`DSC-Vol. 55-1, Dynamic Systems and Control:
`Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems
`Volume 1
`Jan 1994 pp 295..301
`ASME 1994
`
`The PHANToM Haptic Interface:
`A Device for Probing Virtual Objects
`
`Thomas H. Massie and J. Kenneth Salisbury.
`Department of Mechanical Engineering
`Massachusetts Institute of Technology
`Cambridge, Massachusetts
`
`1. Abstract
`This paper describes the PHANToM haptic interface - a device which measures a user’s finger tip position and
`exerts a precisely controlled force vector on the finger tip. The device has enabled users to interact with and feel a
`wide variety of virtual objects and will be used for control of remote manipulators. This paper discusses the design
`rationale, novel kinematics and mechanics of the PHANToM. A brief description of the programming of basic shape
`elements and contact interactions is also given.
`
`2. Introduction
`A dominant focus in robotics research labs has traditionally been the development of autonomous systems - those
`which operate without human supervision or interaction. However, robotic systems which are under direct human
`control have begun to enjoy a resurgence of interest in recent years, in part due to advances in robot and human
`interface technologies. These new interactive systems (telerobotic) promise to expand the abilities of humans, by
`increasing physical strength, by improving manual dexterity, by augmenting the senses, and most intriguingly, by
`projecting human users in to remote or abstract environments. In this paper we focus on our work to develop a
`means for interacting with virtual mechanical objects; this is an important stepping stone toward the development of
`enhanced remote manipulation systems in which simultaneous interaction with real and virtual objects will be
`possible.
`
`At the MIT Artificial Intelligence Laboratory, we have been developing haptic interface devices to permit touch
`interactions between human users and remote virtual and physical environments. The Personal Haptic Interface
`Mechanism, PHANToM, shown in Figure 1, has evolved as a result of this research (Massie, 1993). The
`PHANToM is a convenient desktop device which provides a force-reflecting interface between a human user and a
`computer. Users connect to the mechanism by simply inserting their index finger into a thimble. The PHANToM
`tracks the motion of the user’s finger tip and can actively exert an external force on the finger, creating compelling
`illusions of interaction with solid physical objects. A stylus can be substituted for the thimble and users can feel the
`tip of the stylus touch virtual surfaces. By stressing design principals of low mass, low friction, low backlash, high
`stiffness and good backdrivability we have devised a system capable of presenting convincing sensations of contact,
`constrained motion, surface compliance, surface friction, texture and other mechanical attributes of virtual objects.
`
`3. Three Enabling Observations
`Three observations influenced the basic design of the PHANToM. The first observation established the type of
`haptic stimulation that the device would provide, the second determined the number of actuators that the device
`would require and the third established the volume or workspace that the device would possess.
`
`1. Force and motion are the most important haptic cues. A significant component of our ability to “visualize”,
`remember and establish cognitive models of the physical structure of our environment stems from haptic
`interactions with objects in the environment. Kinesthetic, force and cutaneous senses combined with motor
`capabilities permit us to probe, perceive and rearrange objects in the physical world. Even without detailed
`cutaneous information (as with a gloved hand or tool), the forces and motions imparted on/by our limbs and
`fingers contribute significant information about the spatial map of our environment. Information about how an
`object moves in response to applied force and the forces which arise when we attempt to move objects can
`provide cues to geometry (shape, locality, identity), attributes (constraint, impedance, friction, texture, etc.) and
`events (constraint, change, contact, slip) in the environment. Unlike other sensory modalities, haptic interactions
`permit two-way interaction via work exchange. Controlled work can be performed on dynamic objects in the
`environment and modulated to accomplish tasks.
