ball and socket joints having 3 degrees of
freedom (DOF) each. Type A means that the links are designed to be stiff only for
forces along their structure. This force situation in the links of type A is guaranteed
since a ball and socket joint can not transmit bending or twisting torque to the link it is
connected to.
The actuators in figure 2 are mounted on the fixed platform and the moving part of the
actuators are connected to the links of type A via links of type B. The type B links are
designed to be stiff against also bending and twisting torque. All the links of type B do
not need to be connected to actuators, but 3 of them must, otherwise the actuated
platform cannot be manipulated with 3 DOF.
Each of the links of type B (figure 2) can be connected to one or more of the links of
type A. One could say that each link of type B can connect to a cluster of links of type
A and it is possible to introduce a simple clustering scheme, where for example 2/2/2
means that the links of type A are clustered with 2 links to each of the 3 links of type
B. To achieve parallel movements of the actuated platform (to preserve the tilt
angles), type A links belonging to the same cluster must be parallel and have the same
length. Moreover, to avoid a collapsing parallel arm structure because of kinematic
singularities, the placement of the type A link joints on the actuated platform must be
optimised as well as the relative directions between the type A links of the different
clusters.
Fixed platform
Actuators
Links of type A
Links of type B
Actuated platform
Ball & socket joints
Figure 2: Components for the design of structures
with the same features as the FlexPicker
robot.
Now, the 6 links of type A can be clustered in 3 ways: 2/2/2, 3/2/1 and 4/1/1. The
4/1/1 clustering will not fulfil the kinematic demands for a controllable structure and
can be omitted. However, the 2/2/2 and the 3/2/1 clustering according to figure 3 are
kinematically useful. In figure 3 the actuated platform is flat and the type A links are
jointed to the platform surface in a 2D pattern. Using the 2/2/2 clustering scheme for
the design will end up with the Delta structure. The optimised link placement in this
case is achieved when the lines between the joints of each cluster on the actuated
platform have an angle of 120 degrees between each other. The arm structure will
collapse if instead of 120 degrees the angle between two joint lines is 0 (180) degrees.
If instead the 3/2/1 clustering is used for the design of a parallel arm robot, the
placement of the joints of the type A links on the platform surface is not so critical.
The only demand is that the 3 lower joints of cluster 1 are not allowed to be on a
straight line on the platform. The optimum is achieved when these 3 joints of cluster 1
form a triangle with equal side length. This robustness with respect to the link
placement on the actuated platform opens up new possibilities for the robot design.
Cluster 1
Cluster 2
Cluster 1
Cluster 2
Cluster 3
2/2/2 clustering
Cluster 3
3/2/1 clustering
Figure 3: Useful clustering strategies when the links of type A
are attached to the actuator platform in a 2D pattern.
In figure 3 the actuated platform is considered to have a 2D design, which means that
the links of type A connect to the platform surface in a plane. However, the actuated
platform could also be designed as a 3D framework as depicted in figure 4.
This framework does not need to be a cube as in the figure, but the cube drawing
makes it easier to see the configurations of the links. As in the case with a 2D platform
design, there are also in this case 2 useful clustering possibilities for an actuated 3D
platform: 2/2/2 and 3/2/1. The 3/2/1 clustering will have almost the same properties as
for the 2D platform, but the 2/2/2 clustering will have some added features in
comparison to the 2D platform case.
Of course, it could be possible to design the parallel arm structure for 3 DOF
positioning with more than 6 links of type A. However, then the actuated platform will
be over constrained and it will be difficult to achieve a cost effective design. One
example of an over constrained working clustering is 3/2/1/1. In this case the cluster
with 2 type A links will not be actuated but just connected via a type B link to the
fixed platform, where the actuator is replaced by a passive hinge.
Cluster 1
Cluster 1
3D Frame
3D Frame
Cluster 2
Cluster 2
Cluster 3
Cluster 3
2/2/2 clustering
3/2/1 clustering
Figure 4: Useful clustering strategies when the links of type A
are attached to the actuator platform in a 3D pattern.
