human interaction with small haptic effectsbrl.ee.washington.edu/eprints/220/1/rep173.pdf · 1.2...

16
Jesse Dosher* Blake Hannaford Department of Electrical Engineering BioRobotics Laboratory University of Washington Seattle WA 98195-2500 USA *Correspondence to [email protected] Presence, Vol. 14, No. 3, June 2005, 329 –344 © 2005 by the Massachusetts Institute of Technology Human Interaction with Small Haptic Effects Abstract This research investigates the ability of subjects to detect small haptic effects and the associated gains in task performance with various configurations of hap- tic stimuli. Variations in force amplitude, shape, and pulse duration used to cre- ate the effects are studied. An adaptive-threshold method is used to obtain subjects’ detection thresholds for actively explored haptic icons ranging in size from 3 to 5 mm. Detection thresholds are compared for smooth versus rough actively-explored icons. Subjects’ detection thresholds for “static icons” (force pulses of 50 to 150 ms duration) are also measured. Results indicate that rough (sawtooth) haptic icons are more easily detected by a human subject than smooth (sinusoidal) icons of the same size. Transient vibrotactile cues may con- tribute to these observations. Mean subject performance, as measured by Fitts’ information-processing rate and by clicks per minute, is shown to improve with the amplitude of haptic stimulus. 1 Introduction Haptics is the study of information exchange through the tactile and kin- esthetic modalities. To use a metaphor, a haptic display, or device, is to touch what an optic display, such as a monitor, is to vision. A typical haptic display measures the position and velocity of the end point of the user’s limb or digit and applies force to that limb or digit based on the interaction in a virtual envi- ronment. As an engineering field, haptics encompasses four main areas: the electromechanical system, the mathematical object interaction model, the hap- tic rendering algorithm, as well as many human factors. A new and interesting direction of haptics research involves small-scale hap- tic devices. Small haptic devices could provide a better mouse substitute for laptop computers or augment the ability to fully utilize advanced personal dig- ital assistants (PDAs) and cell phone features. These applications, however, severely restrict the weight, power consumption, and volume of the device. Other applications, such as automotive GPS navigation instruments, or even climate and sound systems (systems that all tend to take drivers’ eyes off the road) could benefit from haptic controls (Burnett & Porter, 2001). Inevitably, a haptic device that meets the strict weight, power, and volume requirements of a handheld device will be capable of producing only very small forces and displacements compared to existing desktop devices. This project sought to quantify the weakest haptic effects that a subject can detect with a practical haptic device, the Fingertip Haptic Display (FHD; Ven- Dosher and Hannaford 329

Upload: others

Post on 31-Jan-2020

11 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

Jesse Dosher*Blake HannafordDepartment of ElectricalEngineeringBioRobotics LaboratoryUniversity of WashingtonSeattle WA 98195-2500 USA*Correspondence [email protected]

Presence, Vol. 14, No. 3, June 2005, 329–344

© 2005 by the Massachusetts Institute of Technology

Human Interaction with SmallHaptic Effects

Abstract

This research investigates the ability of subjects to detect small haptic effects

and the associated gains in task performance with various configurations of hap-

tic stimuli. Variations in force amplitude, shape, and pulse duration used to cre-

ate the effects are studied. An adaptive-threshold method is used to obtain

subjects’ detection thresholds for actively explored haptic icons ranging in size

from 3 to 5 mm. Detection thresholds are compared for smooth versus rough

actively-explored icons. Subjects’ detection thresholds for “static icons” (force

pulses of 50 to 150 ms duration) are also measured. Results indicate that rough

(sawtooth) haptic icons are more easily detected by a human subject than

smooth (sinusoidal) icons of the same size. Transient vibrotactile cues may con-

tribute to these observations. Mean subject performance, as measured by Fitts’

information-processing rate and by clicks per minute, is shown to improve with

the amplitude of haptic stimulus.

1 Introduction

Haptics is the study of information exchange through the tactile and kin-esthetic modalities. To use a metaphor, a haptic display, or device, is to touchwhat an optic display, such as a monitor, is to vision. A typical haptic displaymeasures the position and velocity of the end point of the user’s limb or digitand applies force to that limb or digit based on the interaction in a virtual envi-ronment. As an engineering field, haptics encompasses four main areas: theelectromechanical system, the mathematical object interaction model, the hap-tic rendering algorithm, as well as many human factors.

A new and interesting direction of haptics research involves small-scale hap-tic devices. Small haptic devices could provide a better mouse substitute forlaptop computers or augment the ability to fully utilize advanced personal dig-ital assistants (PDAs) and cell phone features. These applications, however,severely restrict the weight, power consumption, and volume of the device.Other applications, such as automotive GPS navigation instruments, or evenclimate and sound systems (systems that all tend to take drivers’ eyes off theroad) could benefit from haptic controls (Burnett & Porter, 2001).

Inevitably, a haptic device that meets the strict weight, power, and volumerequirements of a handheld device will be capable of producing only very smallforces and displacements compared to existing desktop devices.

This project sought to quantify the weakest haptic effects that a subject candetect with a practical haptic device, the Fingertip Haptic Display (FHD; Ven-

Dosher and Hannaford 329

Page 2: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

ema, 1999; Venema, Matthes, & Hannaford, 2000).1

In these experiments we studied a basic attractive“icon”—a region of space accompanied by a local at-tractive force field. Using a forced-choice protocol withan adaptive threshold-finding algorithm, we determinedthe minimum amplitude for the haptic effect that wasdetectable 71% of the time. This threshold is a com-bined human-machine property. To estimate the effectof the haptic device itself on the results, we comparedthis threshold with selected mechanical properties of thedevice (Dosher, 2001).