`
`2. Many meaningful haptic interactions involve little or no torque. Perhaps the most significant design feature of
`the PHANToM is the passive, 3 degree-of-freedom “thimble-gimbal”, shown in Figure 2. The decision to use the
`
`295
`
`Valve Exhibit 1045
`Valve v. Immersion
`
`
`
`thimble-gimbal was based on the observation that many finger tip interactions with the environment involve little
`or no torque about the finger tip. (Tightening a screw with one’s fingernail is on of the few clear counter-
`examples.) Because the three rotations about the center of the finger tip are neither measured nor actuated by the
`PHANToM, the user’s finger tip can be modeled as a point or frictionless sphere in the virtual environment. The
`same argument applies for the tip of a stylus - the tip of a sharp pencil or pen touching a surface has virtually no
`torque exerted on it by the surface. Introducing three passive freedoms with the "thimble-gimbal" greatly
`simplifies programming as well as mechanism design.
`
`3. A small wrist-centered workspace is sufficient. Many meaningful haptic interactions occur within the volume that
`the finger tip spans when the fore-arm is allowed only limited movement. In order to determine the most suitable
`workspace for a haptic interface, a wooden mock-up consisting of a 3 degree-of-freedom kinematic chain with a 3
`degree-of freedom thimble-gimbal was constructed. Through experience with the mock-up, it was decided that
`the PHANToM should be constructed such that a user could move the wrist freely without encountering the edges
`of the workspace. The size of a mouse pad, a sheet of note-book paper and the computer keyboard are common
`examples of this scale of haptic workspace.
`
`4. Three Necessary Criteria for an Effective Interface
`The following three criteria are necessary for an effective force-reflecting haptic interface device. Independent
`psychophysical testing could establish specifications for each of the three criteria, however available actuator,
`sensor, material and computer technology will ultimately determine the degree to which each of the criteria can be
`met. Furthermore, the three criteria must be considered simultaneously, as improving the specification for one will
`necessarily degrade the specifications for the other two. The PHANToM represents an effort to balance these three
`criteria to achieve an effective, affordable, force-reflecting haptic interface with existing technologies.
`
`1. Free space must feel free. Users must not be encumbered by the device. That is, the device should exert no
`external forces on a user moving through free virtual space. Translated into engineering requirements, this means
`that there should be little back-drive friction, low inertia at the human-machine interface and no unbalanced
`weight. For the PHANToM, we arrived at values for each of these attributes that were perceivable, yet not
`distracting. Static back-drive friction for the PHANToM is less than 0.1 Newton (Nt), inertia is such that the user
`perceives no more than 100 grams of mass at the interface and unbalanced weight is less than .2 Nt for all points
`within the workspace.
`
`2. Solid virtual objects must feel stiff. One metric of a force-reflecting interface is the maximum stiffness of the
`virtual surfaces that it is capable of representing. Because no structure or control loop is perfectly stiff, each
`virtual object compliance is not limited by the stiffness of the structure, but rather by the stiffness of stable control
`that can be achieved. Using the current control algorithm, the PHANToM can reflect a maximum stiffness of
`about 35 Nt/cm, We have found that most users can be convinced that a virtual surface with a stiffness of at least
`20 Nt/cm represents a solid, immovable wall. The maximum obtainable stiffness depends not only on the natural
`frequencies of the device but also on the resolution of the sensors and actuators and the servo rate.
`
`3. Virtual constraints must not be easily saturated. There is nothing as disturbing as leaning against a wall and
`falling through it. In the virtual world, walls should be solid. The maximum exertable force for the human finger
`is on the order of 40 Nt (Sutter, 1989), but during precise manipulation we find that people rarely exert more than
`10 Nt of force, the peak maximum for the PHANToM. In fact, the time average force exerted during normal
`operation is on the order of 1 Newton, while the maximum continuous force capability for the PHANToM is
`about 1.5 Nt.
