High Fidelity Multi Finger Haptic Display Rainer Leuschke Elizabeth K. T. Kurihara Jesse Dosher Blake Hannaford University of Washington, Seattle, USA E-mail:
[email protected],
[email protected],
[email protected],
[email protected]
Abstract The Fingertip Haptic Display (FHD) is a five bar mechanism developed at the University of Washington for haptic interaction with the fingertip of the operator. The twodegree-of-freedom mechanism was computer optimized to accommodate the workspace of the human finger in flexion/extension with a high degree of kinematic isotropy. Direct drive with frictionless flat-coil actuators ensure low torque ripple, inertia, and static friction in the actuation system for high fidelity haptic rendering. This paper describes a redesign of the position sensing arrangement, I/O hardware and software as well as thermal modelling to improve performance and stack four devices for use with index, middle, ring and little finger in a multifinger haptic display. Promising applications are expected to include palpation training for medical personnel, museum displays enabling the visitor to “touch” art, as well as psychophysics research: exploring the limits of human touch perception.
1. Introduction A number of multi finger haptic devices can be found in the literature. Gillespie and Cutkosky [4] developed a virtual piano with haptic feedback. The device is limited to one degree of freedom per finger. Burdea et al. [2] developed a direct drive device using pneumatic pistons on the palm of the users hand. The drawback of the actuator type chosen are low bandwidth and stiffness. Immersion CyberGraspTM [5] is a glove style haptic device for the entire hand that has relatively high friction and low bandwidth. The Fingertip Haptic Display (FHD) is a five bar mechanism developed at the University of Washington for haptic interaction using the fingertip of the operator. A rendering of the device and its workspace are shown in Figure 1. The FHD design uses two flat-coil actuators in the base to achieve low inertia, low friction and good heat dissipation capabilities for rendering virtual objects with a high
Figure 1: Fingertip haptic device and its workspace
degree of realism. The use of flat coils make a narrow device width possible. The unique layout with the coil on the torque output side of the joint axis facilitates a compact base geometry and improved cooling conditions for the coil [12]. The two-DOF device was numerically optimized to accommodate the workspace of the human finger for 95% of the human population in flexion/extension with a high degree of kinematic isotropy [10]. Kinematic isotropy [9], one measure of performance of haptic devices, is ≥0.75 at all points in the workspace. The numerical optimization also penalized link length, which results in a device with low intertia and high stiffness. The FHD has been used in a number of psychophysical studies. Venema investigated haptic perception of first and second degree discontinuities of 2D curves [10]. Dosher, et al., explored small haptic effects and their perception limits as well as the ability of small haptic effects to improve task performance [3]. Icons with attractive forces in the horizontal plane were detectable with peak forces as low as 5 mN and measurably improved task performance as
encoder support
low as 50 mN. These experiments approached the limits of the device in terms of position and force resolution. In the device’s initial implementation, external position encoders extended the thickness dimension and prevented multiple devices from being placed close enough to match human finger flexion-extension planes. While some good results have been obtained with the original device, its true potential is yet to be reached as the original concept was to stack several devices to allow high fidelity haptic interaction for multiple fingers. A redesign of the position sensing arrangement, I/O hardware and software, and thermal modelling enables us to improve performance and stack four devices for use with index, middle, ring and little finger in a multifinger haptic display. This paper describes the engineering of the improved FHD and the mechanical integration of four such devices into a multi-finger haptic device. A new position sensor was specified and the device was modified to fit it inside. Thermal measurements and analysis have clarified the force production capacity of the actuators, and measurements of the human hand have led to a flexible mounting system.