A successful haptic interaction involves a haptic devicein contact with the human operator in a bidirectionalexchange of information. In contrast to the elegant ex-periments of pure psychophysics, in a haptic interactionthe stimulus is not independent of the user inputs. Ourmeasurement depends on the specific haptic device aswell as on the user, and is not a threshold in the classicalpsychophysical sense. Instead we measure a data pointby which we evaluate technology possibilities in an im-portant new application area.

1.1 Literature Review

Many psychophysical experiments have been con-ducted on the sensitivity of human touch. Studies onperceptual acuity and Braille have contributed well-designed techniques and adaptive-threshold algorithmsto the study of human sensory perception (Stevens,Foulke, & Patterson, 1996). Other psychophysical ex-periments have quantified human spatial (Moy, Singh,Tan, & Fearing, 2000; Louw, Kappers, & Koenderink,2002) and surface-texture perception (Lederman,Klatzky, Hamilton, & Grindley, 2000). However, thesestudies measure the threshold of human perception,which may differ from the useful haptic threshold. Un-like psychophysics experiments that seek to remove, asmuch as possible, the effects of the mechanism used, weseek to discover a combined human-machine hapticthreshold level, as well as a level for useful haptic inter-actions. Few experiments have been performed investi-

gating human-machine detection thresholds for hapticeffects.

Utilizing a force-feedback mouse, Hasser et al. (Has-ser, Goldenberg, Martin, & Rosenberg, 1998) reportedperformance gains of 61% in a computer GUI targetingexperiment, and a reduction of targeting errors of 70%.Hasser et al. (Hasser, Martin, & Dennerlein, 2000) alsoreported a 52% speed improvement in a “tunnel” steer-ing task—analogous to navigating a computer GUImenu system. These experiments utilized a consumerforce-feedback pointing device (mouse) with force out-puts of nearly 1 N, and did not investigate the effects ofusing smaller scale haptic effects (both force and size).

Psychological experiments show us that properly ex-ploiting multimodal interactions and haptic illusions canenhance or attenuate haptic effects (Akamatsu, MacKen-zie, & Hasbrouc, 1995; Srinivasan, Beauregard, &Brock, 1996). Increasing the effect of a small hapticdevice without raising its output (i.e., power cost) mayalso aid in development of low-power haptic devices.Goldfarb and Kilchenman (2001) studied the effects offorce saturation and system bandwidth on the quality ofhaptic interactions. They concluded that “haptic inter-face hardware may be capable of conveying significantperceptual information to the user at fairly low levels offorce feedback and system bandwidth,” (p. 1387). Theyfound that system bandwidth above 40 Hz, and end-point forces above 3 to 4 N did not provide further per-formance enhancements in a size-identification task.These experiments investigated levels of force 10 to 100times greater than the force outputs we investigated.

In prior research, the FHD was used to explore hu-man perception of first- and second-order surface dis-continuities as might be found in CAD models (Ven-ema & Hannaford, 2000). This research examined howthe ability to perceive and locate first- and second-ordersurface discontinuities is affected by both the magnitudeof the discontinuity and the chosen set of control gainsused to haptically render the discontinuity. The me-chanical design of the FHD was also discussed.

Dennerlein and Yang (2001) report that when manyicons have force feedback, the presence of “distrac-tors”—haptic icons that a user has to move through toget to a target icon—can cause user frustration and in-creased effort in completing a task. They suggest that a

1. Portions of this work have appeared in Dosher, Lee, and Han-naford (2001) and Dosher and Hannaford (2001).

330 PRESENCE: VOLUME 14, NUMBER 3

Page 3: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

more elegant activation of specific force fields should beused to reduce this interference. Akamatsu and Mac-Kenzie’s (1996) work implies that motor-sensory band-width may suffer in multimodal interactions in desktopenvironments with the overuse of force feedback. Oak-ley et al. (Oakley, Brewster, & Gray, 2001) showed thatdynamically varying the level of haptic stimulus in adesktop menu-navigation environment can increaseoverall productivity while providing a subjectively lesstaxing condition as compared to having full haptic forceat all times.

Results such as these lead us to conclude that it is notoptimal, in all cases, to provide large amounts of forcefeedback in human-computer interfaces. Insight intohow users perceive small haptic effects, and how perfor-mance varies with haptic stimulus, may therefore behelpful in designing better haptically enabled human-machine interfaces, as well as lightweight, low-power,small haptic devices.

1.2 Fitts’ Law

Fitts’ law (Fitts, 1954) was used in this study toevaluate the results of the task-performance experiment.Fitts devised an experiment where subjects rapidlytapped a stylus between two alternate targets, and ex-pressed movement time (MT) in terms of target width,W, and movement amplitude, A, as

MT � a � b log2�2AW �, (1)

where a and b are empirical constants used for data-fitting in other applications of Fitts’ law. We termed therate at which a subject performed a task as the Fitts In-formation Processing Rate (FIPR) for that trial, whichcarries the units bits per second (bps). Rearranging Fitts’Law, we write

FIPR �

log2�2AW �

MT (2)

where A is the movement amplitude from one target tothe next, W is the icon width (5 mm) and MT the time

it took the subject to move from the previous target tothe current target.

Our performance experiment was similar to a Fitts’law task. In Fitts’ (1954) original experiment, subjectstapped a stylus between two targets as rapidly as possi-ble. In our experiment, subjects “tapped” on randomlyselected haptic icons. Our target width was constant;however, the distance between subsequent targets couldbe any one of four values. Our task fell somewhere be-tween a purely experimental Fitts’ law task, and a morerealistic “point-and-click” task, as would be found in atypical human-machine interface.

