Sie sind auf Seite 1von 6

Mechanisms for Haptic Feedback

Raymond Hui, Alain Ouellet, Andrew Wang, Paul Kry, Stefan Williams,
George Vukovich, Walter Peruzzini

Directorate of Space Mechanics


Canadian Space Agency
6767 route de 1ACroport
St-Hubert (QuCbec)
Canada J3Y 8Y9
duir@sp-agency.ca>
Abstract
This article describes work in progress at the Canadian
Space Agency on the design and implementation of haptic
devices. Haptic devices are a special class of robotic
mechanismsfor which structural transparency is a
foremost design criterion. Also notable is the fact that
ofen. three or four degrees of freedom, rather than six as
in general robotic tasks, are suficient for many haptic
applications. Furthermore, in order to make these
devices readily available to many users, it is necessary
their kinematic models be sufficiently simple such that
they can be controlled by inexpensive means. Various
three and four-dof mechanisms, some of which recently
developed by the Canadian Space Agency, are discussed
herein in terms of their suitability for haptic applications.
Forfive or six-dof applications, the concept of the virtual
handle is introduced, reducing the problems and
complexity usually associated with mechanisms with a
high number of degrees of freedom.

1. Introduction
Traditionally, robotic mechanisms have been used to carry
out autonomous tasks, or as master or slave devices in
teleoperation. In particular, various force-reflecting hand
controllers have been designed for remote control of slave
manipulators - see for example [7]. It has recently been
realized that force reflection also allows an operator to
explore and feel a remote or virtual environment - see
for example [2, 3,4, 8. 1 I]. A mechanism used in this
mode is typically referred to as a haptic device, or
alternatively a manipulandum. Thus far, applications
including virtual pushbuttons and potential fields for
computer interface - i.e. a force-reflecting mouse - have
been developed. Tremendous growth in this area within
IEEE International Conference
on Robotics and Automation
0-7803-1965-6/95 $4.0001995 IEEE

the next few years is anticipated. The key elements to the


design of an efficient and effective haptic device include
structural transparency and control facility.
Structural transparency is more important and
desirable in the design of a haptic device than in slave
manipulators in general because the main function of such
a device is to provide tactile and force feedback; hence it
is important not only that the mechanism be
backdriveable, but also that, as much as possible, the
operator discern only the mechanical impedance being
reflected or represented, and not that of the structure of
device. This design criterion emphasizes the importance
of reducing mass and inertia, as well as backlash and
friction, wherever possible. Other factors such as
workspace, payload capacity, etc. which usually come
under consideration in the design of robotic mechanisms
are important but in this case not so much as the
characteristics already mentioned. Positional accuracy is
similarly secondary to fidelity of force feedback.
Furthermore, certain designs lead to non-complex
kinematics and decoupled or nearly decoupled dynamics
which is also highly desirable.
Given the general design criteria as discussed above,
it can be concluded that the conventional fully serial
configuration is inferior for a force-reflecting haptic
device because of the necessity to actuate remote joints.
Motors and gear-reducers constitute the main volume and
mass of the joints of a robot, unless cabled transmission is
used to locate the actuators at the base; in the latter case,
the complexity of cable-routing increases with each
degree of freedom.
In parallel mechanisms, the actuators can often be
kept at the base and in many instances grounded (i.e. only

- 2138 -

the shafts are non-stationary), leading to a lower static


load and inertia. The distribution of loads in addition to
elimination (or at least reduced use) of highly mobile
cantilevered beams provide parallel mechanisms greater
strength and rigidity as well as lighter mass and thus
higher bandwidths than their serial counterparts, provided
the design is of high precision. The main drawbacks of
parallel mechanisms are limited workspace, complexity of
design if a high number of degrees of freedom is required
(usually, the number of universal and spherical joints
required increases, thus increasing friction and backlash),
and complicated dynamic models and forward kinematics
(only special configurations of parallel mechanisms have
closed-form solutions for forward kinematics). In
conclusion, the advantages of parallel mechanisms satisfy
the design criteria of a force-reflecting haptic device but it
is important to make use of them such that their
disadvantages do not hamper the performance of a haptic
device.

