The Answer Is Blowing in the Wind: Directed Air Flow for

Socially-acceptable Human-Robot Interaction

Vincent Zhang

1 a

, Natalie Friedman

1,2 b

, David Goedicke

1,2 c

, Dmitriy Rivkin

1 d

Michael Jenkin

1,3 e

, Xue Liu

1,4 f

and Gregory Dudek

1,4 g

Samsung AI Center, Montreal, Canada

Information Science, Cornell Tech, New York, U.S.A.

EECS, York University, Toronto, Canada

Computer Science, McGill University, Montreal, Canada

Keywords:

Human-Robot Interaction, Air-driven Haptic Cues.

Abstract:

A key problem for a robot moving within a social environment is the need to capture the attention of other

people using the space. In most use cases, this capture of attention needs to be accomplished in a socially

acceptable manner without loud noises or physical contact. Although there are many communication mecha-

nisms that might be used to signal the need for a person’s attention, one particular modality that has received

little interest from the robotics community is the use of controlled air as a haptic signal. Recent work has

demonstrated that controlled air can provide a useful signal in the social robot domain, but what is the best

mechanism to provide this signal? Here, we evaluate a number of different mechanisms that can provide this

attention-seeking communication. We demonstrate that many different simple haptic air delivery systems can

be effective and show that air on and air off haptic events have very similar time courses using these delivery

systems.

1 INTRODUCTION

The real world presents many challenges to a mobile

robot that are not encountered in a controlled lab-

oratory setting. In particular, in the real world the

motion plans of mobile robots are often obstructed

by humans and getting humans to move out of the

way can be challenging. Autonomous vehicles need

mechanisms to attract the attention of humans, per-

ceived as, “obstacles” in a socially acceptable man-

ner. Classic methods that are often found in industrial

settings, like beeping loudly or waiting patiently, are

either disruptive or ineffective (Wogollter, 2006) and

sometimes both when used in a social setting.

A classic example of this problem is the motion of

a robot through a crowd in a social setting as shown

https://orcid.org/0000-0003-1134-045X

https://orcid.org/0000-0003-4751-7739

https://orcid.org/0000-0002-4837-893X

https://orcid.org/0000-0003-4136-4831

https://orcid.org/0000-0002-2969-0012

https://orcid.org/0000-0001-5252-3442

https://orcid.org/0000-0001-5040-4925

Figure 1: A telepresence robot trying to make its way

through a group of individuals who are blocking its path,

but is blocked.

in Figure 1. For the robot to be able to make its way

to the goal, it must acquire the attention of the people

who are obstructing its path and then communicate to

106

Zhang, V., Friedman, N., Goedicke, D., Rivkin, D., Jenkin, M., Liu, X. and Dudek, G.

The Answer Is Blowing in the Wind: Directed Air Flow for Socially-acceptable Human-Robot Interaction.

DOI: 10.5220/0010136001060113

In Proceedings of the International Conference on Robotics, Computer Vision and Intelligent Systems (ROBOVIS 2020), pages 106-113

ISBN: 978-989-758-479-4

1.5

2.5

3.5

4.5

5.5

Focused

Present

Self Confi dent

Nervous

Expre ssive

Dramatc

Relaxed

Draws Others

Ta sk Ori ented

Poise

Smooth

Impa tient

Dictatorial

Memorable

Social Acceptability

7 point Likert scale

No Cue Air Cue

Figure 2: Results from a Wizard of Oz (WoZ) user study on the effect of using haptic air for robot to human interaction

(Friedman et al., 2020b). A questionnaire measured the appropriateness of the robot on a 7 point Likert scale. A score of 1 =

less focused, less present, less self-conﬁdent, etc. A score of 7 = more focused, more present, more self-conﬁdent, etc. Error

bars show standard errors. Haptic air cues made the robot seems more socially acceptable compared to a robot that did not use

this cue over a range of different measures. Results shown here are for the 23 adults (4 female) participants who completed

the questionnaire associated with the study. The mean age of the study participants was 25 (SD=5.8). Full details of the study

can be found in (Friedman et al., 2020b).

them its intention to move through the space. Captur-

ing the attention of users could be accomplished in a

number of ways, including through the use of acous-

tic, visual and haptic cues in various combinations.