2 Methods

All subjects who participated in experiments readand signed an approved human subjects consent form.The Human Subjects Review Committee ApplicationNumber is 01-8235-E01.

2.1 FHD—Fingertip Haptic Display

The Fingertip Haptic Display (Figure 1) is a 2DOF planar haptic device, designed in the BioroboticsLaboratory, with extensive kinematic optimization thatcouples a lightweight, rigid five-bar linkage with directdrive, flat-coil actuators (Venema, 1999; Table 1). Itsdesign encompasses the index finger flexion/extensionworkspace of 95% of the human population (Venema &Hannaford, 2000). About 2.5 cm thick, the FHD isdesigned to be stacked into a four-fingered device that,when completed, will allow the four fingers to be inde-pendently curled or stroked over a virtual surface.

2.2 FHD Software

The software controlling the FHD hardware is runon a quad-processor 200 MHz x86-based PC, runningSolaris. The FHD control code is interrupt based, withthe control loop running at 1000 Hz. All code is writ-ten in the C language.

Previous research (Venema & Hannaford, 2000)found control gains that were optimal for exploringfirst- and second-order surface discontinuities of Kp � 1

Dosher and Hannaford 331

Page 4: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

N/mm and Kd � 3 mN � s/mm. These control gainswere used in all experiments reported in this paper. Thevirtual plane was implemented with a PD control law.

F(y) � ��Ky�y � 100 mm� � by, y � 100 mm

0, y � 100 mm(3)

2.3 Experiment Setup

In all of the experiments, we first created a virtualhorizontal surface with the FHD. The surface was at thesame location for all experiments (Figure 1, y � 100mm). The haptic icons used for a particular experimentwere then located on the horizontal surface. In all ex-periments, subjects wore noise-reducing ear protectionto reduce possible sound cues.

Two forms of haptic cues were used for all experi-

Figure 1. The FHD planar mechanism was optimized to fit finger workspace of 95% of the population. The axes of the FHD’s

workspace has been overlaid, along with a representation of the virtual plane at y � 100 mm.

Table 1. FHD Specifications

Joint torque220 Nmm continuous,450 Nmm peak

Torque output resolution 0.25 NmmPosition resolution 0.0027 degreesWorkspace Approx. 120 � 120 mmKinematic isotropy � 0.75

332 PRESENCE: VOLUME 14, NUMBER 3

Page 5: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

ments: those that were designed for active exploration,and those that were designed for passive (static) interac-tions. All of the actively explored icons (Figure 2) werebased on either a 3-, 4-, or 5-mm sawtooth or 5-mmsinusoid waveform. Sinusoidal force was tangential tothe virtual plane, and a function of displacement fromthe icon’s center, x

F�x� � �F sin�2�

��x � xo�� (4)

The force at each extreme edge was zero, then gradu-

ally increased to its peak value at (x � xo) � ��

4 . Five

mm was chosen as the largest icon size based on theavailable workspace of the FHD, the need to evenlyspace up to five icons, and for consistency with the ideaof “small” haptic devices and workspaces.

In the detection-threshold experiments the iconswere presented in pairs. For example, a subset of trialsduring an experiment would consist of a pair of 4-mmwide sawtooth icons, located on the virtual horizontalsurface, with a center-to-center spacing of 20 mm. Dur-ing a trial, both icons would have basic visual feedback,but only one randomly assigned icon would have hapticforces associated with it. The recorded force value asso-ciated with each trial was the peak force in the iconwaveform.

2.3.1 Adaptive Threshold. The adaptive-threshold method is designed to adjust the stimuluslevel (force output) such that a 71% correct response isobserved from the subject (Stevens et al., 1996). Sub-jects were asked to provide a response based on the per-ception of a force during each trial. Responses were

Figure 2. Three sawtooth-shaped icons, sized 3-, 4-, and 5-mm, and one sinusoidal-shaped icon, 5 mm

in width. Forces were created tangentially to the horizontal axis, pointing inward to the icon’s center.

Force varied with displacement from the icon’s center based on the chosen waveform, creating the

impression of a small divot. All experiments began with peak force levels of 500 mN.

Dosher and Hannaford 333

Page 6: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

deemed to be correct or incorrect based on properlydetermining the presence or absence of a force for agiven trial. Using the adaptive-threshold method, twoor more correct responses decrease the stimulus, while asingle incorrect response increases the stimulus back toits previous value. To obtain more data in the region ofinterest (near subjects’ thresholds), the stimulus levelchanged by 10% for the first 12 trials, while all remain-ing trials changed by 5%. During the experiments, sub-jects held a pushbutton in their opposite hand. Eachtrial ended when the button was pushed.

During the trials, each deviation from the currenttrend in stimulus was termed a stimulus path reversal. Asan example, a subject may give correct responses for thefirst 30 trials, where the force levels are still appreciablyabove his/her threshold level for detection. This resultsin a continuous decrease in the level of stimulus; how-ever, when the subject gives an incorrect response(occurring more frequently as their threshold is ap-proached), this causes the adaptive-threshold algorithmto increase the force back to its previous value, causing abreak in the decreasing trend of the stimulus. Thesewere termed path reversals, as the subject was viewed astraversing a stimulus path, uniquely determined by theirparticular responses during an experiment.

The experiment continued until 12 such stimuluspath reversals occurred. This method is also referred toas an adaptive staircase (Wall & Harwin, 2000) algo-rithm, or an adaptive 2-down, 1-up procedure (Dhruv &Tendick, 2000).

