86

IEEE TRANSACTIONS ON HAPTICS,

VOL. 7, NO. 1,

JANUARY-MARCH 2014

Reorienting Driver Attention with Dynamic Tactile Cues Cristy Ho, Rob Gray, and Charles Spence Abstract—A series of three experiments was designed to investigate whether the presentation of moving tactile warning signals that are presented in a particular spatiotemporal configuration may be particularly effective in terms of facilitating a driver’s response to a target event. In the experiments reported here, participants’ visual attention was manipulated such that they were either attending to the frontal object that might occasionally approach them on a collision course, or else they were distracted by a color discrimination task presented from behind. We measured how rapidly participants were able to initiate a braking response to a looming visual target following the onset of vibrotactile warning signals presented from around their waist. The vibrotactile warning signals consisted of single, double, and triple upward moving cues (Experiment 1), triple upward and downward moving cues (Experiment 2), and triple random cues (Experiment 3). The results demonstrated a significant performance advantage following the presentation of dynamic triple cues over the static single tactile cues, regardless of the specific configuration of the triple cues. These findings point to the potential benefits of embedding dynamic information in warning signals for dynamic target events. These findings have important implications for the design of future vibrotactile warning signals. Index Terms—Haptic I/O, human factors, human information processing, automotive

Ç 1

INTRODUCTION

I

N recent years, there has been a great deal of interest in the development of multisensory driver assistance systems (e.g., see [1], [2]). While the trend for the future would appear to be the development and installation of completely autonomous driver systems that perform the majority of ‘accident-prone’ driving tasks (e.g., parking, lane changing, etc.) at the push of a button or increasingly via voice, or some other form of gestural, control (e.g., [3]), the design of collision warning systems remains an important area of applied research (see [4]). These systems are designed to keep drivers informed with regard to critical driving situations that they may want to intervene in, should the autonomous driver systems (for whatever reason) go astray. The development of effective collision warning systems represents a potentially low-cost solution to reducing the number of accidents on the road, which are often associated with the loss of life and substantial financial costs (see [5]). Research in this area has demonstrated that the presentation of tactile warning signals can offer an effective means of alerting a driver to a potential danger and rapidly orienting their spatial attention to the location, or direction, of interest in the environment (see [6] for a review). It is often said that tactile warnings may be more effective than equivalent warning signals presented in other sensory





C. Ho and C. Spence are with the Department of Experimental Psychology, University of Oxford, Oxford OX1 3UD, United Kingdom. E-mail: [email protected], [email protected]. R. Gray is with the School of Sport and Exercise Sciences, University of Birmingham, Birmingham B15 2TT, United Kingdom. E-mail: [email protected].

Manuscript received 17 July 2013; revised 28 Oct. 2013; accepted 12 Nov. 2013. Date of publication 2 Dec. 2013; date of current version 14 May 2014. Recommended for acceptance by R. Klatzky. For information on obtaining reprints of this article, please send e-mail to: [email protected], and reference the Digital Object Identifier below. Digital Object Identifier no. 10.1109/TOH.2013.62

modalities. The reason for this being that the sense of touch is relatively uninvolved in the task of driving and any other secondary tasks that a driver may engage in, such as talking on a phone or using the in-car navigation system [7]. There is certainly already some evidence to support such a proposal (e.g., [8], [9]). However, one limitation with the use of tactile warnings is the difficulty associated with conveying meaning (or some information about the collision event other than merely a spatial direction) to the driver. In the case of visual and auditory warning signals, the cues themselves are typically semantically meaningful in that they are naturally mapped onto certain well-learnt external events. A red exclamation mark, for instance, is generally understood as some kind of high importance error signal; whereas a car horn sound is interpreted as the driver of a vehicle trying to attract someone else’s attention [10]. Meanwhile, in the case of tactile warning signals, it is unclear whether there exist only a limited number of universally recognised tactile patterns or vibrations that intuitively signify one thing or another (e.g., rumble strips on the road indicating a driver’s approach to a roundabout). It is also unclear whether with people’s increasingly widespread exposure to tactile interfaces, drivers will more easily be able to master various different types of symbolic tactile signals. This is similar to the case of mastering the auditory cues associated with a rear parking sensor system before their meaning become instantly recognisable. In this regard, previous research on the presentation of structured abstract tactile information (known as tactile icons, or ‘tactons’) has demonstrated that people can reliably identify no more than two to three different vibrotactile parameters (e.g., rhythm, roughness, and spatial location; e.g., [11], [12]). This appears to suggest that the number of distinct types of information that can be communicated via vibrotactile cues by a given system may be quite limited. For instance, Jones and her colleagues have

