Effectiveness of Tactile Warning and Voice Command for Enhancing Safety of Drivers

Safety is impaired when drivers are required to perform main driving task (tracking of own car, distance maintenance between own car and a leading car, and response to target objects) and secondary task simultaneously, for example, responding to target cars on the road while operating in-vehicle equipment. A two-factor (presence or absence of tactile warning by input modality (no secondary task, voice command for a secondary task, and manual input for a secondary task)) within-subject design of ten licensed males was used to investigate how to compensate for safety impairments (decreased performance of a main and a secondary task such as increased tracking error during driving or increased reaction time to target cars on the road). We investigated whether the use of tactile warnings transmitted via left and right thighs for detecting road objects and voice command to operate in-vehicle equipment could compensate for safety impairments such as the increased reaction time to target cars on the road, the increase of detection error of target cars, or increased tracking error in driving. The accuracy and speed of responses to target cars encountered during driving were reduced when a driver was asked to perform the main and the secondary task simultaneously compared to situations performing only the main driving task (tracking, distance maintenance, and response to target cars). The availability of a tactile warning system for road objects compensated for these diminished performance measures, including slower response times and the increased detection error of target cars. Likewise, voice command contributed to enhanced performance of the main driving task such as decrease of tracking error.

Based on the literature review on voice command above, 89 we assumed that voice command was more effective for a 90 secondary task to improve the performance decrement of a 91 main driving task as compared to manual input for a sec-92 ondary input. 93 Jung et al. [31] proposed a voice interface with touch 94 pad interactions as a superior method for accomplishing 95 secondary tasks during driving. Although the findings of Jung 96 et al. [31] demonstrated the advantages of a secondary task 97 interface that utilized both voice and tactile senses, they inves-98 tigated only the effect of voice command and tactile display 99 on the secondary task performance and did not explore how 100 these modifications affected driving itself or hazard detection 101 while driving. In other words, they did not examine the effect 102 of tactile warning on both driving task and secondary task. 103 Based on the discussion above, we attempt to explore the 104 effectiveness of tactile warning and voice command within 105 the framework of S-C-R compatibility [22], [23]. The ratio-106 nale of using voice command and tactile warning is summa-107 rized as Hypotheses 1-5 below (see Figure 1). The cogni-108 tive information required to process a target object includes 109 its perception and recognition followed by the appropriate 110 motor responses. According to the S-C-R compatibility prin-111 ciple, performance will be enhanced by avoiding interference 112 between stimulus input modalities, between cognitive (cen-113 tral) processing modalities, and between response (output) 114 modalities. 115 This study focused on ways to improve driving efficiency 116 (safety) by using tactile warning and voice command when a 117 secondary task (e.g., operating in-vehicle equipment) inter-118 fered with the main driving task (tracking, distance main-119 tenance, and response to target cars). Specifically, while 120 verifying Hypotheses 1-5 below, we investigated how the 121 simultaneous use of tactile warnings to detect target cars and 122 voice command to execute a secondary task contributed to 123 compensate for the impaired performance that resulted from 124 performing the two tasks simultaneously to provide some 125 implications for enhancing safety by means of tactile warning 126 and voice command. 128 As mentioned above, tactile warning was superior to visual 129 or auditory warning in response speed and accuracy because 130 of lack of interference of the stimulus input modalities like 131 stimulus (input) modality interference of visual warning with 132 a visual target or that of auditory warning with noise in 133 traffic environment [14], [15], [17]. In Hypothesis 1, there-134 fore, we assume that tactile warning would promote rapid 135 responses to target cars on the road because of the absence of 136 the interference between stimulus modalities associated with 137 a visual object (visual target car and visual warning to this 138 target). 139 Avoidance of stimulus modality competition between 140 visual warning and visual attention to a target car by tactile 141 warning does not mitigate other activities such as track-142 ing (visual stimulus modality) and maintenance of distance 143 between own car and leading car (visual stimulus modal-144 ity) necessary for the main driving task, because resource 145 competition for central processing among the three spatial 146 activities (tracking, distance maintenance, and responses to 147 target cars) during the main driving task cannot be mitigated 148 by avoidance of stimulus modality competition by tactile 149 warning. Therefore, it is expected that tactile warning does 150 not improve the performance of tracking task and mainte-151 nance of appropriate distance between own and the leading 152 cars (Hypothesis 2). It must be noted that Hypotheses 1-2 are 153 related to the main driving task (tracking, distance mainte-154 nance, and response to a target car) and tactile warning.