We computed each subject’s detection threshold us-ing the force values from the last four stimulus-path re-versals created by their responses. By the end of the ex-periment, subjects reached their detection-thresholdlevel, and the level of stimulus oscillated closely aroundthat value (Figure 4, �50� trials). The detectionthresholds we report are the mean of all subjects. Thismethod was used for all experiments except task perfor-mance.

2.4 Experiments with ActivelyExplored Icons

For detection-threshold experiments we used pairsof 5-mm icons, placed 15 mm center-to-center on the

virtual surface. The icon width and displacement werechosen based on the workspace of our device and thedesire to study the potential usefulness of small hapticdisplays with compact workspaces. During each trial,one randomly assigned icon would have a force associ-ated with it, consisting of a sawtooth waveform. A sim-ple LED display located next to the FHD indicatedwhen subjects were inside of either icon A or B, butgave no indication of the shape of the icon, or magni-tude of the force.

With their finger in the FHD, the subject would restgently on the virtual surface while sliding back and forthto explore the two icons. The subject’s wrist rested on aplatform coincident with the virtual plane. Subjects wereallowed to sample both targets indefinitely before indi-cating which target they perceived to contain a force.Subjects indicated their choice by moving into the tar-get and pressing a button held in their opposite hand.This was repeated until completion of the adaptive-threshold procedure.

Six subjects were used, 5 male and 1 female, rangingin age from 22 to 45. All were right-handed except onemale. All subjects were associated with the electrical en-gineering department. Subjects reported no known sen-sory or motor impairments.

We conducted a second set of experiments, using thesame subjects, which sought to determine how shape(roughness) affected detection thresholds for hapticicons. Using a setup identical to the previous experi-ment, we presented subjects with pairs of 5-mm wideicons. However, this time the associated waveform was asinusoid. The detection-threshold results for thesesmooth sinusoidal icons were then compared to that ofthe rough (sawtooth) icons of the same width.

Another set of experiments sought to determine howicon width affects detection thresholds. The same sub-jects repeated the first experiment (described above)three times, using pairs of either 3-, 4-, or 5-mm saw-tooth icons. The detection thresholds were then com-pared for the different icon widths.

2.5 Static Pulses

We also conducted a set of experiments with sin-gle tangential force pulses of 50, 100, or 150 ms. This

334 PRESENCE: VOLUME 14, NUMBER 3

Page 7: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

experiment sought to investigate the potential useful-ness of an impulse-only haptic display. Pulse lengthswere chosen that corresponded to the stimulus lengthsoccurring in the active-exploration experiments at thetypical range of exploration velocities. We termed thesestimuli static icons.

During the experiment, subjects rested their fingermotionless at a specified location. The wrist was sup-ported at the same height and the fingertip was sup-ported by the same virtual plane as in the active-exploration experiments. Subjects were informed thata pulse would occur randomly with each trial, andthat not all trials would contain a pulse. After theirresponse, the next trial would begin after a randomdelay of between 1 and 3 seconds. Both the randomspacing of trials and the assignment of zero-magni-tude trials was done to eliminate subjects’ indicatingthe perception of a pulse simply because they ex-pected one each trial.

During the experiments, pulse amplitude started at500 mN and was varied based on the same adaptive-threshold algorithm used in the active-exploration ex-periments. The experiment continued until completionof the adaptive-threshold procedure. To reduce possibleeffects of learning on the analysis, half the subjects werepresented with an ascending sequence (50, 100, 150ms), and half received a descending sequence (150,100, 50 ms) of pulse durations. Unlike the first experi-ment, in which the stimulus was reset to 500 mN foreach icon width, subjects continued at the present forcelevel when the pulse-duration level was changed, andthe stimulus output tended toward a new thresholdlevel, associated with the change in pulse duration.

Six subjects were used in this experiment, 5 male and1 female, ranging in age from 25 to 45. All were right-handed and 4 of the 6 had participated in the previousexperiments. Subjects reported no known sensory ormotor impairments.

2.6 Task Performance

The second category of experiments sought todetermine performance gains through the use of haptics

in a basic pointing task. These experiments utilized 13subjects (2 female, 11 male, ages 22–42). Five of thesubjects had participated in the previous experiments. Inthis experiment, we created five virtual icons evenlyspaced across the same virtual horizontal surface. Allicons were 5-mm wide with 15-mm center-to-centerdisplacements. Movement amplitude was one of fivepossible values: 15 n, where n � 1,2, . . ., 4 mm. In ourperformance measurement experiments, the targetwidth, W, was always 5 mm. Movement time, MT, wasrecorded (via software) by measuring the number ofmilliseconds between trials.

Utilizing a set of five LEDs, each icon had crude vi-sual feedback in the form of a constantly lit LED wheninside its corresponding icon. During each trial, a ran-domly assigned “target” LED would flash rapidly. Sub-jects were instructed to move their finger as quickly aspossible from their present location to the location ofthe target LED and push a button held in their oppositehand once inside the target. To provide visual feedback,the target LED, as with all icons, would glow steadilywhen the subject was inside the virtual icon. Upon com-pletion, the next trial would begin and a new targetwould flash. Consecutive targets never consisted of re-peated locations, that is, movement amplitudes werealways nonzero between trials. Haptic feedback waspresent only on the target icon.

Unlike the previous experiments involving activeexploration, this experiment consisted of timed trials.Preexperimentation indicated that between 25 and 75trials were necessary to eliminate the effects of subjectvariability in learning curve. The experiment beganwith 100 “warm-up” trials immediately followed byfour contiguous sets of 100 trials each. During thewarm-up trials, the target icon had a force (at the ex-treme edges) of 200 mN, using a sawtooth waveform.Subjects were then presented with constant force lev-els of either 0, 50, 100, or 300 mN in blocks of 100trials. These stimulus levels were selected to provide ameasure of baseline performance (no haptics), perfor-mance near the threshold value (50 mN), twice thatvalue (100 mN), and six times that value (300 mN).Subjects took between 10 to 12 min to complete all500 trials.