`
`5. PHANToM Mechanics
`In its simplest form, the PHANToM can be thought of as a transmission between three DC brushed motors with
`encoders and the human finger tip a shown in Figure 3. The x, y and z coordinates of the user’s finger tip are
`tracked with the encoders, and the motors control the x, y and z forces exerted upon the user. Torques from the
`motors are transmitted through pre-tensioned cable reductions to a stiff, lightweight aluminum linkage. At the end
`of this linkage is a passive, three degrees of freedom gimbal attached to a thimble, Figure 3. Because the three
`passive rotational axes of the gimbal coincide at a point, there can be no torque about that point, only a pure force.
`This allows the user’s finger tip to assume any comfortable orientation. More importantly, because the user can be
`represented by a single point of friction-less sphere within the virtual environment, collisions and resulting
`interaction forces within the virtual environment are easily calculated.
`
`296
`
`
`
`The PHANToM has been designed so that the transformation matrix between motor rotations and endpoint
`translations is nearly diagonal. Decoupling the three motors produces desirable results in terms of back-drive
`friction and inertia. For a haptic interface with perceivable inertia and back-drive friction, it is important that the
`friction and inertia be nearly constant in all directions to minimize the distraction they create for the user (i.e. well
`conditioned inertia matrix and small, non-disparate friction components) (Vertut, 1986).
`
`An interesting design feature of the PHANToM, seen in Figure 1, is that two of the three motors move in such a
`manner as to counterbalance the linkage structure. Because the PHANToM is statistically balanced, there is no need
`to compromise the dynamic range of the device by actively balancing the structure with biased the motor torques.
`Conveniently, the first rotational axis of the PHANToM is located directly above the wrist of the user. This permits
`aligning the inherently spherical workspace of the mechanism with similarly spherical wrist. The complexity of the
`cable reduction mechanism is minimized by using a single cable to "mesh" two motor capstans with another pulley.
`This minimizes mechanism width and tensioning difficulty.
`
`6. Virtual Worlds
`The generation of haptic cues to create virtual objects requires the ability to 1) track motion of the user, 2) detect
`collision between the user controlled probe (virtual finger tip) and the virtual objects, 3) compute reaction forces in
`response to contact and motion and, 4) exert forces on the user. Because 1, 2 and 4 are relatively easy with the
`PHANToM we have been able to focus on the development of rules or control laws for 3 which generate a wide
`variety of interaction sensations.
`
`In general we use a one-sided Hooke’s law relationship to simulate walls. Walls may be combined in a number of
`ways to give the sensation of polyhedral objects. To date, the interiors and exteriors of rectangular solids have been
`created. Figures 4 through 7 illustrate some of the effects which must be dealt with at surfaces, corners, and edges.
`Because of the unusually low friction in the device these surfaces feel distinctly slippery. Spherical surfaces have
`been created as illustrated in Figure 8. Variations in surface stiffness of both of these shapes may be used to give a
`wide range of "feel" to the objects. Regular, local variations in surface geometry have been used to simulate coarse
`textures.
`
`It is also possible to give these objects body properties such as mass, permitting objects to be pushed and bounced
`off the walls of the virtual environment. It has been shown to be relatively easy for users to distinguish between
`massive and low-mass objects by feel alone. By giving bodies stiffness relative to ground, virtual buttons have also
`been created which simulate the "fall-away" feeling of real switches as shown in Figure 9.
`
`One of the more subtle effects to simulate is friction. Simply implementing the classic Coulomb or stiction models
`leads to unstable behavior or flawed sensations due to limitations in resolution and time step size inherent in
`sampled data systems. A method recently developed in our lab permits stable and convincing friction sensations to
`be generated even in the face of these limits.
`
`Interestingly, the contact force state can be used to control additional actions. In the case of a virtual painting
`demonstration the contact force was used to permit intuitive control of the brush width of a virtual paintbrush.
`
`7. Perceptual Observations
`Initially, there was some concern as to how a user would adapt to using his or her finger tip to manipulate a single
`point in virtual space. The action is quite intuitive, with one exception. When feeling the outside of virtual spheres,
`some users are disturbed by the "PHANToM" effect. That is, when using the device, one’s hand can physically pass
`through the volume occupied by a virtual sphere, while only the finger tip is constrained to remain outside of the
`virtual sphere. Some users are quick to use this phenomena to their advantage and begin probing all sides of virtual
`objects, unconstrained by the volume of their hands.