In the ensuing sections, various three and four-dof


mechanisms are described. Three or four degrees of
freedom are often sufficient for haptic applications so long
as the handle provides passive dofs for the hand or
finger(s) to be re-positioned. For applications requiring a
higher number of dofs, a hybrid design involving directdrive and cabled actuation, as well as the concept of the
virtual handle are introduced.

19941; here, a third actuator pivots about the vertical axis


so as to align the vertical four-bar mechanism with the tip
of the horizontal five-bar mechanism; noticeable inertia is
hence expected as an operator moves the device sideways.

Fig. 1

Fig. 2

The third actuator can be grounded by using a


spherical elbow as in Fig. 2 (the spherical elbow is
achieved by three non-parallel simple joints with
intersecting axes). The resulting design is nonetheless not
completely satisfactory as the entire five-bar linkage acts
as a cantilever beam due to motion in the vertical plane.
The resulting bending stress can be eliminated for
downward motion by supporting the tip of the five-bar
mechanism on a surface; however, there is no simple way
of achieving the same effect for upward motion.

2. Three-Dof Mechanisms
The five-bar mechanism is the best known parallel
mechanism in robotics, having been popularized by the
work of Asada et al. [I]. In general, its forward
kinematics is rather complex for a two-dof device. This
can be simplified if the two grounded actuators are coaxial and the lengths of opposite links are equal. The
resulting forward and inverse kinematics are both
straightforward. The dynamic model is rather simple in
comparison to that of a general five-bar mechanism, and
is potentially decoupled and free of nonlinear terms. It is
clear that simple dynamics leads to facility of control, a
quality which obviates the need for great computational
power and hence makes a given mechanism more
accessible to popular use.
The simple dynamics and relatively low inertia of the
five-bar mechanism are limited to the plane. We set off to
extend this planar design to three dimensions and at first
obtained the designs in Figs. 1 and 2. The design in Fig.
1 is in some ways contrary to the conventional approach
of adding a third joint to rotate a five-bar mechanism
about a vertical axis, e.g. the Phantom [Sensable Devices,

--

Fig. 3

Fig. 4

The Delta mechanism invented by Clavel (US patent


no. 4,976,582, 1990) [ IO] with alterantive embodiments
shown in Figs. 3 (only one branch is shown to minimize
drawing complexity) and 4 is an interesting positioning
structure with grounded actuators. It consists of a base
and an end-effector platform connected by three skewing
parallelogram sub-mechanisms. On one hand, the design
is very interesting in that the end-effector platform
remains parallel to the base at all times, which makes it a
useful positioning mechanism for manufacturing
applications. On the other hand, the closed-form forward
kinematics solution is not straightforward [ 101.

To varying degrees, the above mechanisms meet the


criterion of near structural transparency for haptic

2139 --

applications. On the other hand, their forward kinematics


solutions are not simple even when a closed form
solutions exists - see [lo] in the case of Delta. We
continued to pursue the idea of extending the five-bar
mechanism to three dimensions and eventually arrived at
the nearly di-tetrahedral design in Fig. 5 . This design
consists of three or more branches each made up of a
primary link connected to a secondary one by a spherical
elbow. The primary links are driven, preferably directly,
by rotary actuators while the secondary links are
connected to the end-effector by simple revolute joints.
It turned out that this design has already been
disclosed in relatively unknown patent (US patent
4,65 1,589, 1987). In addition, a similar mechanism based
on linear actuation has been the subject of previous
studies 161. It is surprising the nearly di-tetrahedral form
in Fig. 5 has not received greater attention in the
literature for, as discussed below and in greater detail in
[9], the model of the mechanism becomes greatly
simplified in special cases.

is also greatly simplified (although not necessarily


simple), as each secondary link is the reflected image of
the primary link with which it is contiguous. In fact, this
reduction in complexity of kinematics and dynamics is
possible whenever there is congruency between the
primary and secondary structures, regardless of the
relative lengths of the primary links, and the relative
positions and orientations of the actuators.
We find this simplicity in modeling an important
advantage of the di-tetrahedral mechanism over the Delta
mechanism which has the advantage of a constantly
horizontal end-effector - an important property in certain
pick-and-place robotic tasks but can be replaced by a
pivoting handle in a haptic device as illustrated in Fig. 6.