155
There is no resource competition for central processing 156 between a secondary graphical user interface (GUI) task 157 (verbal processing) and a reaction task to target cars on 158 the road (spatial processing), because the former is verbal 159 and the latter corresponds to a spatial task. Therefore, it is 160 expected that fast spatial response to target cars on the road 161 by tactile warning will enhance the performance of a GUI 162 task irrespective of whether a GUI task is performed either 163 manually or with voice command (Hypothesis 3).

164
The S-C-R compatibility principle also assumes that simul-165 taneous performance of the main driving task and the 166 secondary task with manual responses represents a case of 167 interfering response modalities and such an interference will 168 lead to impaired performance (safety), including an increase 169 in the tracking errors and a decrease in the fraction of time 170 spent at the appropriate distance from the leading car. There-171 fore, it is expected that the use of voice command for a 172 secondary task would eliminate the interference between the 173 two manual response modalities (manual driving and manual 174 response to a GUI task) and thus might compensate for perfor-175 mance (safety) impairments associated with the main driving 176 task in a multi-task situation. When voice commands are 177 used to perform a secondary task, this interference does not 178 occur. Therefore, we anticipate that voice command results 179 in improvements in efficiency of the main driving task in a 180 multi-task situation. This hypothesis is also supported by past 181 findings [24], [26], [27], [28], [29], [30] that voice commands 182 were effective under a multi-task situation. Therefore, it is 183 expected that avoidance of response modality competition 184 between manual input to a GUI task and manual tracking, 185 manual distance maintenance and manual response to target 186 cars enhances performances of tracking, distance mainte-187 nance, and response to target cars on the road and contributes 188 to the improvement of the three tasks related to driving 189 (Hypothesis 4).

190
It is also expected that avoidance of the response modality 191 competition above by voice command enhances performance 192 of a GUI task itself (Hypothesis 5). This hypothesis is con-193 cerned with only the secondary GUI task and voice command. 194

196
Ten healthy 21-23-year-old males (graduate or undergradu-197 ate students) were recruited from Dept. of Intelligent Sys-198 tems, Okayama University and participated in the experi-199 ment. All participants were licensed drivers for 1-5 years. All 200 participants provided written informed consent after receiv-201 ing a brief explanation of the aim and content of the experi-202 ment. The experiment was approved by the Ethical Commit-203 tee, Department of Intelligent Mechanical Systems, Okayama 204 University, Japan (Approval No.2019-sys-04). Although only 205 male participants took part in the experiment, we judged that 206 gender difference would not affect the results and the research 207 hypotheses 1-5.   to Murata et al. [13]. The SOA of 1 s meant that a warn-231 ing was presented to the participant 1 s before a target car 232 appeared.

233
The approximate layouts of the in-vehicle displays and the 234 GUI task are shown in Figure 2. As Japanese driver's seats are 235 located on the right, the driver seat was located on the right as

250
Participants were required to perform a simulated main 251 driving task (including tracking, distance maintenance, and 252 response to target cars) with or without a secondary task (GUI 253 task) (see Figure 3) as accurately and rapidly as possible.