Dosher and Hannaford 335

Page 8: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

2.7 Data Analysis

2.7.1 Detection-Threshold Experiments. Thefirst experiment involved a nontimed, forced-choiceprotocol, with subjects choosing between one of twohaptic “icons,” created by a sawtooth waveform. At theconclusion of the experiment we computed a singledetection-threshold value (in mN) for each subject bytaking the mean of the last four path reversals. Themethod of the second experiment was identical, exceptfor the use of a (smooth) sinusoid waveform. Thesepopulation means are shown in Table 2.

A two-way analysis of variance was computed on themeasured detection thresholds for the 6 subjects andthree sizes of rough icons, while a pairwise t test wascomputed between the measured detection thresholdsfor the 5-mm rough versus smooth icons.

Analysis of variance was also computed on the detec-tion thresholds for the static icon experiment involving50-, 100-, and 150-ms duration pulses.

2.7.2 Performance Measurement Experi-ments. We computed performance levels in two ways.The first took an information-processing view of humanmovement performance, based on Fitts’ law. We usedthe equation

FIPR �

log2�2AW �

MT (5)

where FIPR stands for the Fitts’ Information ProcessingRate. We defined movement amplitude, A, to be thecenter-to-center distance between targets. The FIPRvalue (in bps) reported for each subject is the meanFIPR value of all 100 trials at each haptic stimulus level(0–300 mN).

Subject performance was also computed as the num-ber of trials, or clicks per minute (CPM). The CPMvalue for a subject at a given level of haptic stimulus wascomputed by dividing the total time taken for all trialsby 100 (the number of trials at a given force level).

Analysis of variance was computed on CPM and FIPRperformance results, followed by pairwise t tests.

3 Results

3.1 Detection Thresholds for ActivelyExplored Icons

We plotted the motion, velocity, and forces in atypical exploration during the two-target experiment(Figure 3). The top trace shows position versus time inmilliseconds for the exploration of two icons. Each icon,centered at �110 mm and �90 mm in the FHD work-space, is represented by a pair of horizontal lines in thefigure. In the trial shown, the subject first moved to theicon located at �110 mm, overshot it, then moved tothe icon at �90 mm.

As can be seen by comparing the first and third traces,the icon at �90 mm has a haptic force (created with asawtooth waveform), while the one at �110 mm doesnot. Peak forces of 250 mN can be seen at the edge ofthe active icon; forces decrease as the subject moves to-ward the icon’s center. The set of forces occurring justafter 1500 samples (or 1.5 s) illustrates nicely the forceprofile of the sawtooth icon (see Figure 2). Subjectsgenerally used velocities between 50 mm/s and �50mm/s when exploring icons. Small spikes are visible inthe velocity signal at the time of abrupt force changesand may contribute to transient, high-frequency cues discernible to subjects in the case of“rough” icons.

Figure 4 shows the stimulus trends incurred for 6 testsubjects. Each line represents the peak magnitude of

Table 2. Mean Detection Thresholds in mN for 6 SubjectsExploring Three Rough and One Smooth Haptic Icons

Subject3 mmrough

4 mmrough

5 mmrough

5 mmsmooth

1 17.9 17.7 25.5 43.42 25.9 40.6 32.0 49.83 27.9 22.7 28.8 89.14 26.7 20.1 31.7 32.95 42.1 29.4 34.6 62.26 40.0 28.9 41.7 55.9Mean 30.1 26.6 32.4 55.6SD 9.2 8.3 5.5 19.3

336 PRESENCE: VOLUME 14, NUMBER 3

Page 9: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

force (at icon edges) present for each trial during onerun of the experiment. Subjects started each experimentat 500 mN of force. The force output then changedbased on their response to stimulus. Subjects generallytook between 75 and 85 trials per experiment beforeincurring the necessary 12 stimulus-path reversals.

When actively exploring for a 3-, 4-, and 5-mm at-tractive icon with our FHD, many subjects showedlower detection thresholds on smaller icons (Table 2);however, analysis of variance (ANOVA) on the first 3columns of Table 2 showed no statistical significancebetween rough-icon width and threshold value F(2,15) � 0.83, p .01.

3.2 5-mm Sinusoid (Smooth) Icons

In the second experiment, the same 6 subjectsactively explored a smooth sinusoidal icon of 5-mmwidth. This experiment was conducted in the same

manner as the first, but using only one icon width. Fig-ure 5 shows the adaptive-threshold stimulus paths takenfor the 6 subjects. The mean threshold value for all sub-jects was 55.4 mN, as compared to 32.4 mN for the5-mm sawtooth icon (Table 2).

The t test revealed that subjects had a lower detectionthreshold when exploring the 5-mm rough icon, versusthe 5-mm smooth icon, t(10) � 2.829, p .05.

3.3 Detection Thresholds for StaticPulses

In the third experiment, subjects held their fingermotionless and were subject to a short pulse tangentialto the virtual surface. Three pulse durations (50, 100,and 150 ms) were used. Table 3 shows the mean forcethreshold associated with each pulse duration for allsubjects.

An analysis of variance computed on the population

Figure 3. Typical exploration behavior of a subject for two icons, indicated by pairs of horizontal lines

(top trace). Icon A (position � �110 mm) had no force feedback, icon B (position � �90 mm) had a

sawtooth attractive force field. Time axis is in milliseconds.