Fig. 6 The first CSA di-tetrahedral haptic device


Fig. 5 The nearly di-tetrahedral mechanism
In general, the forward kinematics solution of the
mechanism in Fig. 5 is not simpler than that of other
parallel mechanisms such as the Delta. However, there
are two cases for which the model reduces to a very
simple form. The first case is the design in which the
secondary stage comprised of the secondary links and the
end-effector structure is congruent to the primary one
comprising the primary links and the actuators. The
result is that the two structures are mirror images about
the plane formed by the three spherical joints.
Determination of the end-effector's position and
orientation is then reduced to finding the image of the
actuation base which can be selected as the origin without
loss of generality. Since the plane can be specified in
terms of the controlled angles, the position of the endeffector can be similarly expressed. The dynamic model

Fig. 7 Two congruent tetrahedrons stacked together


such that they are mirror images
We use the true di-tetrahedral form in Fig. 7
(although it is not obvious, this ideal design can be
realized in practice as shown in Fig. 8) to elucidate the
simplified kinematics of this mechanism. To find the
image of the base vertex about the dividing plane, it is

- 2140 -

first necessary to find the vector normal to the plane


which is the normalized cross product of any two vectors
making up the plane. Then the tip of the mechanism can
be located by projecting one of the primary links onto this
normal vector, and multiplying the result by two. More
specific details are provided in [Peruzzini et al., 19951.
An alternative solution to the position of the ideal ditetrahedral mechanism is based on the intersection of the
secondary links at a common point. This point can be
located by finding the intersection of the three spheres
centred at the three spherical elbows with the secondary
links as radii - a relatively straightforward procedure.
Such a solution to the position of the di-tetrahedral
mechanism is possible only if the secondary links indeed
intersect at a common point. One mechanism which
satisfies this condition is shown in Fig. 8. In this design,
the distance between each spherical elbow and a reference
point on the end-effector as indicated by a broken line is
fixed, but the secondary links are offset from these virtual
lines. In the case of a nearly di-tetrahedral design as in
Fig. 5 and 6, the primary and secondary stages have to be
congruent so that the equally simple reflection method
can be used to find the end-point position.

which have four degrees of freedom. The one shown in


Fig. 9 was invented by Hayward under contract the
Canadian Space Agency [3]. It allows control of all three
rotations as well as a translation along the axis of the roll
rotation. This mechanism is ideal as a wrist structure.

Fig. 9 The Hayward mechanism


The mechanism in Fig. 10 is a different type of fourdof mechanism; it allows positioning of its tip in threedimensional space with the end-effector able to rotate
about the vertical direction. It is more suitable than the
Hayward mechanism above for haptic applications in
which positioning control is more essential (to provide
feedback of the surface of a virtual or remote object).

Fig. 10 The Compass mechanism


Fig. 8 The design of a true di-tetrahedral mechanism
In Lamberts patent, the di-tetrahedral mechanism is
described as one which provides control of one
positioning and two orientation degrees of freedom. We
have thus far discussed it as a positioning structure. In
fact, it can be used in either mode, but not both
simultaneously. As a haptic device, a pivoting handle
should be added to make the mechanism more ergonomic.

3. Four-Dof Mechanisms
The mechanisms introduced above all have three degreesof-freedom. We discuss in this section two mechanisms
- 2141

The operational principle is based on controlling the


tips of the vertical compass-like structure (hence the name
Compass) to desired positions with five-bar mechanisms
(or any other mechanism which can move the tips in the
horizontal plane). The height of the end-effector is
determined by the distance between the two tips.
Displacement of the end-effector in the horizontal plane
can be effected by the moving the tips together in the
same direction by the same distance. Rotation about the
vertical is achieved by moving the tips such that their
distance from the projection of the tip to the base plane
remains fixed - i.e. effectively drawing a circle. The

passive universal joints at the legs of the compass to rotate


as necessary while remaining in a vertical plane.
It is straightforward to find the position of the endeffector of the four-dof Compass mechanism. The
positions of the tips of the two five-bar mechanisms can
be found from their respective pair of controlled base
angles. Finding the position and orientation after that is a
simple exercise in trigonometry.