The clustering approach can also be used to design robots with other than 3 DOF
positioning of the actuated platform. For example, a 2 DOF robot with the FlexPicker
features except feature 4 can be achieved with a 3/3 clustering. One of the 2 clusters
will then be as cluster 1 in the 3/2/1 clustering in figure 3 while the other cluster will
be as cluster 2 in the same figure, but with an added diagonal link mounted in the same
plane as the original cluster 2 links.
Kinematic design principles with 3/2/1 clustering on an actuated 2D platform.
To take care of the 3 links cluster (cluster 1 of the 3/2/1 clustering in figure 3), an
intermediate platform is needed between the fixed platform and the actuated platform.
This intermediate platform will, besides controlling one positioning DOF of the
actuated platform, also control the 2 tilting DOF of the platform. Thus, if only parallel
movements of the actuated platform are desired, then the link arrangement between the
actuator and the intermediate platform must work in such a way that the tilting angles
of the intermediate platform are preserved when the actuator moves the intermediate
platform. For a linear actuator this is achieved automatically and for a rotating actuator
this can be achieved by for example a parallelogram arrangement.
The 2 links cluster (cluster 2 of the 3/2/1 clustering in figure 3) has the responsibility
of controlling a second positioning DOF together with the orientation DOF of the
actuated platform. To be able to do this, the links of cluster 2 are not allowed to be in
parallel with the links of cluster 1. In cluster 2 the upper joints of the 2 type A links
will be connected to a connection bar, and the direction of this bar will control the
orientation of the actuated platform. In the case of a rotating actuator, the orientation
of the actuated platform will be kept constant if the rotation axis of the actuator is in
parallel with the direction of the connection bar, which in turn is parallel with a line
between the joints on the actuated platform.
The 1 link cluster (cluster 3 of the 3/2/1 clustering in figure 3) will control the third
positioning DOF only and there will not be any additional actuator restrictions other
than performing positioning.
Design examples for the 3/2/1 clustering on an actuated 2D platform.
Since the 3 links cluster connected to the intermediate platform mentioned above takes
care of both of the tilting DOF of the actuated platform, the 3/2/1 clustering structure
Actuator 1
Actuator 4
Actuator 2
Actuator 3
Parallelogram
links of type B
Connecting
bar
Intermediate
platform
Links of
type A
Actuated
platform
Ball & socket
joints
Figure 5: Parallel arm robot design using 3/2/1 clustering and a 2D
platform. The fixed platform is not shown in the figure, but all
the four actuators are mounted on the fixed platform.
permits more flexibility in the design than the 2/2/2 clustering. This can be utilised to
achieve a more compact robot than the FlexPicker robot and one possibility is to
design an arm configuration with the arms working in a T structure instead of a Delta
structure. The T-structure will make it easy to put robots close to each other or to have
the robots side mounted (wall or column mounted). The intermediate platform can also
be used to manipulate the tilting angles of the actuated platform selectively and since
the tilting control is kinematically separated from the orientation control, the
orientation can selectively be controlled by changing the orientation of the connecting
bar to the 2 links clustering.
Figure 5 shows one design example for the 3/2/1 clustering. Here the type B links are
configured to form a T-structure (seen from above) and moreover the design includes
the actuation of one tilting DOF. All actuators are of rotating type and they are all
mounted on a fixed platform (not shown in the figure). Actuators 1- 3 manipulate the
position and actuator 4 manipulates one tilt angle of the actuated platform. The link of
type B for actuator 1 is part of a parallelogram. The tilt angle is manipulated by
actuator 4, which changes the form of the parallelogram. For more information about
this design, see (3).
Actuator 3
Fixed platform
Actuator 2
Actuator 1
Actuated platform
Figure 6: SCARA type parallel arm robot design using 3/2/1
clustering and an actuated 2D platform.