Dosher and Hannaford 337

Page 10: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

of 6 subjects showed no statistical significance betweenthe pulse widths used in the experiment and subjects’detection-threshold values, F(2, 12) � 1.29, p .05.

3.4 Task-Performance Gains withHaptics

Nearly all subjects showed an increase in task per-formance (measured as FIPR) as the level of hapticstimulus was increased over the range from 0 mN to300 mN. Subjects had FIPRs ranging from 1.63 to 2.73bps at the “no-force” (0 mN) level, while rates rangedfrom 2.34 to 3.90 bps at the 300 mN force level. Subjectsas a whole showed an increase from 2.2 bps to 3.2 bps for0 and 300 mN, respectively (see Table 4).

Subject performance, as clicks per minute, was com-puted by observing the total time required to complete100 trials at a corresponding force level. The same trend

of increasing performance with haptic force is observ-able (mean CPM increases from 30.3 to 52.3 CPM asforce increases from 0 to 300 mN). ANOVA computedfor both the FIPR and CPM revealed that subject per-formance improves as the level of stimulus increases,F(3, 48) � 16.72, p .0001 and F(3, 48) � 25.26,p .0001, respectively. Tables 5 and 6 show the resultsfrom t tests analyzing subject performance differencebetween each stimulus level, t(23)�tstat, p x. All ttests revealed that subject performance increased witheach increase in force, except in the case of 0 mN versus50 mN force level.

4 Discussion

In summary, we examined the ability of humansubjects to detect small haptic effects in order to better

Figure 4. Adaptive threshold stimulus trends for 6 subjects during the sawtooth (rough) 2-icon

experiment. The adaptive threshold method converged in approximately 50–85 iterations to a threshold

corresponding to a 71% correct response level.

338 PRESENCE: VOLUME 14, NUMBER 3

Page 11: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

understand how these effects may be utilized in smalldevices. We chose three basic methods to convey hapticinformation to subjects, and examined their effective-ness. We found that small variations in the size of icons,or the length of static pulses had minor effects, if any.However, the smoothness of an icon had a marked ef-fect on detectability.

When subjects actively explored for a rough (sawtooth-shaped) 3-, 4-, or 5-mm attractive icon with our FHD,

we measured an average detection threshold of 30.1,26.6, and 32.4 mN, respectively, for a population of6 subjects. The slight trend of lower detection thresh-olds for smaller icons was not statistically significant.However, the mean detection thresholds for a smooth(sinusoidal) 5-mm icon was 55.4 mN—1.7 times ashigh, and this difference was significant, t(10) � 2.829,p .05.

Figure 5. Stimulus trends incurred by 6 subjects for a 2-icon force experiment with sinusoidal (smooth)

haptic icons. Adaptive threshold method converged in 75–100 iterations to a threshold corresponding to

a 71% correct response level.

Table 3. Static Pulse-Detection Thresholds

Pulse duration Detection threshold

50 ms 22.5 mN100 ms 25.6 mN150 ms 28.9 mN

Table 4. Variations in Performance at Different Levels ofHaptic Stimulus for 13 Subjects

Force FIPR CPM

0 mN 2.2 bps 30.350 mN 2.3 bps 32.4

100 mN 2.7 bps 41.3300 mN 3.2 bps 52.3

Dosher and Hannaford 339

Page 12: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

Our device is a force-based haptic display, as com-pared to a vibrotactile-based haptic display. Informationabout a haptic feature (or a surface) is conveyed by forceand changes in displacement, as sensed by the entirefinger of a subject. A vibrotactile display, on the otherhand, conveys haptic information by small deformationsand vibrations at a localized area—commonly the fin-gertip (Hasser & Weisenberger, 1993). A haptic displaygenerally has a lower bandwidth than a vibrotactile dis-play, due to greater inertia; however, vibrations can stillbe present in a haptic interaction due to high-frequencyresponse of the device, control instabilities, and jointbacklash. Instantaneous changes in commanded forcelevels (in control software), such as at the edge of oursawtooth icons, can produce high-frequency contentthat may be perceivable by a human operator (notespikes in velocity in Figure 3).

Figure 3 shows the position, velocity, and forces en-countered during the exploration of a 5-mm sawtoothicon. The first exploration of the �90-mm icon occursbetween 1546 and 1666 ms and is 120 ms in length.This corresponds to an average exploration speed ofabout 42 mm/s. A 512-point Fast Fourier Transform(Figure 6, 1st trace) of the the forces during this explo-ration (Figure 6, 5th trace) shows frequency contentvery similar to the analytical Fourier Transform of anidealized 120-ms sawtooth icon of equal width (2nd

trace) with magnitude 0.25 (corresponding the maxi-mum amplitude of the force data, 250 mN).

The continuous Fourier Transform of our sinusoidalicon, explored at a constant velocity, contains a singlefrequency (Figure 6, 3rd trace). Assuming the same ex-ploration velocity as the sawtooth above, 42 mm/s,then the temporal frequency of our sinusoid icon will be8.33 Hz. The sawtooth icon clearly exhibits muchgreater spectral content.

We found that by using a rough waveform (saw-tooth), we could convey the presence of a haptic iconusing a lower amount of force. It is well known thattouch is highly sensitive to vibration at frequencies upto 1000 Hz or more (Shimoga, 1993). From 0 to 250Hz, the threshold decreases, with a peak sensitivity ataround 250 Hz (Bolanowski, Gescheider, Verrillo, &Checkosky, 1988; Tan & Rabinowitz, 1996). Our dataare consistent with the hypothesis that increasing fre-quency content of a haptic effect, for example by chang-ing the force profile to a rougher one, could result inlower thresholds, potentially lowering the maximumforce output required by a haptic device. Conversely,using waveforms with reduced spatiotemporal frequencycontent could result in haptic features with higher de-tection thresholds.