254
Our study involved two tasks, including the main driving 255 task (tracking, distance maintenance, and response to target 256 cars) and the secondary task (GUI task). There were three 257 lanes on the display of driving simulator. In the main driving 258 task, participants were required to drive a middle lane, follow 259 a leading car and suppress the deviation of own car from 260 the center of the 2nd (center) lane as much as possible. 261 Speed was to be maintained at a constant 80 km/h based on 262 readouts provided by a display. Participants were instructed 263 to maintain a distance of 60-100 m between their cars and 264 the leading car. Each participant was informed that he was 265 maintaining an appropriate distance by the display of a green 266 rectangular frame around the image of the leading car. If the 267 distance between the cars fell below 60 m or exceeded 100 m, 268 the participant was informed of this change in status by a 269 change in the color of the rectangular frame.

270
For the detection task of target cars, each participant was 271 instructed to identify a specified car that was displayed on the 272 in-vehicle monitor located near the display for the GUI task 273 (secondary task) by pressing a button located on the right side 274 of the steering wheel (see Figure 2). The decision to place the 275 in-vehicle monitors at the side mirrors near the display of the 276 GUI task was based on previous studies [32], [33]. Potentially 277 dangerous situation occurs also in front of the vehicle. The 278 reaction to a front or rear object is shown to be based on nearly 279 the same protocol [34]. Moreover, we judged that the attentive 280 level in front of own car could be assessed by the percentage 281 of time spent at the appropriate distance. Therefore, this study 282 dealt with only the rear target. The detection task of target 283 cars behind own car proceeded as follows: While driving in 284 the second (middle) lane, participants encountered a white, 285 a black, or a red sedan that appeared randomly from the rear 286 of the participant's car in either the first (left) or the third 287 (right) lane and that eventually passed the participant's car. 288 The approaching cars caught up with the participant's car 289 VOLUME 10, 2022 FIGURE 4. Explanation of the overlaps used to examine the impact of interference on the main task (driving task and detection task of target cars) and the GUI task (secondary task). 6 seconds after their appearance. These target cars appeared driving task (tracking, distance maintenance, and response 328 to target cars). Significant overlap of these two tasks might 329 impair driving safety. Previous studies (for example, [21], 330 [26], [35]), however, did not control for the potential overlap 331 of two tasks. Wickens et al. [21] and Murata [26] used a 332 continuous task as a main task and a discrete task as a sec-333 ondary task, and investigated the effects of voice command 334 or manual input to a secondary task on the dual task perfor-335 mance. The dual-task condition of these studied included both 336 temporal overlap of the main (continuous) and the secondary 337 (discrete) task (both tasks were simultaneously conducted) 338 and temporal non-overlap of both tasks (only a main task 339 was conducted), and regarded both temporal overlap and non-340 overlap as a dual task condition. It is desirable that only the 341 temporal overlap is analyzed as a dual-task condition. There-342 fore, our study used an experimental design that controlled 343 the temporal overlap of the main and the secondary task as 344 shown in Figure 4. 345 The experiments were carried out for the following four 346 conditions that consisted of the presence or absence of tactile 347 warning and input modality of secondary task (voice com-348 mand or manual input). In other words, experiments included 349 (I) a simultaneous dual-task situation with tactile warning 350 for the detection of target cars and voice command for the 351 secondary task, (II) a simultaneous dual-task situation with 352 tactile warning for the detection of target cars and manual 353 input for the secondary task, (III) a simultaneous dual-task 354 situation without tactile warning for the detection of target 355 cars and with voice command for the secondary task, and (IV) 356 a simultaneous dual-task situation without tactile warning 357 for the detection of target cars and with manual input for 358 the secondary task. To characterize the safety impairments 359 resulting from the overlap of the main and the secondary task, 360 we controlled the overlap and non-overlap of two tasks as 361 shown in Figure 4. It must be noted that each experiment 362 included both the overlap and the non-overlap condition of 363 the main and the secondary task as shown in Figure 4. 364 Although the above four experiments were conducted, 365 there were actually the following six conditions (presence 366 or absence of tactile warning (two levels) by input modal-367 ity (three levels: no secondary task (non-overlap), voice 368 command for the secondary task, and manual input for the 369 secondary task): (i) non-overlap with tactile warnings for 370 detecting target cars, (ii) non-overlap without tactile warnings 371 for detecting target cars, (iii) overlap with tactile warnings for 372 detecting target cars and voice command for the secondary 373 task, (iv) overlap with tactile warnings for detecting target 374 cars and manual input for the secondary task, (v) overlap 375 without tactile warnings for detecting target cars but with 376 voice command for the secondary task, and (vi) overlap with-377 out tactile warnings but with manual input for the secondary 378 task.