4. Implementation Issues
It is important to point out that for the above mechanisms
the orientation of the end-effector is either only partially
or not at all controllable. Nevertheless, this is not
necessarily a disadvantage for applications in remote
control and haptic interface because passive joints can be
placed at the tip such that the orientation of the
controlling hand or finger can be varied to suit the
comfort of the operator without changing its position.
This is quite natural for haptic applications since it is
common for humans to change the orientation of a finger
or probe while exploring the surface of an object.
We felt initially that even with a light structure, it
was necessary to compensate for the weight of the links of
the mechanisms described above so that the operator does
not feel it. However, extensive tests with the ditetrahedral mechanism in Fig. 6 demonstrated that our
design was sufficiently light that the operator barely felt
its weight. In fact, fatigue usually arose due to the lack of
support for the operator's arm rather than the weight of
the mechanism. Attempts to compensate for its weight
resulted in poor performance of the haptic device because
the cogging torque in the brushless motors used led to
"preferred" positions.

In addition to its barely noticeable weight, the design


of the di-tetrahedral mechanism proved to be a success in
other aspects of structural transparency mentioned above.
The mechanism "feels" virtually devoid of backlash,
friction and inertia. The only noticeable backlash is found
in the commercially obtained universal joint mounted at
the base of the handle to allow it to pivot. This handle is
further flawed due to the frequency with which it finds
itself in a singular position. A custom design solution has
been found.

5. Virtual Handle
One motivation for designing only three or four-dof forcereflecting mechanisms is that six-dof mechanisms are
inevitably complex. Whether serial or parallel, the

kinematic and dynamic models of a mechanism increases


at a faster than linear rate as the number of degrees of the
freedom increases. Furthermore, a totally parallel six-dof
mechanism usually has limited range of motion. On the
other hand, there are many force-reflection applications,
including the traditional master-slave manipulation tasks,
which require more than three or four degrees of freedom.
One way to provide six degrees of freedom to a hand
controller or haptic device without incurring mechanical
and modeling complexity is to define a virtual handle
using two or more three or four-dof devices, controlling
the thumb and index finger respectively. Although
multiple mechanisms have been previously used in haptic
applications [ 121, they have been used to grasp virtual
objects rather than to simulate a single virtual hand
controller, as suggested herein. In the theory of grasping,
it is known that two point contacts with friction can
provide five instantaneous degrees of freedom to a
grasped object. The missing rotation about the line
joining the two points of contact can be effected with a
third mechanism or if each grasping finger has a fourth
degree of freedom. By applying this principle to the
virtual handle, the thumb and index finger can be made to
feel as if they are gripping the handle of a force-reflecting
mechanism which has five or six degrees of freedom. The
complexity of the models of two three or four-dof
mechanisms, and the difficulty in controlling them, are in
general much more tractable than dealing with a six-dof
mechanism, particularly if the latter is based on a parallel
design.
*
\

/
/
\ ,,.....--........._,_
i......__..._.___...
i /
J

Fig. 10 The virtual handle concept (haptic devices not


shown for simplicity)
The virtual handle can be simulated by defining its
geometry in virtual space corresponding to the workspace
of the mechanisms involved. The operator can approach
and grip it, perhaps guided by graphical representation on
a computer screen. The feeling of gripping the handle is

2142 -

provided by force reflection exerted on each digit by the


corresponding mechanism. Since only forces are
necessary, a three-dof mechanism such as the ones in the
di-tetrahedral class can be used. Either digit can move
along the surface of the handle, or off it, to locate and
press the available virtual buttons which can enhance the
functionality of the system. This concept can be used to
simulate any force-reflecting mechanism which can be
gripped between two digits. Additional digits can be
included in the grasp by providing more mechanisms,
although this may introduce the problem of interference of
linkages.