Of icon width, peak force, and stiffness, only two canbe varied independently. It is interesting to note that

Table 5. P-Values for Performance Experiment when Analyzed as CPM. Each Cell Shows tstat/p-value.

50 mN 100 mN 300 mN

0 mN �0.996/p .01 �4.19/p .005 �8.43/p .000150 mN – �2.93/p .01 �6.74/p .0001

100 mN – – �3.78/p .001

Table 6. P-Values for Performance Experiment when Analyzed as FIPR

50 mN 100 mN 300 mN

0 mN �0.774/p .01 �3.32/p .01 �6.23/p .000150 mN – �2.42/p .05 �5.34/p .0001

100 mN – – �3.28/p .005

340 PRESENCE: VOLUME 14, NUMBER 3

Page 13: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

Figure 6. Fast Fourier Transforms (FFTs) of force data from a typical exploration of a sawtooth icon (top trace),

an idealized sawtooth of equal magnitude (2nd trace), idealized sinusoid icon at 8.3 Hz, and three static pulses.

The lower image shows the exploration data used for the FFT (top trace) with a sinusoid overlaid for comparison.

Dosher and Hannaford 341

Page 14: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

the local stiffness of the sinusoid at �x � 0 is�F�

2.5 mm,

while the 5-mm sawtooth stiffness is�F

2.5 mm (Figure 2)—

a factor of � greater stiffness. Vibrotactile effects, specifi-cally the higher frequency content of the rough icons,rather than stiffness, may explain the lower detectionthreshold for rough icons as compared to smooth icons.

The mean detection thresholds for 50-, 100-, and150-ms static pulses were 22.5, 25.6, and 28.9 mN,respectively, for a population of 6 subjects (Table 3).Although they appear to indicate a trend in lower detec-tion thresholds for smaller pulses, analysis of variance(ANOVA) did not show the differences to be statisti-cally significant, F(2, 12) � 1.29, p .05. An FFT ofthe three static pulses (magnitude 0.25) shows greaterspectral content, as the pulse length is decreased (Figure6, 4th trace).

An interesting subjective observation was that as themagnitude of these force pulses was reduced, subjectsbecame aware of the circulatory pulse in their fingertips.Near threshold, the heartbeat was sometimes hard todistinguish from the stimulus.

4.1 Performance Gains with Haptics

We conducted experiments assessing performancegains due to haptics to better understand how smallhaptic effects could be used in a human-machine inter-face to improve the rate at which a task can be per-formed. We found that even small levels of haptic force(100 mN) provide an improvement in task performance.Task performance continued to increase as the level ofhaptic stimulus was increased. The improvement in taskperformance continued up to the maximum level usedin these experiments, 300 mN.

We used the results of the detection-threshold experi-ments to choose our lowest nonzero force level for thefive-target performance experiment. While we found30.8 mN to be the mean detection threshold for 5-mmsawtooth icons, we observed (Figure 4) that subjectsbegin making an increasing number of errors below 50mN, and thus chose 50 mN as the minimum nonzeroforce level for the performance experiments. Addition-ally, as subjects approach their threshold level (after ap-

proximately the 50th trial), they never rise above 50mN again (Figure 4).

5 Conclusions

Any haptic display will have, to at least some de-gree, constraints placed on its weight, power, and size.Whether these constraints are a product of constructionmethods, available power, space limitations, or otherdesign requirements, it is useful to maximize the effec-tiveness of the haptic display. Very few haptic displayshave continuous force outputs that exceed that of a hu-man operator. Those haptic displays that do have largeforce outputs suffer from other limitations, such as highfriction and inertia, and large space requirements (Han-naford, Marbot, Buttolo, Moreyra, & Venema, 1996;Adams, Moreyra, & Hannaford, 1999). In all cases, uti-lizing methods that can convey useful information to auser in the most efficient manner will make optimal useof a haptic display’s abilities. This is especially importantwhen considering the design of small haptic interfaces,where the tightest constraints on weight, power, andsize are placed.

A goal in our research was to determine a minimumforce level that a human subject could detect with onehaptic device. That a haptic feature is detectable doesnot necessarily mean it is “useful,” or provides an im-provement in performance in a human-computer inter-face. Our performance measurement experimentsshowed that, while 50 mN is above the mean detection-threshold level, it is not high enough to provide a statis-tically significant increase in subject performance. Datafrom the remaining force levels (100 and 300 mN)showed that small haptic forces can increase perfor-mance in a simple point-and-click interaction. Otherinteresting experiments could be conducted that seek todetermine if there is a force level (or range) that corre-sponds to a peak performance level; high force levels onnontarget icons or features may contribute to a distract-ing effect.

One key idea in adding haptic interfaces to devices isto improve the quality and efficiency of the human-machine interface. Smaller devices, some of which havelimited interfaces, may benefit from augmenting their

342 PRESENCE: VOLUME 14, NUMBER 3

Page 15: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

interface with haptic information. Our research hashopefully provided insight into some of the issues in-volved in using small haptic effects in human-machineinterfaces.

References

Adams, R., Moreyra, M., & Hannaford, B. (1999, Novem-ber). Excalibur, a three-axis force display. ASME WinterAnnual Meeting Haptics Symposium, Nashville, TN.

Akamatsu, M., & MacKenzie, I. S. (1996). Movement charac-teristics using a mouse with tactile and force feedback. In-ternational Journal of Human-Computer Studies, 45, 483–493.

Akamatsu, M., MacKenzie, I. S., & Hasbrouc, T. (1995). Acomparison of tactile, auditory, and visual feedback in apointing task using a mouse-type device. Ergonomics, 38,816–827.