1939-1412 ß 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

HO ET AL.: REORIENTING DRIVER ATTENTION WITH DYNAMIC TACTILE CUES

demonstrated that people can be trained to recognise patterns presented on the skin. However, it should be pointed out that trained users were tested in Jones et al.’s study, while the aim of the current study was to investigate tactile information presentation that would be more intuitive and automatic, by taking into account the affordances of the environment. Note also that the constraints governing tactile information presentation in cars versus in military applications are quite different. In particular, military personnel can be required to wear an extra vest or receive extensive training with a tactile system. This is not really a practical solution for the typical car driver. In any case, for the design of in-car warning signals, it is important that drivers be informed about the nature of the event so that they can react appropriately in a time-critical manner. Leo et al. [13] reported that looming sounds can be particularly effective in terms of acting as a salient cue to improve a person’s visual sensitivity in the cued direction. In this case, performance was enhanced in an orientation discrimination task (i.e., in a laboratory-based psychophysical task). Leo et al. suggested that looming auditory cues might have tapped into specific ‘threat-related’ circuitry that presumably enhanced the early stages of visual information processing. It is thus interesting to examine whether the presentation of looming tactile stimuli could also be used to tap into the same ‘threat-related’ brain circuitry and thus enhance a driver’s visual perceptual sensitivity. In the applied domain, [14] recently demonstrated the potential benefits of the presentation of looming intensity auditory warning signals in terms of improving the speed and accuracy of a driver’s response to an impending front-to-rear-end collision event (see also [15]). The theoretical rationale underlying the present study was to determine whether tactile signals could be programmed in a manner that represents an approaching/ looming event and that automatically captures a person’s attention. In the literature on attention, for instance, various studies have demonstrated that visual and auditory events that involve the movement of an object toward a person are considered to have greater biological salience (i.e., they are more attention capturing) than equivalent events that involve movement away from the observer (e.g., [16]). It would therefore be interesting to investigate whether distance information encoded tactually is processed by similar mechanisms as in the case of visual and auditory looming stimuli. This may have important theoretical implication given that vision and audition are typically considered as ‘distal’ senses while touch is a ‘proximal’ sense. In the present study, we hypothesized that the presentation of tactile signals that contained inherent motion/approach information might facilitate a driver’s speeded braking responses.

2

EXPERIMENT 1

In Experiment 1, participants completed a simulated object following task directly analogous to the car-following scenarios described in [17]. The participants in the present study had to follow a ‘car’ (in this case represented by a ball) travelling in a simulated trajectory in front of them in three-dimensional space (created using a simulated optic flow elicited by means of the presentation of a field of

87

Fig. 1. Screenshot of the visual stimuli presented in the present study. The background dots moved continuously out from the centre toward the edge of the screen. The speed at which the dots travelled was modelled as a function of the angle at which the accelerator and brake pedals were depressed. The initial size of the lead object was varied on each trial. Note that white arrows were added here to illustrate the flow direction of the grey dots. These arrows were not present in the actual visual stimuli seen by the participants.

moving dots; see Fig. 1). The participants had to follow the lead object at a set speed (i.e., distance) by depressing the accelerator pedal. The colour of the lead object changed between yellow (appropriate) and grey (too fast or too slow) so as to indicate to the participants the appropriateness of their speed. The travelling speed of the moving field of dots was simulated as a function of the driven speed of the participant’s ‘car’ in order to simulate the windscreen view of objects moving in the environment as in the case of actual driving. The participants were instructed to maintain a safe distance with respect to the lead object, and to depress the brake pedal in response to the onset of a closing situation (i.e., a sudden rapid expansion of the visual size of the lead object). We hypothesized that the presentation of a triple vibrotactile warning signal (that translated into a vertical upward directional movement toward the participant’s head) might give rise to a larger performance benefit than the presentation of either a double or single vibrotactile warning signal of the same aggregated objective intensity and duration (note though that the subjectively perceived intensity and duration might differ as a function of the warning type). We also hypothesized that the benefits of the presentation of warning signals over a no-warning condition would be larger when the participants were looking away (i.e., when they were performing a secondary colour discrimination task) in the periphery than when they were looking straight ahead.

2.1 Methods Sixteen participants took part in this experiment (mean age of 27 years, age range from 23-40 years; seven males and nine females). The participants reported having, on average, eight years of driving experience (range from 2-23 years). Half of the participants reported that they would normally drive on the left side of road (i.e., as is normally the case in England and most Commonwealth countries), while the remainder reported that they normally drove on the right side of the road instead. All of the participants reported having full colour vision, normal or corrected-to-normal

88

vision, normal hearing, a normal sense of touch, and no history of neck pain. The experiment lasted for about 30 min and the participants received £5 in return for taking part. The experiment was conducted in accordance with the ethical guidelines laid down by the Central University Research Ethics Committee, University of Oxford. The participants were seated in the centre of a soundproof experimental booth. A steering wheel (Logitech MOMO Racing Force Feedback Wheel; Logitech, Inc., Fremont, CA) was placed in front of them, with accelerator and brake pedals placed on the floor at a comfortable distance from their right foot. The accelerator and brake pedals were used to record the participant’s driving responses, while two hand paddles on each side situated behind the steering wheel were used to record the participant’s responses to the speeded colour discrimination task. A 23-inch computer monitor positioned at a viewing distance of 100 cm directly in front of the participants was used to present the frontal simulated driving stimuli. A tricolour (red-green-blue) light-emitting diode (LED) was placed about 115 cm behind the participant’s back at each of the two corners of the experimental booth (at an elevation that roughly matched the participant’s shoulder level when seated). These LEDs served as the visual targets for the colour discrimination task. Vibrotactile stimuli were presented via three tactors (VBW32, Audiological Engineering Corp., Somerville, MA) fastened to the participant’s waist using a Velcro belt. The tactors were aligned vertically (the distance between adjacent tactors was 23 mm center-to-center such that the tactors were close but not touching one another when placed on a flat surface; due to body movement during the experiment the inter-tactor distance might not always be exactly the same for each participant; note that the two-point acuity threshold around the belly at navel level has been estimated to be about 19 mm for a person aged 20 years and 24 mm for a 40-year-old; see [18]) with the bottom tactor placed just above the participant’s bellybutton. The tactors were driven by a 250 Hz sinusoidal signal at 1/4 of the full intensity of the tactor that was clearly perceivable by our participants. Three types of vibrotactile stimuli were presented: single, double, and triple vibrotactile cues. The single cue consisted of the operation of one of the three tactors for 300 ms. The double cue consisted of the sequential operation of two of the three tactors, each for 150 ms, with the tactor situated closest to the bottom always being activated before the tactor situated closer to the top (head). The triple cue consisted of the sequential operation of all three tactors, from bottom to top, each for 100 ms. Two side loudspeakers positioned to the left and right side, 100 cm in front of the participant, were used to present the auditory stimuli. The auditory stimuli consisted of double tone bursts (modified from a phone ring tone with 15 ms on, 10 ms off, and 15 ms on) presented at 66 dB(A). These stimuli were used to indicate the direction in which the driver was required to turn his/her head to respond to the secondary colour discrimination task. Pink noise was also delivered at 60 dB(A) from the two side loudspeakers throughout the experimental session to mask any noise produced by the operation of the tactors.