actuated link
scale
main support encoder
Figure 2: CAD model detail of encoder integration
limited by the wave length of the light source and other optical factors. To increase resolution, the analog outputs of the sensing array is sometimes interpolated to obtain intermediate positions. This process is highly sensitive to physical alignment of optical grid, light source and sensor array. Calibration techniques to improve upon the interpolation results are available [11] but do require alignment within reasonable limits. In the 1999 design, these difficulties led to the addition of external optical encoders that are manufactureraligned. Their additional bulk precluded stacking for multi finger use. Miniaturized interferometric encoders have become available in recent years. They can overcome some of the issues that plague conventional designs applied to high resolution sensing. These encoders use a laser diode light source and reflective diffraction gratings (similar to CD/DVD technology) to generate interference patterns. This technique uses a large number of grid lines at any given time to reduce the effect of grid accuracy and can therefore accommodate fine scale graduations and high alignment tolerances. The product selected for this project (MicroE Systems Mercury encoder) fits the space available in the FHD frame with some modifications. The high grid line count and interpolation factors available permit resolutions in excess of the previously achieved specification and even exceed the detection threshold of around 1 µm mentioned above. Table 1 shows some of the specifications for the two encoder models used in the multi finger device (M1500 and M2000) compared to the encoders of the first generation device (CP200). The interferometric encoder used in the FHD has relatively large tolerance requirements (Table 1). To ensure proper alignment of the encoder package, the grating disk
2. Position Sensor Integration Psychophysics research indicates that features as small as 1 µm can be detected by test subjects [6]. Other investigations into limitations of haptic perception indicate detection thresholds below 10 µm depending on the shape of the stimulus [7]. To accommodate experimentation at or near the human perception threshold, we may need device resolution of 1 µm or better. Taking into account the geometry of the five bar linkage of the FHD, joint position measurement with a resolution of 10 µrad or better is required. An additional design goal for the Fingertip Haptic Display was to keep the device narrow enough to allow stacking of several devices for multi finger operation. The voice coil actuator with drive side torque output developed for this device [10] fulfills this requirement. The support with permanent magnets and backing irons attached is approximately 27 mm wide. Some device internal space is available for placement of position sensors. Integration of a suitable modular sensing solution proved to be difficult with the technology available at the time. Meeting all requirements for resolution, robustness and space for the position sensors was not achieved. The conventional optical encoder uses an LED light source, a reflective or transmissive optical grating, a mask and a photo detection array to produce two sine waves phase shifted by 90◦ . These signals can be converted to square waves and the pulses counted with a dedicated IC. With this technique four counts per grid line as well as the direction of motion can be resolved. The grid spacing is 2
lines per revolution interpolation factor counts per revolution resolution accuracy (short range) alignment tolerance θz θy θx z (standoff) y (radial)
CP 200 2048 64 131072 48 µrad n/a n/a n/a n/a n/a n/a
M1500 M2000 4096 4096 40 256 163840 1048576 38 µrad 6.0 µrad ±9.2µrad
to activate and cure the epoxy. The resulting coil is stiff and self contained, with good packing density. It is mounted into the link with a structural thermal cement (Omegabond 600) that facilitates heat transfer to the link. Heat transfer is important because the maximum torque will be limited by resistive heating of the coils and we would like to have the highest possible peak forces. In order to maximize the available torque without overheating we need a thermal model of the actuator. This will enable us to determine maximum steady state heat dissipation and accurately track heating during operation with changing currents. The heat produced in the actuator coil is I 2 R irrespective of torque τ and angular velocity ω.
±2◦ ±2◦ ±1◦ ±0.15 mm ±0.1 mm
Table 1: Encoder specifications
Hin = P = I 2 R
(1)
link A0
The heat produced flows through the thermal system driven by the difference between link temperature (TL ) and ambient temperature (TA ).
#1
coil core
Hout =
#4
coil
TL − TA RT
(2)
where RT is the thermal resistance of the link. In addition, heat is stored in the link, representing “thermal mass” MT . The link temperature can be expressed as Z t 1 (Hin − Hout )dt + TA (3) TL = MT 0
#2
#3
Let T = TL − TA and substitute (1) and (2) to get Z t 1 T )dt T = (P − MT 0 RT
Figure 3: Thermocouple #1 through #4 locations for thermal testing
(4)
Taking the LaPlace transform results in the transfer function for a first order thermal model with time constant RT MT T (s) RT = (5) P (s) 1 + sRT MT
placement as well as the encoder placement are referenced from the same side of the main support. This necessitates an adapter plate that positions the encoder above the grating disk. Fine adjustments of the sensor alignment can be achieved in 4 DOF. The standoff height (z) of the sensor can be adjusted with spacers on the main shaft. Adjustments in x, y and θz of the encoder package are possible by using the oversized mounting holes of the adapter plate. After assembly and testing of some of the sensors, it appears that the encoders were mounted sufficiently accurately without the need for further adjustment.
Figure 4 shows the preliminary results of steady state thermal testing of an actuated link. The link was held in a fixed position and a DC current applied. The temperature of the link was measured at location #1 through #4 as shown in Figure 3 with a type T thermocouple embedded next to the coil. During testing the DC current is ramped up in discreet steps and temperatures are measured when steady state is reached. The thermal limit of the link is given by the maximum operating temperature of the epoxy bonding the coil at 130◦ C. The graph shows a near linear relationship between input power and link temperature in the operating range. Small differences in temperature depending on thermocouple location can be observed. The maximum steady state power was determined to be 20.5 W and thermal resistance RT =5.2◦ /W. Conservatively we rate the thermal resistance of the coil as 7.5◦ /W.