References

Another interesting idea is to control the output


forces of the mechanisms such that as the operator moves
hisher fingers, the shape of a manipulator can be felt, and
hence the operator can move it link by link. This is
potentially useful in controlling redundant manipulators.
More details on the applications and implementation of
the virtual handle are presented in [ 5 ] .

Asada, H. and Youcef-Toumi, K., Direct-Drive Robots


Theory and Practice, Cambridge, MA: MIT Press, 1987.
Hayward, V., Choksi, J., Lanvin, G.and Ramstein, C.,
Design and Multi-Objective Optimization of a Linkage
for a Haptic Interface, Proc. of the 4th Workshop on
Advances in Robot Kinematics, Ljubljana, Slovenia, 1994.
Hayward, V. and Reynier, F., Report on the Kinematic,
Dynamic and Structural Design for the Development of a
Miniature Hand-Controller with Force and Tactile
Feedback Capability, report for Canadian Space Agency
contract no. 9F009-1-1441/01-SR, 1992.
Hirata, Y. and Sato. M., 3-Dimensional Interface Device
for Virtual Work Space, Proc. of the IEEWRSJ Int. Con$
on Intelligent Robots and Systems, Raleigh, NC, USA,
1992, pp. 889-896.
Hui, R., The Virtual Handle Concept, submitted to
IROS, 1995b.

6. Conclusions
Herein, two basic classes of mechanisms are introduced.
They are respectively useful as three and four-dof forcereflecting mechanisms. The design of these mechanisms
is based on structural transparency and simplified
kinematics. Structural transparency allows the operator to
feel for the most part only the force reflection determined
by the software driving the mechanism, and not its
weight, inertia and friction. Furthermore, a structurally
near-transparent mechanism can be controlled effectively
without modeled-based compensation. Simplified
kinematics allows the mechanism to be controlled with
computational overhead affordable by a standard PC
(control rates of 1.5 kHz or higher has been achieved with
the nearly di-tetrahedral mechanism in Fig. 6 and a 48666 PC).
The virtual handle is for applications requiring five
or six degrees of freedom. The main advantage of this
concept is that the complexity, costs and decrease in

Lee, K. and Shah, D., Kinematic Analysis of a Three


Degrees of Freedom In-Parallel Actuated Manipulator,
Proc. of IEEE Int. Con$ on Robotics and Automation,
1987, pp. 345-350.
McAffee, D. A. and Fiorini, P., Hand Controller Design
Requirements and Performance Issues in Telerobotics,
Proc. of International Conference on Advanced Robotics,
Pisa, Italy, 1991, pp. 186-192.
Millman, P. and Colgate, J., Design of a Four Degree-ofFreedom Force-Reflecting Manipulandum with a Specified
Forcemorque Workspace, Proceedings of the IEEE Int.
Con$ on Robotics and Automation, Sacramento, CA,
USA, 1991, pp. 1488-1493.
Peruzzini, W., Oueilet, A. and Hui, R., Analysis of the
Di-Tetrahedral Mechanism, to be presented at IFToMM,
1995.
Pierrot, F., Robots Pleinement ParalltYes Ugers:
Conception, Modilkation et Commande, Doctoral Thesis,
Universit6 Montpellier 11, 1991.

performance associated with designing six-dof


mechanisms can be avoided. From this idea, it can be
seen that other virtual mechanisms can be similarly
simulated.

SensAble Devices Inc., Brochure of the Phantom ForceReflecting Haptic Interface, Vanceburg, Kentucky, 1994.

We are currently conducting experiments with both


the nearly di-tetrahedral mechanism shown in Fig. 6 and
the Compass mechanism in Fig. 10. The detailed design
of a true di-tetrahedral mechanism has recently been
completed.

- 2143 -

Yoshikawa, T., Zheng, X.and Moriguchi, T., Display of


Operating Feel of Dynamic Virtual Objects with Frictional
Surface, Proc. of Int. Con$ on Intelligent Robots and
Systems (IRQS), Munich, Germany, 1994, pp. 731-738.

Das könnte Ihnen auch gefallen