IEEE TRANSACTIONS ON HAPTICS,

VOL. 7, NO. 1,

JANUARY-MARCH 2014

The frontal driving simulation consisted of a moving dot field with grey dots travelling from the centre to the edge of the screen against a black background. The dots were programmed to move as a function of the angle at which the accelerator was depressed, calculated with the formula: Dot speed ¼ (Angle of accelerator depression  1 þ 1)  Max simulated speed of dots/Simulated speed of dots Each dot travelled on its randomized path and when the accelerator was depressed, the moving dot fields moved continuously. A three-dimensional ball object presented at the centre of the screen represented the ‘car’ in front. The colour of the lead object changed between yellow and grey depending on whether the participants were depressing the accelerator at the appropriate speed (yellow) or that they were travelling too fast or too slow (grey). The lead object normally had a radius of 2  0.5 degree of visual angle, which was randomly chosen prior to the start of each trial. The radius was varied in order to generate some randomness in the approach of the lead object. This feature of the design thus ensured that the participants could not perform the task simply by comparing the current visual size of the lead object to a particular constant size of the lead object. When a critical closing-in event was presented, the radius of the lead object would rapidly expand by 3 degree within 1,000 ms to simulate a lead vehicle suddenly hitting the brake. The experimental session consisted of a block of 22 practice trials and a block of 165 experimental trials. The warning signal type (single, double, or triple cues) and head position (forward, turned to the left, or turned to the right) were fully crossed, and each condition (nine critical trials in total for each warning signal type per each head position) was randomly chosen and presented equiprobably within the experiment. Note that half of the trials after a head turn were followed by a warning signal (or a closingin event), and for the other half of the trials after a head turn (66 trials in total, 33 for each side) this was followed by a colour discrimination task. Catch trials (nine trials in total, three for each head position) in which there was no ‘closing-in’ after the presentation of a warning signal were presented at a ratio of 1:9 to that of the critical trials. Baseline trials (nine trials in total, three for each head position) in which no warning signal was presented before the closing-in event were presented at a ratio of 1:9 to that of the critical trials. The response mapping of the left and right hand paddles to the colour target blue or green was counterbalanced across participants. The participants were instructed to imagine that they were following a car in front of them (represented by the ball presented at the centre of the screen). They had to try and maintain a safe distance (i.e., speed) with respect to the lead object by keeping the accelerator depressed halfway with their right foot such that the colour of the lead object always stayed yellow (rather than grey). The participants were instructed to lift their right foot off the accelerator and hit the brake pedal as rapidly as possible when they detected any closing-in with the object in front. This was represented by the closure of the lead object toward them (i.e., by the expansion of the image). The participants were also instructed to turn their head to the side indicated

HO ET AL.: REORIENTING DRIVER ATTENTION WITH DYNAMIC TACTILE CUES

by the double tone auditory cue and to check the red light behind them. The participants were instructed to imagine this as checking the blind spot behind while driving. They were instructed to keep looking at the red light and pull the corresponding left or right paddle as soon as they detected the colour change to either blue or green. Each trial began with the presentation of the vibrotactile warning signal for 300 ms (in the forward head position) or the presentation of the double tone burst (in the headturn trials). In the head-turn trials, the onset of the auditory cue coincided with the onset of the red light on the cued side. In 50 percent of these trials, the red light turned blue or green after a random interval of 601-1599 ms. The participants had to indicate the colour change by pulling the corresponding hand paddle. The target light remained on until after a participant’s response had been recorded, or if no response was detected within 1500 ms of the onset of the colour change. On the remaining 50 percent of trials, the vibrotactile warning signal (300 ms) was presented after the onset of the auditory cue signalling a head turn following a random interval of 601-1599 ms. The participant had to turn their head back to the forward position and decide as rapidly as possible whether there was a closing-in critical event. In the critical trials involving a closing-in event (in both the forward head position and headturn trials), the onset of the warning signal coincided with the onset of the expansion of the lead object. The expansion motion lasted for a maximum of 1000 ms and the lead object would remain at its expanded size until the end of the trial. The participant had to depress the brake pedal as soon as possible should they detect the closing-in. A trial was terminated after a participant’s response or if no response had been recorded within 2 s of the onset of the warning signal. The lead object then resumed to the next randomized size of 2.5-3.5 degrees of visual angle during the intertrial interval of 5 to 10 s. The participants were given a break upon completion of every 55 trials during the experimental block.