380
The experimental design included two independent within-381 subject factors. The first factor (independent variable) was the 382 presence or absence of tactile warnings for detection of target 383 cars. The second factor (independent variable) was the input 384 modality (no secondary task, voice command for the sec-385 ondary task, and manual input for the secondary task). The  The practice data were excluded from the further analysis.  The participant was also required to maintain an appropri-448 ate 60-100 m distance between own car and the leading car. 449 This distance was also recorded every 100 ms. The percentage 450 of time in which an appropriate distance between the two cars 451 was maintained was calculated as the duration during which 452 appropriate car distance was maintained divided by the total 453 duration of the experiment (12 min).

454
From the perspective of road safety, conditions that facil-455 itate more rapid and accurate responses to any given road 456 hazard are desirable outcomes. Overlooking a hazard while 457 driving suggests an inattentive state and thus represents an 458 undesirable event. The time that elapsed from the appearance 459 of a target car on either the left or the right of the participant's 460 car until the response button was pressed was measured as the 461 reaction time.

462
A total of 72 cars emerged either to the left (36 cars) 463 or to the right (36 cars) of the participant's car during one 464 experimental session. Among these 72 cars, the target car (red 465 sedan) appeared randomly 12 times to the left and 12 times to 466 the right of the participant's car. The participant was required 467 to detect the target car while it was on the left or right display. 468 When the participant could not detect the target car until the 469 participant's car was overtaken by the target car, this was 470 regarded as a detection error. The percentage detection error 471 by each participant was calculated for either left or right 472 side of own car as the ratio of the number of target cars not 473 detected to the total number of target cars (n = 12).

474
The task completion time and the error trials were used as 475 performance measures in the GUI task. At the end of each 476 experiment, the participant was asked to provide a subjective 477 rating on his state of concentration while driving. The scores 478 were recorded on a five-point scale from (1) unable to con-479 centrate on driving at all under these conditions to (5) able 480 to concentrate sufficiently on driving under these conditions. 481 They were also asked to score their ease of situational aware-482 ness during a task under these conditions, with scores from 483 (1) indicating that it was very difficult to maintain situational 484 awareness to (5) indicating that it was very easy to maintain 485 situational awareness. It must be noted that the participants 486 were asked to report subjective ratings of concentration and 487 situational awareness not only for the four experimental 488 conditions (iii)-(vi) (overlap) but for the two experi-489 mental conditions (i)-(ii) (non-overlap (without secondary 490 task) with and without tactile warning) above mentioned 491 (see Figure 3).

FIGURE 5.
Tracking error recorded in the presence or absence of tactile warnings and input modality (no secondary task, secondary task with manual input, secondary task with voice command).

494
A two-way (the presence or absence of tactile warnings by 495 input modality (three levels: no secondary task, manual input 496 to a secondary task, and voice command to a secondary task)) 497 analysis of variance (ANOVA) was carried out on the tracking 498 errors. The results of the ANOVA are summarized in Table 1.

499
The ANOVA for the tracking error revealed only a significant 500 main effect of input modality, and no significant interaction 501 was detected.