3. Thermal Modelling The FHD is powered by custom designed voice coil actuators with 90◦ of motion range. The coils were manufactured in our lab by winding 150 turns of AWG 28 magnet wire with epoxy coating (MWS epoxy bond) on a bobbin consisting of an aluminum coil core and bobbin plates. After winding, the coil was resistively heated to above 130◦ C 3
140 120
FHD
temperature [°C]
I/O board 100
encoder counters DACs
80 60
USB 2.0
host computer
40
thermocouple #1 thermocouple #2 thermocouple #3 thermal limit
20 0 0
5
10 15 power [W]
20
25
temperature [°C]
120 100 80 60 40 thermocouple #4 thermal limit 200
400 time [s]
600
control VE
limitations of the hardware made it desirable to transition to a newer control platform, using state of the art equipment and software. The central piece of the upgraded solution is a USB interface board in development at the University of Washington. This board is equipped with encoder counters, digital to analog converters and analog to digital converters that can all be accessed and updated from the host via the USB 2.0 interface. Figure 6 shows an overview of the control system under development. The encoder counters have maximum counter width of 24 bits to accommodate high resolution applications, like the FHD. The DACs have a resolution of 16 bits over a ±10 V range. The control loop on the host computer can be run at 1 kHz with very low jitter using the RTAI Linux extension. This hard real time approach guarantees the performance of the controller.
140
0 0
RTAI
HCI graphics data log
Figure 6: I/O schematic for Multifinger Haptic Display
Figure 4: Link temperatures for constant power input, steady state response (3 sensor locations)
20
Linux
800
Figure 5: Link temperature for constant power input (20W), transient response
5. Device performance
To obtain the time constant for the first order model the transient response to a step input is recorded and shown in Figure 5. The input power was set to 20 W and adjustments made to ensure constant power with increasing resistance of the coil due to heating. The time constant for the link was determined to be 160 s and MT =30.6 Ws/◦ . The thermal model can be used for open loop tracking of device temperature to ensure operation within thermal limits.
The generational improvements of the optical encoders, I/O hardware and software and thermal modelling results in a new version of the FHD with potential for very high fidelity haptic interaction. Some of the performance characteristics are summarized in Table 2. To measure the torque characteristics of the actuators, the FHD was mounted horizontally and a force gauge attached to an actuated link. The near constant torque factor of the voice coil actuators can be seen in Figure 7. The torque is ripple free and varies just 7% across the range. Small improvements in packing density of the coil wire result in a slightly higher torque constant for the new devices over the first generation (0.117 Nm/A vs. 0.1 Nm/A). The steady state torque limit is increased by 9% using the thermal modelling data. The peak torque of the original version of the device was limited by current capabilities of the power amplifiers (3 A). The new power amplifier is sized for 6 A currents. With the help of an accurate thermal model it should be
4. USB 2.0 based I/O Hardware The first version of the FHD is controlled by a multiprocessor PC running Solaris [10]. The control loop is running on one processor to ensure sufficient resources, while graphics run on another processor. Loop latency and jitter can not be guaranteed with this solution due to the inherent properties of the general purpose operating system but were experimentally verified. Encoder counters were limited to 16 bits and DAC resolution is 12 bits. The age and 4
150
torque factor [Nmm/A]
125 100
f 75 50 25 0 0
a
experimental polynomial fit 15
30
45 θj [°]
60
75
90
b c
Figure 7: Torque factor over the actuator range
d
actuator torque max steady-state current torque output resolution joint position resolution finger tip position resolution kinematic isotropy
v1999 150 Nmm 1.5 A 0.25 Nmm 48 µrad ≈5 µm
v2004 164 Nmm 1.4 A 0.036 Nmm 6.0 µrad
e
Figure 8: Hand geometry and device positions
≈0.6 µm ≥ 0.75
lines at the center of the fingers (a) are drawn between the metacarpal-phalangeal joint (MCP) and the pad of the fingertip. The hand is oriented such that the line for middle and ring finger are symmetric with respect to the center line of the platform. Line (b) is perpendicular to the middle finger at the MCP. Line (c) is the edge of the hand rest supporting the palm near the wrist and (e) is parallel to (d) at the wrist joint. From this basic geometric information the desired position of the center of each FHD can be obtained by extending the lines (a). The points for 3 different hands as well as their mirror images (accounting for left and right handed users) are shown at (f). Based on this simple examination of hand geometry, two slots are located to roughly match the required de-
Table 2: FHD specifications possible to safely push the FHD past the previous current limits to render force spikes accurately, allowing peak torques of over 600 Nmm depending on duty cycle.