2.2 Results Braking reaction times (RTs) defined as the time after the onset of a critical event when the participant initiated a braking response by depressing the brake pedal for over 1/3 of its complete depression was measured. The arbitrary value of 1/3 was chosen to ensure that any recorded response was intended to be a braking response by the participant and not random noise such as the participant gently hitting the edge of the brake pedal for whatever reason. A within-participants analysis of variance (ANOVA) was performed on the braking RT data in order to assess the relative effectiveness of the presentation of the different types of vibrotactile warning signals in orienting the participants’ attention to the front for a speeded collision avoidance response. Those trials in which the participants failed to respond before the termination of a trial (i.e., misses) were discarded from the analysis of the RT data. On average, these trials accounted for less than 1 percent of all trials in which a warning signal was presented. The two factors in the experimental design were warning signal type (single, double, or triple cues) and head

89

Fig. 2. Mean latency of speeded braking responses (RTs in ms) as a function of the type of warning signal and head position in Experiment 1. Error bars indicate the standard errors of the means.

position (forward, turned toward the left, or turned toward the right). The analysis of the data revealed a significant main effect of the type of warning signal that was presented, F ð2;30Þ ¼ 3:5; p ¼ 0:042, Cohen’s f ¼ 0:47 (see Fig. 2). Post-hoc pairwise comparison revealed no statistical significant difference between the single ðM ¼ 852 ms) and double (M ¼ 852 ms) vibrotactile cue conditions, p ¼ 0:99. Importantly, however, the differences between the triple cue condition (M ¼ 824 ms) and the other two conditions were significant, p ¼ :034, Cohen’s d ¼ 0:17 and p ¼ 0:032, Cohen’s d ¼ 0:17, respectively, for the single and double cue conditions. These differences indicate that the participants were significantly more responsive to the triple vibrotactile cues than to either the single or double cues (that lasted for the same total duration of 300 ms). The analysis also revealed a significant main effect of head position, F ð2;30Þ ¼ 4:1; p ¼ 0:027, Cohen’s f ¼ 0:51. Post-hoc pairwise comparisons revealed that participants responded significantly more rapidly when they were looking forward (M ¼ 790 ms) than when they were looking to the left (M ¼ 869 ms, p ¼ 0:049, Cohen’s d ¼ 0:50Þ or right ðM ¼ 869 ms, p ¼ 0:053, Cohen’s d ¼ 0:48) at the onset of the warning signal. Unsurprisingly, the difference between the two conditions in which the participant’s head was turned was not statistically significant, p ¼ 0:99. The interaction between warning signal type and head position failed to reach statistical significance, F ð4;60Þ ¼ 1:5; p ¼ 0:23, Cohen’s f ¼ 0:31. Overall, the participants missed an average of 0.0, 12.5, and 6.3 percent of the trials in which no warning signal was presented prior to the onset of the closing-in event in the forward, turned to the left, and turned to the right head positions, respectively. Pairwise comparison of their RTs in response to the closing-in event revealed that they responded significantly more slowly when there was no warning signal (M ¼ 1;078 ms) than when a warning signal was presented, all ps < 0:001, suggesting that the

90

presence of any warning signal was beneficial when compared to the absence of any signal, just as expected. The participants correctly responded to the colour discrimination task with a mean RT of 724 ms (SE ¼ 35 ms) and an accuracy of 93.4 percent (SE ¼ 1:3%).

2.3 Discussion Experiment 1 was designed to investigate whether the presentation of vibrotactile cues with certain embedded meaning in terms of stimulus motion (i.e., with vibrotactile cues travelling along the body surface) might facilitate a driver’s collision avoidance responses, as opposed to continuous vibrotactile cuing presented from exactly the same position on the participant’s body surface. This proposal was based on the hypothesis that the vertical motion information might constitute some kind of natural mapping to the potential collision event occurring directly in front of the driver, which, in turn, might facilitate our participants’ behavioural performance in response to the collision event. Previous studies have demonstrated the potential usefulness of directional waypoint vibrotactile cuing for navigation by mapping waypoint directions on the location of the vibrations around the waist [19]. Consistent with our earlier studies demonstrating the effectiveness of vibrotactile cuing for front-to-rear collision (e.g., [8], [20]), the results of the present study revealed a significant advantage for vibrotactile cuing in facilitating a participant’s speeded braking response to closing-in events in a simulated driving task. What’s more, the results revealed that the triple vibrotactile cues were significantly more effective in terms of facilitating the participant’s response, than the single and double cuing conditions. These findings would appear to suggest that the upward directional information embedded in the triple cuing condition might have contributed to the response facilitation. One might also argue that the response facilitation could have been a direct consequence of stimulating a wider area on the body surface overall than stimulating just a single spot. However, this is unlikely to be the case given that the continuous (relatively longer) stimulation over a single position might actually have been expected to give rise to a more intense overall sensation than when the relatively shorter vibrations traversed the body. It is also generally the case that the higher the intensity of a vibrotactile cue, the more alerting it is, the faster the expected response is. Note here that we chose to fix the overall cue duration at 300 ms, given that the participant’s attention might have been captured by the source of the vibrations, so if the cue remained on until a participant’s response (as with many existing auditory alarm systems), we thought it possible that the participants might perceive the vibrotactile cues as a cause of distraction that could therefore potentially cause interference to their speeded response. In contrast to our initial hypothesis that double cuing might offer a performance advantage over single cuing, given that it should also have provided the upward motion cue just as in the triple cuing condition, we failed to observe any difference between single and double cuing. It is possible that any illusion of tactile apparent motion generated by double vibrations was simply not

IEEE TRANSACTIONS ON HAPTICS,

VOL. 7, NO. 1,

JANUARY-MARCH 2014

salient enough and thus our participants might have failed to perceive the intended upward motion cue. It is, however, also possible that our participants were unable to perceive the temporal sequence of the two vibrations, rendering its motion information ineffective. With triple vibrations, we suspect that our participants could probably have more reliably identified the direction indicated by the cues than in the case of double vibrations. One of the factors that may have contributed to the apparent advantage of the dynamic triple cues relates to their travel path toward the participant’s head. It is likely that the perception of vibrotactile cues moving toward (i.e., approaching) the head would have been more behaviourally relevant and thus given rise to an alerting effect that would facilitate the participant’s speeded response. In order to investigate whether the nature of the directional cuing had any effect on the effectiveness of the triple vibrotactile cues in facilitating a driver’s collision avoidance response, we conducted a second experiment in which we compared vibrotactile cues that travelled in an upward versus downward direction.

