WalkingVibe: Reducing Virtual Reality Sickness and Improving Realism while Walking in VR using Unobtrusive Head-mounted Vibrotactile Feedback

Intro clip:

We investigate ways to improve the Virtual Reality (VR) experience of walking, which is arguably the most fundamental locomotive experience in real life. A key challenge towalking in VR is that motion simulation induces VR sickness, with symptoms such as headache, nausea, dizziness, disorientation, and fatigue.

WalkingVibe holds 2-sided vibrotactile design. The motors are attached on the elastic band and controlled by microcontroller such as Aruduino Nano.

To balance effectiveness with unobtrusiveness, we looked for vibrotactile frequencies that have good vestibular response and low perceived annoyance. Prior studies have shown that the maximal vestibular response to bone-conducted vibration is 200-400Hz. Moreover, perceived annoyance and discomfort increases linearly with vibration frequency and that frequencies above 150Hz should not be used for vibrating the head. We therefore selected the frequency of 150Hz in our vibrotactile design.

We aimed to reduce VR sickness and discomfort while improving the realism of walking in VR, by exploring different vibrotactile stimulation designs that can be readily streamlined into a VR headset. We conducted a 240-person study to compare total 8 feedback conditions by combined different visual, audio and tactile stimulation while users seated statically and navigated passively through VR environment.

Feedback Conditions

System Design

We describe the implementation of visual and tactile feedback provided in our study.

Visual feedback

To decrease the predictability found in VR experiences, we extended the prior works’ approach for supporting various moving speeds, and also needed to examine the effects of the stride frequency in terms of speed for generating appropriate timings of the vibration cues. As striding behavior varies per person, we used a polynomial regression model for translating speed to stride length, which can then be translated back to frequency through arithmetic division. While previous works only described the relationship between speed and vertical translation by up to 2.2m/s; no other work had covered this relationship beyond walking speed. Therefore, we first recorded and tracked the HMD positions of three users with different walking and running speeds, and then generated a linear model for amplitude multiplication.

Tactile feedback

Due to varying head sizes among users wearing the VR sys-tem’s head-mounted display (HMD), we attached Velcro stripsonto a headband that can be adjusted for individual fit and achieve flexible positioning from our prototype’s haptic vibrations. On the outer rim of the headband, we added aplastic strip layer and sewed together an array of cells of fixed lengths of 15 mm. Each cell has an opening on the top of the headband, which provides an enclosure for easy installation of the vibration motors. For vibrations, we used Parallax servos —– 12 mm coin-style, 3.3 V/90 mA, freq=9000 rpm — that operate at 150 Hz, which lies within the suggested range of 32 to 150Hz for preventing head discomfort from vibrations and whose amplitudes can be controlled by voltage. To control the servos, we placed an Arduino Nano board on top of the HMD. This board was connected to the computer via the HMD’s USB-C port, and listened to signals sent from the controlled VR application that was written in Unity 2019.1.4f1.

The setup for each category of tactile feedback:  (a) 2-sided vibration, (b) backside vibration, (c) 2-sided tapping feedback replicated from PhantomLegs project.


During the experiment, participants were seated statically and passively walked through the virtual environment. For the study task, participants passed through virtual checkpoints that were scattered in three designed VR scenes: a city, a forest, and a science fiction-themed passage. In total, each participant passed through 54 checkpoints in their VR session.

Screenshots of three scenes and paths of the virtual walkingenvironment: (a)(d) the city, (b)(e) the forest, and (c)(f) the sci-fi Passage.

We had three measurements in our 240-person, 8-condition between-subject study, including Simulator Sickness Quetionnarie (SSQ), Percieved Discomfort (DS) and Walking Realism. Our general hypothesis for the study was that compared to non-haptic stimulation,tactile feedback can effectively reduce VR sickness and discomfort for passive VR walking experiences by employing a natural visual head-bobbing pattern.


We calculated the perceived discomfort score, relative SSQ (post-test SSQ - pre-test SSQ), and experience realism to further evaluate how different feedback affect the VR walking experiences.

Discomfort score (DS)

The average and standard deviation (in parentheses) of dis-comfort scores (from left to right) are: 1.81 (0.61), 1.53 (0.75), 1.70 (0.92),1.36 (0.53), 1.07 (0.62), 1.45 (0.63), 1.86 (0.74), 1.61 (0.75).

We analyzed the discomfort level by averaging all the participants’ discomforts cores—–which were recorded at the same checkpoint—–as the overall discomfort level for that checkpoint under a specific feedback. In total, there was 27 averaged discomfort values for each condition. We performed such an analysis to avoid the impact of cumulative effect over time.

Based on these results, we employed a non-parametric Kruskal-Wallis test for determining whether there was any significant difference among the data. Our results demonstrated thatthere were indeed significant differences between the conditions (df=7; p=0.001346<0.05). We then ran a post-hoc pair-wise Conover test to compare the different paired conditions. Our results demonstrated that 2-sided vibration (synchronized) feedback received significantly less perceived discomfort compared to all other conditions. Furthermore, 2-sided tapping (synchronized) feedback had significantly lower discomfort than visual-only (without head-bobbing) feedback.

Relative SSQ

The average and standard deviation (in parentheses) of relative sickness score (from left to right) are:  22.19 (18.60), 18.70 (19.17), 18.20 (15.47), 6.73(15.46), 9.35 (16.75), 9.60 (14.18), 16.95 (12.16), 14.34 (20.55).

Our results demonstrated that there was significant differences among all feedback conditions by running one-way ANOVA: F(7, 29)=3.0636;p=0.0042<0.05. Specifically, three tactile conditions (i.e.,2-sided tapping (synchronized), 2-sided vibration (synchronized and random)) significantly reduced participants’ VR sickness—compared with two visual-only (with and without head-bobbing), audio, and backside vibration (synchronized) conditions—–by running post-hoc pairwise Tukey’s HSD test. The significances are shown in figure above.

Walking realism

The average and standard deviation (in parentheses) of realism score (from left to right) are: 4.17 (1.86), 4.93 (2.08), 5.70 (2.31), 4.27(2.03), 5.67 (2.08), 3.50 (1.69), 4.30 (2.31), 4.23 (1.98)

There were significant differences in terms of perceived realism under the feedback conditions by employing one-way ANOVA: F(7, 29)=4.1186; p=0.0003<0.05. Two feedback conditions—–audio and 2-sided vibration (syn-chronized)—–provided significantly higher realism than most of the other conditions except for visual-only feedback: 2-sided vibration (synchronized) vs. visual-only: 0.093; audio vs. visual-only: 0.095. We also conducted pair-wise comparisons as evaluated by a Tukey’s HSD post-hoc test. The significances are shown in figure above.

Please refer to our paper, video and talk for detailed information.

Publication / Media