6. Multi-Finger Platform With the integration of the position sensors, the FHD is stackable to provide multi-finger support. Four devices were fabricated to accommodate index, middle, ring and little finger. To design a platform that is adjustable to the variations in hand geometry and finger splay, a basic characterization of hand geometry is needed. The FHD mechanism was optimized for the stochastic reachable workspace of the human fingers [10]. However, stacking the devices side by side with origins aligned, will constrain finger motion to parallel planes at a distance set by the width of the device (27 mm). A comfortable setup will require a small amount of splay since the flexion/extension planes of the fingers are not completely parallel. To quantify the necessary range of adjustments, a small sample of subject hands was scanned with fingers extended comfortably. The hand geometry was measured off the scan and transferred into the CAD model for the platform. Figure 8 shows the platform and geometric relationships to characterize variations between user hands. The four
Figure 9: Multifinger Haptic Display 5
vice center locations. The base of each device is mounted to the platform with two thumbscrews. The FHD can be moved side to side with both screws loosened. The base can be rotated around the front screw when it is tightened and the rear screw loosened. This simple 2 DOF arrangement allows adjustment of the device for a varying amount of splay. The devices can be oriented up to ±12◦ from the centerline. To support the hand of the operator an adjustable hand rest is mounted to the platform. The hand rest can be adjusted in 4 DOF to accommodate different poses and support positions. Figure 9 shows the multifinger haptic display as it will appear in operation.
Arash Aminpour, Hawkeye King and Kenneth Fodero for their help with this project.
References [1] Colgate, J.E., and Brown, J.M., “Factors Affecting the Z-Width of a Haptic Display”, IEEE International Conference on Robotics and Automation, San Diego, California. May 1994, pp. 3205-10. [2] Burdea, G.C., “Force and Touch Feedback for Virtual Reality”, John Wiley & Sons, New York, 1996 [3] Dosher, J., Lee, G., and Hannaford, B., “How low can you go? Detection thresholds for small haptic effects.”, Touch in Virtual Environments, Proceedings USC Workshop on Haptic Interfaces, Prentice Hall, Feb 23, 2001.
7. Discussion After initial fabrication, we noted significant physical damping by working the actuator back and forth with our fingers. We attributed this to eddy current damping in the aluminum coil core. To test this theory, we machined out the insert and noted significantly less damping. Theoretical work has suggested that physical damping in the device itself (even when partially counteracted by negative control damping) benefits stable rendering of high stiffness contact [1]. Translating the findings into usable design guidelines is an issue not addressed well in previous work. To study design implications of device internal damping effects, the core of the actuator coil was machined as shown in Figure 3 to provide as little inherent damping in the core material as possible for low eddy current losses. To increase device internal damping, a number of aluminum inserts will be mounted into the core in future work. The additional material will increase eddy current losses and therefore damping. Haptic interfaces are commonly designed for pen grip based interaction. The multi finger haptic device allows interaction with the fingertips of the user making it suitable for a wide range of applications. The low inertia, low friction, high power and resolution enable high fidelity force rendering and therefore a high quality user experience. The high performance design makes this hardware an ideal testbed for psychophysics research and haptic perception studies. Experiments to explore the effect of device damping, resolution limitations and velocity filtering are planned. Another possible application of this device is in palpation training for medical personnel. Last but not least we are looking forward to using this device for surface exploration of art and other technology demonstrations.
[4] Gillespie, B., and Cutkosky, M., “Interactive Dynamics with Haptic Feedback”, Proceedings of ASME WAM, DSC v49, ASME, New York, pp. 65-72., 1993 [5] Immersion Co., “The CyberGrasp: Groundbreaking haptic interface for the entire hand”, url: http://www.immersion.com/3d/products/cyber grasp.php, 2004. [6] Johansson, R.S., and LaMotte R.H., “Tactile detection thresholds for a single asperity on an otherwise smooth surface”, Somatosens Res., 1983, v1(1), pp. 21-31. [7] Louw S., Kappers A.M.L., and Koenderink J.J., “Haptic detection thresholds of Gaussian profiles over the whole range of spatial scales”, Experimental Brain Research, 2000, v132, pp. 369-374. [8] MicroE Systems, “Mercury Encoders: Alignment of Rotary Scales”, Technical Note, Natick, Massachusetts, url: http://www.microesys.com/pdf/TNAlignment of Rotary Scales.pdf. [9] Stocco, L.J., Salcudean, S.E., and Sassani, F., “Mechanism Design for Global Isotropy with Applications to Haptic Interface”, Proceedings of the ASME Dynamic Systems and Control Division, 1997, pp. 115-122. [10] Venema, S.C., “Experiments in Surface Perception Using a Haptic Display”, Ph.D. Thesis, University of Washington, Department of Electrical Engineering, April, 1999. [11] Venema, S.C., and Hannaford, B. “Kalman Filter Based Calibration of Precision Motion Control”, Proceedings of IROS-95, Pittsburg, PA, August 1995.
8. Acknowledgments
[12] Venema, S.C., Matthes, E., and Hannaford, B., “Flat Coil Actuator having Coil Embedded in Linkage”, U.S. Patent 6,437,770, 2002.
The authors gratefully acknowledge the support from Samsung Advanced Institute of Technology and the National Science Foundation (grant IIS-0303750). The authors would also like to thank Steven Venema, Sean Hoyt, 6