3

EXPERIMENT 2

3.1 Methods Twelve participants took part in this experiment (mean age of 29 years, age range from 21-43 years; three males and nine females). The participants had an average of 10 years of driving experience (ranging from 1-26 years), by selfreport. Eight of the participants reported that they would normally drive on the left, while the remainder reported that they would normally drive on the right side of road. The apparatus, materials, design, and procedure were identical to those used in Experiment 1, with the exception that the warning signal type now consisted of upward versus downward moving triple vibrotactile cues. The upward triple cue was the same as the triple cue presented in Experiment 1. The downward triple cue consisted of the sequential operation of all three tactors, from top to bottom, each presented for 100 ms. The experimental session consisted of a block of 16 practice trials followed by a block of 150 experimental trials (12 critical trials for each warning signal type per each head position; with a break once every 50 trials). Catch trials were presented at a ratio of 1:8 to that of the critical trials. Baseline trials were presented at a ratio of 1:8 to that of the critical trials. 3.2 Results Similar analyses were performed on the braking RT data from Experiment 2. An ANOVA with the within-participants factors of warning signal type (upward versus downward triple cue) and head position (forward, turned to the left, or turned to the right) was performed in order to compare the relative effectiveness of the two types of moving vibrotactile cues in orienting participants’ attention for a speeded braking response for targets in front. On average, participants missed less than 1 percent of all trials in which a warning signal was presented. The analysis of the RT data revealed a significant main effect of head position, F ð2;22Þ ¼ 6:1; p ¼ 0:008, Cohen’s f ¼ 0:71 (see Fig. 3). Post-hoc pairwise comparisons once

HO ET AL.: REORIENTING DRIVER ATTENTION WITH DYNAMIC TACTILE CUES

Fig. 3. Mean latency of speeded braking responses (RTs in ms) as a function of the warning signal type and head position in Experiment 2. Error bars indicate the standard errors of the means.

again revealed that participants responded significantly more rapidly when looking forward (M ¼ 832 ms) than when their head was turned to the left (M ¼ 938 ms, p ¼ 0:008, Cohen’s d ¼ 0:62Þ or right ðM ¼ 933 ms, p ¼ 0:037, Cohen’s d ¼ 0:60Þ at the onset of the warning signal. There was no statistically significant difference between the two conditions in which the participant’s head was turned, p ¼ 0:84, Cohen’s d ¼ 0:03. However, the analysis failed to reveal any significant main effect of the type of warning signal, F ð1;11Þ < 1, n.s., suggesting no performance benefit of one type of cuing over the other ðM ¼ 902 ms for upward moving and M ¼ 900 ms for downward moving tactile cues). The interaction between warning signal type and head position also failed to reach statistical significance, F ð2;22Þ ¼ 2:0; p ¼ 0:16, Cohen’s f ¼ 0:41. Overall, on average, the participants missed 0.0, 11.1, and 2.8 percent of the trials in which no warning signal was presented prior to the target event in the forward, turned to the left, and turned to the right head position, respectively. Pairwise comparison of the RTs in the presence versus absence of the vibrotactile warning signal once again revealed that the participants responded significantly more slowly when there was no warning signal (M ¼ 1;115 ms) than when a warning signal was presented, both ps < 0.001, as predicted. The participants correctly responded to the colour discrimination task with a mean RT of 795 ms (SE ¼ 48 ms) and with an accuracy of 91.9 percent (SE ¼ 1:5 percent). Their performance in this task was similar to that observed in Experiment 1.

3.3 Discussion Experiment 2 was designed to investigate whether the presentation of vibrotactile cues that moved vertically over the participant’s waist in front travelling in an upward direction might offer potential benefits in term of speeding a driver’s braking responses than vibrotactile cues that ran in a downward motion (i.e., away from the participant’s head). Contrary to our hypothesis that the

