Keep your finger on the pulse: Better rate perception and gap detection with vibrotactile compared to visual stimuli

Abstract

Important characteristics of the environment can be represented in the temporal pattern of sensory stimulation. In two experiments, we compared accuracy of temporal processing by different modalities. Experiment 1 examined binary categorization of rate for visual (V) or vibrotactile (T) stimulus pulses presented at either 4 or 6 Hz. Inter-pulse intervals were either constant or variable, perturbed by random Gaussian variates. Subjects categorized the rate of T pulse sequences more accurately than V sequences. In V conditions only, subjects disproportionately tended to mis-categorize 4-Hz pulse rates, for all but the most variable sequences. In Experiment 2, we compared gap detection thresholds across modalities, using the same V and T pulses from Experiment 1, as well as with bimodal (VT) pulses. Visual gap detection thresholds were larger (3[Formula: see text]) than tactile thresholds. Additionally, performance with VT stimuli seemed to be nearly completely dominated by their T components. Together, these results suggest (i) that vibrotactile temporal acuity surpasses visual temporal acuity, and (ii) that vibrotactile stimulation has considerable, untapped potential to convey temporal information like that needed for eyes-free alerting signals.

Keywords: Interval timing; Multisensory integration.; Rate perception; Temporal perception; Temporal variability; Vibrotactile.

Fig. 1

Apparatus for stimulus presentation. The LED and tactor were embedded in a bespoke hand cradle and used for presentation of visual and vibrotactile stimuli, respectively. Multiple interchangeable components in the hand cradle allowed us to accommodate hands of different widths and finger lengths in order to ensure that the tip of a subject’s index finger was positioned atop the vibrating tactor.

Fig. 2

In Experiment 1, series of 50-ms pulses (grey vertical bars, A & B) were presented on each trial. The stimulus terminated when either 10 pulses had been presented or if the subject made a response. Pulse sequences presented in the 0% noise condition (A) were isochronous, with each IPI fixed at a nominal IPI for the given rate (C). On trials with noise, each nominal IPI in the trial sequence was independently perturbed by a random variate (B). Scaling noise by the nominal rate produced IPI sampling distributions for 4-Hz and 6-Hz trials that were unequal in variance (D-G). Sampling distributions were truncated to prevent IPIs $\le 1$ ms. Note: Diagrams of pulse sequences (A & B) are not drawn to scale.

Fig. 3

Sensitivity (A) and decision criterion (B) by noise level for each modality. On average, subjects were better able to categorize rate information conveyed by T pulse sequences than by V pulse sequences. Subjects’ decision criteria varied between modalities, revealing they were more likely to erroneously categorize V sequences as “fast”. Error bars reflect 95% confidence intervals, n=28.

Fig. 4

Error rate for each combination of modality and noise level. A. Subjects had similar accuracy on 4-Hz and 6-Hz T trials. B. Subjects made more errors on 4-Hz V trials than 6-Hz V trials, that is, they were more likely to label a V pulse rate as “fast” overall, regardless of nominal rate. Error bars reflect within-subject standard error, n=28.

Fig. 5

Diagram of the trial structure in Experiment 2. Subjects received either a stimulus comprising a single, continuous pulse ($\mathit{S}{\mathit{P}}_{\mathit{i}}$) followed by a double-pulse stimulus ($\mathit{D}{\mathit{P}}_{\mathit{i}}$; top), or $\mathit{D}{\mathit{P}}_{\mathit{i}}$ followed by $\mathit{S}{\mathit{P}}_{\mathit{i}}$ (bottom). The IPI ${\mathit{X}}_{\mathit{i}}$ separating pulses in $\mathit{D}{\mathit{P}}_{\mathit{i}}$ ranged 2-32 ms. All pulses presented in a single trial came from the same modality condition (V, T, or VT). After the second stimulus, subjects had up to 2 s to indicate whether $\mathit{D}{\mathit{P}}_{\mathit{i}}$ had occurred first or second

Fig. 6

Psychometric modeling of Experiment 2 data. A: Group-level logistic psychometric functions (PFs) fit using maximum likelihood estimation. Ribbons around each PF represent the PF’s standard error. Points show the mean measured accuracy in each condition; error bars around each point represent within-subject standard error. Dashed vertical lines denote the estimated sensory threshold in each modality, which we defined at 76% correct ($d^{\prime }=1$; black dashed horizontal line). B: Average thresholds and bootstrapped standard errors for V, T, and VT stimuli.

Fig. 7

Correlation of subject-level unimodal and bimodal performance in Experiment 2. A: Tactile vs. bimodal accuracy. B: Visual vs. bimodal accuracy. Points are mean accuracy for individual subjects. The dashed black line represents the value expected if bimodal performance (on the vertical axis) were determined entirely by the unimodal performance (on the horizontal axis). Also shown in each panel is the 95% confidence region around the maximum likelihood line of best fit.