Bolanowski, S. J., Gescheider, G. A., Verrillo, R. T., & Chec-kosky, C. M. (1988). Four channels mediate the mechanicalaspects of touch. Journal of the Acoustical Society of Amer-ica, 84 (5), 1680–1694.

Burnett, G. E., & Porter, J. M. (2001). Ubiquitous comput-ing within cars: Designing controls for non-visual use. In-ternational Journal of Human-Computer Studies, 55, 521–531.

Dennerlein, J. T., & Yang, M. C. (2001). Haptic force-feedback devices for the office computer: Performance andmusculoskeletal loading issues. Human Factors, 43, 278–286.

Dhruv, N., & Tendick, F. (2000). Frequency dependenceof compliance contrast detection. Proceedings of theASME Dynamic Systems and Control Division, 69 (2),1087–1093.

Dosher, J. (2001, December). Detection thresholds and per-formance gains for small haptic effects. Unpublished. MSEEThesis, University of Washington, Seattle, WA.

Dosher, J., & Hannaford, B. (2001, October). Detectionthresholds for small haptic effects. SPIE Intelligent Systemsand Advanced Manufacturing Symposium, Boston.

Dosher, J., Lee, G., & Hannaford, B. (2001). How low canyou go? Detection thresholds for small haptic effects. In M.McLaughlin (Ed.), Touch in virtual environments, Proceed-ings of the USC Workshop on Haptic Interfaces, Feb. 23, 2001.Upper Saddle River, NJ, Prentice Hall.

Fitts, P. M. (1954). The information capacity of the human

motor system in controlling the amplitude of movement.Journal of Experimental Psychology, 47, 381–391.

Goldfarb, M., & Kilchenman, M. (2001). Force saturation,system bandwidth, information transfer, and surface qualityin haptic interfaces. Proceedings of the 2001 IEEE ICRA, 2,1382–1387.

Hannaford, B., Marbot, P., Buttolo, P., Moreyra, M., & Ven-ema, S. (1996). Scaling properties of direct drive serialarms. International Journal of Robotics Research, 15 (5),459–472.

Hasser, C., Goldenberg, A., Martin, K., & Rosenberg, L.(1998). User performance in a GUI pointing task with alow-cost force-feedback computer mouse. Proceedings ofthe ASME Dynamic Systems and Control Division, 1, 151–156.

Hasser, C., Martin, D., & Dennerlein, J. (2000). Force-feedback improves performance for steering and com-bined steering-targeting tasks. CHI 2000 Conference Pro-ceedings on Human Factors in Computing Systems, 1, 423–429.

Hasser, C., & Weisenberger, J. (1993). Preliminary evaluationof a shape memory alloy tactile feedback display. Proceedingsof the Symposium on Haptic Interfaces for Virtual Environ-ments and Teleoperator Systems, ASME Winter Annual Meet-ing, 1, 73–80.

Lederman, S. J., Klatzky, R. L., Hamilton, C., & Grindley, M.(2000). Perceiving surface roughness through a probe: Ef-fects of applied force and probe diameter. Proceedings of theASME Dynamic Systems and Control Division—2000, 2,1065–1071.

Louw, S., Kappers, A., & Koenderink, J. (2002, 12 July).Haptic discrimination of stimuli varying in amplitude andwidth. Experimental Brain Research, 146, 32–37.

Moy, G., Singh, U., Tan, E., & Fearing, R. (2000, February18). Human psychophysics for teletaction system design.Haptics-e, The Electronic Journal of Haptics Research, 1(3).Available from: www.haptics-e.org.

Oakley, I., Brewster, S. A., & Gray, P. D. (2001). Solvingmulti-target haptic problems in menu interaction. ExtendedAbstracts of ACM CHI 2001, 357–358.

Shimoga, K. (1993, September). A survey of perceptualfeedback issues in dexterous telemanipulation: Vol. 1.Finger force feedback (263–270). Proceedings IEEEVRAIS-93.

Srinivasan, M. A., Beauregard, G. L., & Brock, D. L. (1996).The impact of visual information on the haptic perceptionof stiffness in virtual environments. Proceedings of the ASMEDynamic Systems and Control Division, 58, 555–559.

Dosher and Hannaford 343

Page 16: Human Interaction with Small Haptic Effectsbrl.ee.washington.edu/eprints/220/1/Rep173.pdf · 1.2 Fitts’ Law Fitts’ law (Fitts, 1954) was used in this study to evaluate the results

Stevens, J. C., Foulke, E., & Patterson, M. Q. (1996). Tactileacuity, aging, and braille readings in long-term blindness.Journal of Experimental Psychology: Applied, 2(2), 910–106.

Tan, H. Z., & Rabinowitz, W. M. (1996). A new multi-fingertactile display. Proceedings of the ASME Dynamic Systemsand Control Division, 58, 515–522.

Venema, S. (1999, April). Experiments in surface perceptionusing a haptic display. Unpublished doctoral thesis, Univer-sity of Washington, Seattle, WA.

Venema, S., & Hannaford, B. (2000, July). Experiments infingertip perception of surface discontinuities. InternationalJournal of Robotics Research, 19(7), 684–696.

Venema, S., Matthes, E., & Hannaford, B. (2000). Flat coil ac-tuator having coil embedded in linkage. U.S. Patent Pending.

Wall, S. A., & Harwin, W. S. (2000). Effects of physical band-width on perception of virtual gratings. Proceedings of theASME Dynamic Systems and Control Division, 2,1033–1047.

344 PRESENCE: VOLUME 14, NUMBER 3