91

upward moving cue would carry certain behaviourally relevant information that might facilitate the driver’s collision avoidance response to a frontal event, the results of Experiment 2 failed to reveal any significant difference between the two types of vertically moving triple vibrotactile cues, despite the observation of a significant advantage for the triple vibrotactile cues over both the double and single vibrotactile cues in Experiment 1. While it seemed possible that an upward moving cue might be particularly alerting in terms of initiating a fight or flight response, the downward moving cue could have primed the appropriate behavioural response (i.e., lifting their right foot off the accelerator and hitting the brake pedal; see [21]). Given that the potential collision event in the present study always consisted of a braking event, we were unable to tell whether or not the present pattern of results was affected by response priming. One might also argue that the upward and downward moving cues might not map onto the intended motion/approach of the target frontal closing-in event, given that the moving cues travelled on the coronal plane while the approach information of the visual target event travelled on the sagittal plane. In any case, the results of Experiment 2 could be interpreted as suggesting that the advantage arising from the presentation of triple vibrotactile cues in Experiment 1 could be attributable to the intrinsic property of the movement per se, without regard to the direction of that motion relative to the collision event. Based on the results of the first two experiments, it is unclear whether our results can be taken to suggest that a continuously moving triple cue connoted the motion of an external event approaching the driver, or whether instead the triple vibrotactile cues simply offered the drivers three chances to detect a warning that, by a process of probability summation, might have created a benefit relative to both the double and single warning signal conditions. In order to answer this question, we conducted a third and final experiment in which the triple vibrotactile cue travelled in a random (i.e., non-sequential) manner, together with double and single vibrotactile cues similar to those presented in Experiment 1. We hypothesized that if the directional moving triple cue conveyed some form of approach information to the participants that was not inherit in a random triple cue, then we would expect to observe no statistically significant advantage of the presentation of the random triple cue over either of the double and single cues in Experiment 3. By contrast, if the superiority of the triple vibrotactile cues was attributable to its redundancy or to the fact that it conveyed motion information that was not available in the single vibrotactile cue condition, then one would expect to observe a performance advantage in the random triple cue condition, just as in the triple upward moving cue presented in Experiment 1.

4

EXPERIMENT 3

4.1 Methods Sixteen participants took part in this experiment (mean age of 26 years, age range from 20-40 years; 5 males and 11 females). The participants had an average of seven years of driving experience (ranging from 1-23 years), by self-report.

92

Fig. 4. Mean latency of speeded braking responses (RTs in ms) as a function of the type of warning signal and head position in Experiment 3. Error bars indicate the standard errors of the means.

Nine of the participants reported that they would normally drive on the left side of road, while the remainder reported that they would normally drive on the right. The apparatus, materials, design, and procedures were identical to those used in Experiment 1, with the exception that the triple warning cue now consisted of the operation of the three tactors, each for 100 ms, in one of the following three random orders: bottom-top-middle, middle-top-bottom, and topbottom-middle. Thus, the triple cue did not travel in a particular (or predictable) direction.

4.2 Results and Discussion Similar analyses to those reported in Experiment 1 were performed on the braking RT data from Experiment 3. An ANOVA with the within-participants factors of warning signal type (single, double, or random triple cues) and head position (forward, turned to the left, or turned to the right) was performed in order to examine the relative effectiveness of the random triple cues as compared to the single or double cues in terms of orienting the participants’ attention for a speeded braking response for targets occurring in front. On average, the participants missed less than 1 percent of all trials in which a warning signal was presented. As in Experiment 1, the analysis of the RT data revealed a significant main effect of warning signal type, F ð2;30Þ ¼ 3:4; p ¼ 0:047, Cohen’s f ¼ 0:46 (see Fig. 4). Posthoc pairwise comparison revealed no statistically significant difference between the single (M ¼ 860 ms) and double (M ¼ 850 ms) vibrotactile cue conditions, p ¼ 0:40, Cohen’s d ¼ 0:08. Importantly, however, the participants responded significantly more rapidly to the random triple cue (M ¼ 829 ms) than to the single cue condition, p ¼ 0:017, Cohen’s d ¼ 0:26. The difference between the double and random triple cue conditions failed to reach statistical significance, p ¼ 0:13, Cohen’s d ¼ 0:17. The analysis once again revealed a significant main effect of head position, F ð2;30Þ ¼ 8:1; p ¼ 0:002, Cohen’s f ¼ 0:71. Post-hoc pairwise comparisons revealed that participants

IEEE TRANSACTIONS ON HAPTICS,

VOL. 7, NO. 1,

JANUARY-MARCH 2014

responded significantly more rapidly when they were looking forward (M ¼ 790 ms) than when they were looking to the left (M ¼ 874 ms, p ¼ 0:008, Cohen’s d ¼ 0:70Þ or right ðM ¼ 875 ms, p ¼ 0:006, Cohen’s d ¼ 0:70Þ at the onset of the warning signal. The difference between the two conditions in which the participant’s head was turned was not statistically significant, p ¼ 0:96, Cohen’s d ¼ 0:01. The interaction between warning signal type and head position failed to reach statistical significance, F ð4;60Þ ¼ 1:3; p ¼ 0:27, Cohen’s f ¼ 0:29. Even though this interaction term, as in Experiment 1, was not statistically significant, it is clear that the advantage for triple cuing over the other two cuing conditions was not present in the forward head position. One possibility could be a statistical floor effect. Besides, the large number of closing-in events following head turns (on half of the trials) might have provided a motivation to the participant to try to multi-task as much as possible. One might then expect this to operate against the potential warning effect in these conditions. However, this was actually not the case and the strongest effect of the presentation of warning signals was observed when the participants were multi-tasking. Overall, on average, the participants missed 0.0, 4.2, and 6.3 percent of the trials in which no warning signal was presented prior to the onset of the closing-in event in the forward, turned to the left, and turned to the right head position conditions, respectively. Pairwise comparison of their braking RTs revealed that participants responded significantly more slowly when no warning signal (M ¼ 1;090 ms) was presented prior the onset of the closing-in event than in the warning signal conditions, all ps < 0:001. The participants correctly responded to the colour discrimination task with a mean RT of 762 ms (SE ¼ 35 ms) and with a mean accuracy of 91.8 percent (SE ¼ 1:6 percent). Note that this pattern of results was similar to that observed in the other two experiments.