`
`In one sense, representing the finger tip as a point within the virtual environment effectively increases haptic
`resolution beyond that of a finger tip in the real world. Cutaneous receptors aside, when a person manipulates and
`probes in the real world the effective resolution limited by the finger volume. However, using the PHANToM, the
`width of a user’s finger tip in the virtual world can be made as small as data quantization permits. Manipulation
`remains intuitive, though, because the thimble-gimbal allows the small virtual point to be spatially located within the
`user’s finger tip.
`
`297
`
`
`
`A few common haptic experiences with the PHANToM seem to evoke strong reactions from users. One such
`experience is sliding along the top of a virtual block that one cannot see, becoming comfortable with the fact that the
`invisible block will support one’s finger and then inadvertently sliding off the top of the block. Several users find
`this falling sensation very powerful. Another experience that users find pleasantly disturbing occurs when one has
`been probing an invisible virtual environment for several minutes only to have the virtual environment disappear
`without warning. The feeling is similar to sitting down in a chair only to find that it has been pulled out from
`beneath you!
`
`Users of the PHANToM provide evidence that our visual, haptic and auditory senses are closely linked and that all
`three sensory modes are required for navigation within virtual environments. Even without visual feedback, many
`users claim that they can "see a sphere" after touching a virtual sphere with the PHANToM. Also, users describe the
`non-linear force characteristics of the virtual buttons they touch as "feeling the buttons click," even when no sound
`is present.
`
`Even though solid objects within the virtual environment are slightly compliant, users are often willing to accept that
`the objects are solid. Perhaps users tolerate this amount of compliance because it is on the order of the compliance
`of the human finger pad. Also, the fact that users can effortlessly slide tangent to the walls seems to re-enforce the
`illusion of a solid surface.
`
`8. Conclusions
`The development of the PHANToM device has demonstrated the feasibility of a relatively low-cost system which
`can provide convincing sensations of interactions with virtual object. The relative ease with which users can learn to
`use the device and immediately begin perceiving and rearranging virtual objects suggests that we have crossed an
`important performance threshold. Performance which permits distinct sensations of free space and constrained
`motion results from a proper balance of mechanism properties such as friction, inertia, force, resolution and
`bandwidth. It is an important research question as to how this balance scales with the size of the interface
`workspace. Larger versions of the PHANToM are under development and will help in determining the appropriate
`balance of performance qualities needed at new scales.
`
`The PHANToM is currently in use in several labs at MIT as well as a number of government and industrial research
`labs. We expect in the near future to see demonstrations of multiple finger interactions and multiple finger
`interactions and multiple user interactions in shared workspaces with the device. Our own work will focus on the
`stability and programming issues which arise when two fingers grasp objects to perform assembly tasks as well as
`use of the device to permit tool interactions such as screwdrivers and pliers.
`
`9. Acknowledgements
`This work was supported in part by DOD National Defense Science and Engineering Graduate Fellowship, NTSC
`Contracts N61339-93-C-0083, N61339-93-M-1961, N61339-93-M-1961 and ONR/URI Grant N00014- 92-J-1814.
`In addition we would like to acknowledge the input of Dr. David Brock and Craig Zilles at the AI Lab for
`contribution to the development of a number of haptic effects.
`
`10. References
`Brooks, T.L. and A.K. Bejczy, 1985, "Hand Controllers for Teleoperation," Technical Report JPL Publication 85-11,
`Jet Propulsion Laboratory, Pasadena, CA.
`
`Durlach, N.I. et al, 1992, "Virtual Environment Technology for Training," Virtual Environment and Teleoperator
`Research Consortium (VETREC), MIT, BBN Report 7661.