502
In Figure 5, the tracking error is plotted as a function of the 503 presence or absence of tactile warnings to target cars on the 504 road and the input modality. Tracking error was significantly 505 impaired when the secondary task performed with manual 506 responses interfered with the performance of main driving 507 task (tracking, distance maintenance, and response to target 508 cars). Tracking error was reduced when the secondary task 509 was executed with voice command. As shown in Figure 5 and   Table 2. 520 TABLE 2. Results of two-way anovas conducted on the percentage appropriate car distance. FIGURE 6. Percentage of time during which appropriate distance from the leading car was maintained in the presence or absence of a tactile warning and input modality (no secondary task, secondary task with manual input, secondary task with voice command).
The ANOVA for this evaluation measure revealed only a 521 significant main effect of input modality, and no significant 522 interaction was detected. 523 Figure 6 shows the percentage of time spent at the appro-524 priate distance from the leading car as a function of the 525 presence or absence of tactile warning and the input modality. 526 When the performance of secondary task interfered with that 527 of the main driving task (tracking, distance maintenance, 528 and response to target cars), this specific safety measure 529 was impaired. Similar to our findings on tracking error, the 530 percentage of time spent at the appropriate distance from the 531 leading car improved when secondary task was performed 532 using voice command. The tactile warnings for the detection 533 of target cars in the main driving task also had no impact on 534 this evaluation measure (Table 2).

536
A two-way (the presence or absence of tactile warnings by 537 input modality (three levels: no secondary task, manual input 538 to a secondary task, and voice command to a secondary task)) 539 ANOVA was conducted on the reaction times to target cars 540 presented to both the right and left sides of the participant's 541 car. The results are summarized in Table 3. The ANOVA for 542 this evaluation measure revealed significant main effects of 543 presence or absence of tactile warning and input modality, 544 and no significant interaction was detected.  in Table 4. The ANOVA for this evaluation measure revealed 562 significant main effects of presence or absence of tactile 563 warning and input modality and a significant interaction.

564
In Figure 8, the detection error of target cars is plotted 565 as a function of presence or absence of tactile warning and 566 the input modality. The use of tactile warnings resulted in 567 major improvements of detection error for both conditions 568 with and without a secondary task. A significant interaction 569 between the presence or absence of tactile warnings and the 570 input modality can be interpreted as follows. While the input 571 modality (no secondary task, voice command, and manual 572 input) of a secondary GUI task did not affect the detection 573 error of target cars when the tactile warning was present for 574 the detection of target cars, the improvements in detection 575 error of target cars associated with the absence of tactile 576 warning were significantly larger (in order) in cases in which 577 VOLUME 10, 2022 FIGURE 8. Detection error of target cars in the presence or absence of tactile warning and input modality (no secondary task, secondary task with manual input, secondary task with voice command).
the secondary task was conducted with manual input, cases 578 in which the secondary task was conducted with voice com-579 mand, and when the secondary task was not conducted.  We evaluated the subjective ratings provided by the partic-609 ipants on conditions for secondary task that included tac-610 tile warnings with voice command, tactile warnings without 611 voice command (with manual input), no tactile warnings 612 but with voice command, and no tactile warnings without 613 voice command (with manual input) using a Scheffe's mul-614 tiple comparison (see Table 5(1)). The subjective ratings 615  TABLE 5. Results of non-parametric tests conducted on the subjective ratings on concentration during driving.

FIGURE 10.
Subjective ratings of participant's concentration while driving as a function of presence or absence of tactile warning and input modality (no secondary task, secondary task with manual input, secondary task with voice command).
for driver concentration were evaluated with a focus on the 616 presence and absence of tactile warning using a Wilcoxon 617 non-parametric test when no secondary task was executed 618 (see Table 5(2)). The subjective ratings on driver concen-619 tration were compared to one another for conditions that 620 included tactile warnings with secondary task, tactile warn-621 ings without secondary task, no tactile warnings with sec-622 ondary task, and no tactile warnings without secondary task 623 using a Scheffe's multiple comparison (see Table 5(3)).