5

GENERAL DISCUSSION

In contrast to the idea that a random triple vibrotactile cue that travelled in a non-specific direction might not offer any advantage over a static single vibrotactile cue, we observed a significant performance advantage of random triple cuing over single cuing in Experiment 3. These results therefore suggest that the presentation of random non-sequential triple vibrotactile cues (an overall 23.9 percent reduction in braking RT relative to the no-warning baseline) was just as effective as the use of directional triple vibrotactile cues (an overall 23.5 percent reduction in braking RT relative to the no-warning baseline in Experiment 1) in facilitating a driver’s response to a critical closing-in event. As in the case of Experiment 2, it is possible that the straightforward task design in which tactile cues always signalled a frontal braking event (as used in the present study) might have mitigated the benefits of directional information inherent in the cues. We anticipate that an additional advantage would be observed from the directionality information built-in in these cues in those scenarios where multiple tactile cues are used to signal different events. It is also possible that the directionality effect, if any, will be more prominent if a

HO ET AL.: REORIENTING DRIVER ATTENTION WITH DYNAMIC TACTILE CUES

93

blocked design was used instead of the current randomized design. This suggestion is based on the assumption that a blocked design may give rise to a consistent mapping of specific cue type with the target event. The findings of the present study are consistent with earlier studies showing that dynamic tactile information can be used to reorient visual attention (e.g., [22]). In particular, dynamic motion processing in the proximal sense of touch and the distal sense of vision appear to be linked. While the present pattern of results did not seem to favour directional triple cuing over random triple cuing, the findings reported here demonstrate the potential benefits arising from triple over single cuing. Note that in the current design, the triple cues were presented consecutively in time (i.e., there were no temporal gaps between consecutive vibrations). It would therefore be interesting in future experiments to investigate what would happen when three temporally separated cues are presented from the same location. This will help answer the questions as to whether the present results demonstrate a benefit of triple over single cuing, or whether instead our results should be attributed to the advantage that may be associated with more spatially distributed, as compared to more narrowly focused cuing. Given that the effectiveness of vibrotactile cuing is influenced by a number of factors including the body location being stimulated and the force of the tactor loading on skin surface, it would appear to be advantageous to introduce moving vibrotactile warnings using a matrix of tactile devices that cover a large surface area in touch with the user. In that case, the intensity of each segment of vibrations can possibly be lower than when stimulating the whole matrix of tactile stimulators at high intensity. It will minimize any potential misses of vibrotactile cuing due to a variety of factors such as driving postures and anthropometric characteristics. Taken together, the results reported in the present study provide evidence supporting our original hypothesis that the presentation of tactile warning signals with inherent motion information could facilitate drivers’ speeded braking responses to impending closing-in events. That is, the presentation of dynamic tactile cues offers added benefits over and above static tactile information in terms of alerting and orienting a driver’s attention to a dynamic driving event. Here we have demonstrated that the crossmodal links between touch and vision in motion processing can potentially be utilized in car interface design by embedding approach information in the warning signals themselves. In our future research we hope to examine and fine-tune the properties of this new class of warning signals using more realistic driving simulator setup.

[2]

ACKNOWLEDGMENTS

[20]

This work was supported by a grant (EP/J008001/1) from the Engineering and Physical Sciences Research Council (EPSRC) to Rob Gray and Charles Spence.

REFERENCES [1]

C. M€ uller and G. Weinberg, “Multimodal Input in the Car, Today and Tomorrow,” IEEE MultiMedia, vol. 18, no. 1, pp. 98-103, Jan. 2011, doi:10.1109/mmul.2011.14.

[3]

[4]

[5] [6] [7]

[8]

[9]

[10]

[11]

[12]

[13]

[14] [15]

[16] [17] [18] [19]

[21]

[22]