`
`Massie, T.H., 1993, "Design of a Three Degree of Freedom Force-Reflecting Haptic Interface ," SB Thesis,
`Department of Electrical Engineering and Computer Science, MIT.
`
`McAffee, D.A. and P. Fiorini, 1991, "Hand Controller Design Requirements and Performance Issues in
`Telerobotics," Proceedings of International Conference on Advanced Robotics (ICAR), Pisa.
`
`Minsky, M., and M. Ouh-young, O. Steele, F.P. Brooks, and M. Behensky, 1990, "Feeling and Seeing: Issues in
`Force Display," Computer Graphics, Vol. 24, pp. 235-243.
`
`298
`
`
`
`SensAble Devices, Inc., 1993 "The PHANToM," literature from SensAble Devices Inc. 225 Court St., Vanceburg,
`KY 41179.
`
`Sutter, P.H. J.C. Iatridis and N.V. Thakor, 1989, "Response to Reflected -Force Feedback to Fingers in
`Teleoperations," Proc. Of the NASA Conf. On Space Telerobotics, NASA JPL.
`
`Vertut, J. and P. Coiffet, 1986, Robot Technology, Teleoperation and Robotics: Evolution and Robotics:
`Application and Technology, Vol. 3B, Prentice Hall, Englewood Cliffs, NJ.
`
`299
`
`
`
`
`
`
`
`300
`
`PERELEUTA TLON 48roece oe|AOsrrov
`
`FORCE =f(POSTION,
`veLocry,
`FTE?
`
`Figure 3: Block diagram of control system.
`
`
`Figure 1: The PHANToM haptic interface.
`
`x
`
`PULSES
`
`ANGLES
`
`
`
`User Point Violating Virlu alWali
`
`Crogs Seclion of Sphere
`
`Force is Normal to Surface
`
`c
`sh
`
`\ L ~ User-pointviolatesvoluniecfsphere
`
`Figure 4: Simulation of virtual planes. Force, F = Ka, is
`exerted normal to plane when the user pushes into virtual
`surface.
`
`Plane 1 Forces sum nicely 10 creale the correct restoring force
`
`Plane 2
`
`Figure &: Forces generated by contact with sphere. As
`with planes, force is proportional to penetration depth
`and in direction normal to surface. Because force is al-
`ways normal, spheres feel very slippery.
`
`
`
`
`“— Force ais away in this regionlo gre a “hapli¢ click”
`=
`
`Wall when the bitlon baloms oul
`
`Sutton Displacement
`
`Figure 5: Forces at interior corner defined by 2 planes.
`Forces sum properly to create correct restoring force when
`planes meet at obtuse angle.
`
`Figure 9: Virtual buttons. Nonlinear force-deflection
`curve for button demonstration showing “fall-away” force
`characteristic.
`
`
`
`Figure 6: Virtual Cubes. Figure at left illustrates reason-
`able force vectors for cube with slightly complhant sur-
`face. For the point shown in figure at right, it is not clear
`which of forces 1, 2 and 3 should be exerted. This is path
`
`dependent.
`
`11 Appendix 1:
`PHANToMSpecifications
`
`Force resolution:
`Noininal spatial resolution:
`Peak force:
`Continuous force:
`Backdrivefriction:
`Max/iniu force (Dyn. range):
`Inertia at tip:
`Workspace:
`Max. object stilliess
`
`[2 Dat
`ANQ dpi
`10 Nt
`£.5 Nt
`O.1 Nt
`|
`100:
`LOU gi
`Sx l7 x 25cm
`35 Nt/ow
`
`Figure 7: Solution to corner problem with cubes. Divid-
`ing cube into regions shown provides simple solution to
`path ambiguity problem. In 3-D, regions are pyramid in
`shape and permit stable behavior at edges and corners. If
`large forces are exerted at corners, probe point may move
`into the adjacent region and be pushed off object giving
`the sensation of “plucking” the corner.
`
`301
`
`TM
`
`
`
`