Figure 6 shows another design example with a SCARA like parallel arm robot, using
the 3/2/1 clustering. In this case the orientation of the actuated platform will depend on
the outgoing angle of actuator 1, but the tilting angles will be kept constant. As can be
seen from the figure, all 3 actuators are mounted on the fixed platform and the axes of
actuators 1 and 2 are lined up along a common rotation axis. If the type B link of the
single link cluster is actuated via a gear transmission of 90 degrees, the rotation axis of
actuator 3 can also line up with the rotation axes of actuators 1 and 2. Then the whole
robot structure can be rotated around the fixed platform column, giving quite a big
working range in relation to the arm structure volume.
Design example for the 2/2/2 clustering on an actuated 3D platform
The 2/2/2 clustering for an actuated 3D platform will result in a prismatic
configuration of the joints connected to the platform. This configuration will give the
robot designer more possibilities concerning the choice of the arm structure than with
the 2/2/2 clustering for a 2D platform. For example, it will be possible to design a
robot with a T-configuration of the arm system and as can be seen from figure 7 it is
also possible to make a SCARA type robot design with this clustering scheme. The
Actuator 1
Actuator 3
Fixed platform
Type B link in a
parallelogram
Actuator 2
Actuated
platform
Figure 7: SCARA type parallel arm robot design using 2/2/2
clustering and an actuated 3D platform.
actuated 3D frame is shown as a cube in figure 7, but of course only a fraction of the
cube geometry will be used if a real robot is designed. Actuator 1 controls the cluster
on the cube diagonal and is mainly responsible for the vertical movements of the
actuated platform. Actuators 2 and 3 will mainly change the distance between the
actuated platform and the fixed platform column. To preserve the orientation of the
actuated platform the link of type B for actuator 3 is designed as a parallelogram. If
the orientation is not important, for example in the case when a fourth rotating
optional axes is used as in the FlexPicker case, the parallelogram can be omitted and
the same type B link structure as in figure 6 can be adopted.
Applications for the new robot structures.
Since the robot structures put forward in this paper have some new features compared
with the FlexPicker, then not only the robot engineers but also the application
Figure 8: Possible future assembly cell using a robot structure
based on 3/2/1 link clustering.
engineers will probably find new design possibilities. The compact T-structure can for
example be used to have 2 robots working in a pair above a conveyor to increase the
production rate. The possibility to selectively actuate one tilting DOF may give new
opportunities when implementing assembly tasks. However, the most interesting new
feature is probably the possibility to have a FlexPicker robot working as a SCARA
type robot, especially if the actuators are lined up to be able to rotate the whole arm
system around a fixed platform column. With this design the robot has the possibility
to work with several conveyors and feeders placed around the robot, as outlined in
figure 8. In this figure a SCARA type parallel arm robot with the 3/2/1 clustering on an
actuated 3D platform is used. To be able to control the orientation of the objects, a
fourth rotating actuator is placed on the actuated platform. By using appropriate
transmission means the actuator could of course be placed on the fixed platform,
compare the arrangement for the optional fourth axis in the FlexPicker case.
The robot in figure 8 is thought to perform vision guided assembly operations on
components delivered by 3 conveyors. The assembly is made directly on one of the
conveyors and the assembled parts are placed on an outgoing conveyor. This is just
one example of how a SCARA like parallel arm robot could be used and it could
perhaps be possible to increase the production rate in an existing production line just
by replacing conventional SCARA robots used today with its parallel arm cousins.
Summary.
Using a link clustering approach it has been possible to derive new parallel arm
structures with the same features as the FlexPicker robot. Moreover, these new
structures have some optional properties, which will make it possible to design more
compact parallel arm robots, to include selective tilting or rotation of the actuated
platform and to design parallel arm robots with SCARA like kinematics. These new
features may make it easier to obtain a future broader use of parallel arm robots in
industry.
References:
(1)
Hunt, K.H., Constant-velocity shaft couplings: a general theory. Trans. Am.
Soc. Mech. Engrs. 95, Series B (Journal of Engineering for Industry),
455- 464 (1973): See Table 2, Class C.
(2) Clavel, R., Dispositif pour le déplacement et le positionnement d'un élément
dans l'espace. Patent CH 672 089 (1985)
(3)
Brogårdh, T., A device for relative movement of two elements.
Patent WO 97/33726 (1996)