C. Spence, “Drive Safely with Neuroergonomics,” The Psychologist, vol. 25, no. 9, pp. 664-667, 2012. C.A. Pickering, “The Search for a Safer Driver Interface: A Review of Gesture Recognition Human Machine Interface,” IEEE Computing and Control Eng., vol. 16, no. 1, pp. 34-40, Feb./Mar. 2005, doi:10.1049/cce:20050109. € uner, C. Stiller, and K. Redmill, “Systems for Safety and U. Ozg€ Autonomous Behavior in Cars: The DARPA Grand Challenge Experience,” Proc. IEEE, vol. 95, no. 2, pp. 397-412, Feb. 2007, doi:10.1109/jproc.2006.888394. M.A. Regan, J.D. Lee, and K.L. Young, Driver Distraction: Theory, Effects, and Mitigation. CRC Press, 2009. C. Ho and C. Spence, The Multisensory Driver: Implications for Ergonomic Car Interface Design. Ashgate, 2008. J.B.F. Van Erp and H.A.H.C. Van Veen, “Vibrotactile in-Vehicle Navigation System,” Transportation Research Part F: Traffic Psychology and Behaviour, vol. 7, no. 4/5, pp. 247-256, 2004, doi:10.1016/j.trf.2004.09.003. C. Ho, N. Reed, and C. Spence, “Assessing the Effectiveness of “Intuitive” Vibrotactile Warning Signals in Preventing Frontto-Rear-End Collisions in a Driving Simulator,” Accident Analysis and Prevention, vol. 38, no. 5, pp. 989-996, 2006, doi:10.1016/ j.aap.2006.04.002. R. Mohebbi, R. Gray, and H.Z. Tan, “Driver Reaction Time to Tactile and Auditory Rear-End Collision Warnings while Talking on a Cell Phone,” Human Factors, vol. 51, no. 1, pp. 102-110, 2009, doi:10.1177/0018720809333517. C. Ho and C. Spence, “Assessing the Effectiveness of Various Auditory Cues in Capturing a Driver’s Visual Attention,” J. Experimental Psychology: Applied, vol. 11, no. 3, pp. 157-174, 2005, doi:10.1037/1076-898X.11.3.157. L.M. Brown, S.A. Brewster, and H.C. Purchase, “Multidimensional Tactons for Non-Visual Information Presentation in Mobile Devices,” Proc. MobileHCI ’06, pp. 231-238, 2006, doi:10.1145/1152215.1152265. L.A. Jones, B. Lockyer, and E. Piateski, “Tactile Display and Vibrotactile Pattern Recognition on the Torso,” Advanced Robotics, vol. 20, no. 12, pp. 1359-1374, 2006, doi:10.1163/ 156855306778960563. F. Leo, V. Romei, E. Freeman, E. Ladavas, and J. Driver, “Looming Sounds Enhance Orientation Sensitivity for Visual Stimuli on the Same Side as Such Sounds,” Experimental Brain Research, vol. 213, no. 2-3, pp. 193-201, 2011, doi:10.1007/ s00221-011-2742-8. R. Gray, “Looming Auditory Collision Warnings for Driving,” Human Factors, vol. 53, no. 1, pp. 63-74, 2011, doi:10.1177/ 0018720810397833. C. Ho, C. Spence, and R. Gray, “Looming Auditory and Vibrotactile Collision Warnings for Safe Driving,” Proc. Seventh Int’l Driving Symp. Human Factors in Driver Assessment, Training, and Vehicle Design, pp. 551-557, 2013. D.A. Hall and D.R. Moore, “Auditory Neuroscience: The Salience of Looming Sounds,” Current Biology, vol. 13, no. 3, pp. R91-R93, 2003, doi:10.1016/s0960-9822(03)00034-4. C. Ho and C. Spence, “Using Peripersonal Warning Signals to Orient a Driver’s Gaze,” Human Factors, vol. 51, no. 4, pp. 539-556, 2009, doi:10.1177/0018720809341735. J.C. Stevens and K.K. Choo, “Spatial Acuity of the Body Surface over the Life Span,” Somatosensory and Motor Research, vol. 13, no. 2, pp. 153-166, 1996, doi:10.3109/08990229609051403. J.B.F. Van Erp, H.A.H.C. Van Veen, C. Jansen, and T. Dobbins, “Waypoint Navigation with a Vibrotactile Waist Belt,” ACM Trans. Applied Perception, vol. 2, no. 2, pp. 106-117, 2005, doi:10.1145/1060581.1060585. C. Ho, H.Z. Tan, and C. Spence, “Using Spatial Vibrotactile Cues to Direct Visual Attention in Driving Scenes,” Transportation Research Part F: Traffic Psychology and Behaviour, vol. 8, no. 6, pp. 397-412, 2005, doi:10.1016/j.trf.2005.05.002. J. Navarro, F. Mars, and J.-M. Hoc, “Lateral Control Assistance for Car Drivers: A Comparison of Motor Priming and Warning Systems,” Human Factors, vol. 49, no. 5, pp. 950-960, 2007, doi:10.1518/001872007x230280. R. Gray and H.Z. Tan, “Dynamic and Predictive Links between Touch and Vision,” Experimental Brain Research, vol. 145, no. 1, pp. 50-55, 2002, doi:10.1007/s00221-002-1085-x.

94

Cristy Ho received the BCogSc degree from the University of Hong Kong in 2001, the MSc degree in human-computer interaction with ergonomics from the University College London, United Kingdom, in 2003, and the DPhil degree in experimental psychology from the University of Oxford, United Kingdom, in 2006. She is currently a postdoctoral research scientist at the Crossmodal Research Laboratory, University of Oxford. She received the American Psychology Association’s New Investigator Award in Experimental Psychology: Applied, in 2006.

Rob Gray received the BA degree in psychology from Queens University in 1993, and the MS and PhD degrees in psychology from York University, Canada, in 1995 and 1998, respectively. He was a research scientist with Nissan Cambridge Basic Research from 1998 to 2001 before joining the Department of Applied Psychology at Arizona State University. From 2006 to 2010, he served as department head of the Department of Applied Psychology and as a Research Psychologist for the United States Air Forces. He is currently a senior lecturer in the School of Sport and Exercise Sciences at the University of Birmingham, United Kingdom. He is the author of more than 50 published refereed journal articles and chapters. He serves as an associate editor for the Journal of Experimental Psychology: Human Perception and Performance and is an editorial board member for Human Factors. In 2007, he was awarded the Distinguished Scientific Award for early career contribution to Psychology from the American Psychological Association and the Earl Alluisi Award for early career achievement in the field of applied experimental & engineering psychology.

IEEE TRANSACTIONS ON HAPTICS,

VOL. 7, NO. 1,

JANUARY-MARCH 2014

Charles Spence received the PhD degree in experimental psychology from the University of Cambridge, United Kingdom, in 1995. He is a university professor in the Department of Experimental Psychology at the University of Oxford. He is the head of the Crossmodal Research Laboratory in the Department of Experimental Psychology, University of Oxford. He has published more than 500 articles in top-flight scientific journals over the last decade. He has been awarded the 10th Experimental Psychology Society Prize, the British Psychology Society: Cognitive Section Award, the Paul Bertelson Award, recognizing him as the young European Cognitive Psychologist of the Year, and the Friedrich Wilhelm Bessel Research Award from the Alexander von Humboldt Foundation in Germany. He is currently an associate editor for Multisensory Research. " For more information on this or any other computing topic, please visit our Digital Library at www.computer.org/publications/dlib.