624
The results of the analyses for the concentration ratings 625 are summarized in Table 5. Figure 10 corresponds to the 626 subjective concentration ratings plotted as a function of the 627 presence or absence of tactile warning and the input modal-628 ity (no secondary task, manual input to a secondary task, 629 and voice command to a secondary task). Similar results of 630 FIGURE 11. Subjective ratings of participant effort required to maintain situational awareness during driving as a function of presence or absence of tactile warnings and input modality (no secondary task, secondary task with manual input, secondary task with voice command). statistical tests performed to evaluate the subjective ratings 631 of situational awareness are shown in Table 6. In Figure 11, 632 these subjective ratings were plotted as a function of the 633 presence or absence of tactile warning and the input modality 634 (no secondary task, manual input to a secondary task, and 635 voice command to a secondary task).

636
For both sets of subjective ratings (i.e., driver concentration 637 ( Figure 10) and situational awareness (Figure 11)), several 638 important tendencies were observed. When secondary task 639 interfered with the performance of the main driving task 640 (tracking, distance maintenance, and response to target cars), 641 both driver concentration and situational awareness were 642 impaired. In such a situation, tactile warnings improved both 643 concentration and situational awareness. Moreover, concen-644 tration and situational awareness increased when tactile warn-645 ings were combined with voice command in the secondary 646 task.  Hypothesis 5 is concerned with voice command and the 700 GUI task, and predicts that avoidance of the response modal-701 ity competition between manual inputs in GUI tasks and 702 manual responses during driving will enhance performance 703 of a GUI task itself. Voice command did not lead to more 704 quick and accurate response in a GUI task than manual 705 input, which did not support Hypothesis 5 (see IV.RESULTS, 706 E and Figure 9). The GUI task must have been too sim-707 ple to demonstrate an advantage of voice command over 708 manual input. However, it must be noted that the advantage 709 of voice command over manual input appeared not in the 710 secondary GUI task ( Figure 8) and the response to a target car 711 (Figures 6 and 7) but in the tracking error ( Figure 4) and the 712 distance maintenance ( Figure 5). While Owens et al. [27], 713 Miller et al. [28], Alvarez et al. [29] command. This must indicate that voice command is not 718 always effective and that the effectiveness depend on the 719 situation. The future research should take the workload of a 720 secondary GUI task into account and examine whether the 721 advantage of voice command over manual input is observed 722 when the workload of GUI task is higher than that in this 723 study.

724
Tactile warnings contributed to more rapid and accurate 725 responses to target cars irrespective of absence or presence 726 of a secondary task (see Figures 7 and 8) and faster responses 727 in a GUI task (Figure 9) according to Hypotheses 1 and 3, 728 respectively. Hypothesis 3 is also regarded as representing an 729 interaction between the main driving task and the secondary 730 GUI task. While voice command contributed to reductions in 731 tracking errors and increased the percentage of time spent at 732 an appropriate distance from a leading car as shown according 733 to Hypothesis 4, voice command did not contribute to the 734 improvements of response speed and accuracy to target cars 735 during the main driving task.

736
As already mentioned, tactile warning improved speed and 737 accuracy of response to target cars and the speed of secondary 738 GUI task and voice command improved tracking error and 739 percentage of appropriate distance maintenance. In short, 740 tactile warning and voice command differently contributed 741 to improvements of impaired safety. It must be noted that 742 unlike these results, both voice command and tactile warning 743 contributed to enhance subjective ratings of concentration 744 and situational awareness during driving.

745
As far as this study is concerned, appropriate countermea-746 sures to prevent performance decrements potentially leading 747 to a crash are to prevent competitions of stimulus modalities 748 and competitions of central processing modalities by tactile 749 warning (based on Hypotheses 1 and 3) and to prevent com-750 petition of response modalities by voice command (based 751 on Hypothesis 4). This indicates that tactile warning and 752 voice command should be used for different effects based on 753 Hypotheses 1 and 3 (performance enhancement of responses 754 to target cars and secondary GUI task by tactile warning) 755 and Hypothesis 4 (performance enhancement of tracking and 756 distance maintenance by voice command), respectively.

757
The limitations of this study can be summarized as follows.

758
The detection of target cars in this study is different from 759 real-world situations to detect hazards. As a hazard appears 760 unexpectedly in a real-world situation, this study might not 761 reflect such a situation. However, it must be noted that the