But not all strategies are socially acceptable. For ex-

ample, loud noises are generally unacceptable in a

social setting and visual cues are ineffective except

when the robot is in direct line of sight. Cues based

on direct physical contact are likely to be inappropri-

ate in many social contexts. Touching or tapping is

generally a questionable behavior with strangers, es-

pecially in some cultures where physical contact is

frowned upon. Even the nature of acceptable touching

varies from culture to culture. For example, in some

cultures, it is acceptable to touch a stranger with an

open palm, but potentially rude to do it with a ﬁnger

tip. In the context of a robot, it is unknown if tapping

a person on the shoulder would be a socially appro-

priate strategy for an autonomous robot.

One potential alternative to an actual physical

touch is to use an alternative medium such as directed

air as a conduit for this haptic touch. In Friedman et

al. (2020a, b) a haptic air cue was integrated with an

audio cue and visual instructions in a robot interaction

experiment. This Wizard of Oz (WoZ) “in the wild”

experiment was centred around a robot attempting to

encourage participants in the space to move out of the

way as the robot followed a pre-computed path. This

study explored the use of air-based haptic cues com-

bined with audio and visual cues to both attract atten-

tion and then to communicate intent to participants

blocking the robot’s path. Providing the robot with a

mechanism to attract attention (haptic air and audio

cues) coupled with explicit instruction to the user to

move out of the way was found to be an effective strat-

egy to enable the robot to move through the crowd. In

this study participants also completed a questionnaire

that asked on a 7-point Likert scale (Burgoon et al.,

1998) questions related to the observer’s perception

of the robot’s behaviour. Figure 2 plots participant re-

sponses from the study where a score of one indicates

that the participant found the robot to be less focused,

less present, less self-conﬁdent, etc., while a score of

seven indicates that the participant found the robot to

The Answer Is Blowing in the Wind: Directed Air Flow for Socially-acceptable Human-Robot Interaction

107

(a) Funnel 1 (3D Printed) (b) Funnel 2 (3D Printed) (c) Common fan

Figure 3: Fan designs. (a) and (b) show the 3D printed cowls of Funnel 1 and Funnel 2. These are mounted to the basic fan

shown in (e).

be more focused, more present, more self-conﬁdent,

and so on. Each of the conditions showed a consis-

tent trend over the majority of the conditions with the

addition of air there is a trend for the robot to appear

more focused, more present, etc.

The work reported in Friedman (2020a, b) used a

very simple fan (shown in Figure 3(c)) to generate the

haptic air cue. This choice of fan design was one of

convenience. But is this the best mechanism for pro-

viding the most appropriate mechanism to generate a

haptic air cue? Here we explore a range of technolo-

gies that might be exploited in generating this cue.

2 RELATED WORK

The use of controlled air has been explored as an in-

teraction medium in the virtual reality/haptics space

in a number of different domains and applications.

Speciﬁc examples include the use of ultrasound trans-

ducers (Hoshi et al., 2009; Hasegawa et al., 2017)

and arrays of air jets (Suzuki and Kobayashi, 2005)

to generate local haptic events. With the goal of sim-

ulating large-scale wind sensations, there have also

been a number of efforts to build simulated wind en-

vironments including Mowaﬁ et al. (2015), Tolley et

al. (2019) and Verlinden et al. (2013). An air puff has

also been considered as a stimulus in other contexts,

for example to elicit responses from children suffering

from autism (Dakopolos and Jahromi, 2019). While

fans and compressed air are the most commonplace

technologies for creating controlled air ﬂows, they are

not the only options. Alternative approaches to sys-

tems based on mechanical fans and compressed air

exist as well. For example Sodhi et al. (2013) and

Gupta et al. (2013) describe approaches that use audio

subwoofer speakers and an air chamber to generate

ring vortex air patterns that are stable and have been

found to be capable of being directed out to 1.25m

from the emitter.

Although there exist a range of different technolo-

gies that might be exploited to generate a controlled

haptic air event, perhaps the most straightforward is

to use an electric fan coupled with some sort of fo-

cusing and aiming device to generate the haptic cue.

Even within this controlled design space there are a

range of different questions that must be addressed.

In particular, are there advantages in terms of shaping

the structure of the air column generated by a fan in

terms of presenting the cue to the observer? This is

the question we consider here.

3 FUNNEL DESIGNS AND

EVALUATION

An effective controlled air source should provide a

constant, non-turbulent ﬂow at a distance of at least a

meter to the human participant. Ensuring that the ﬂow

is non-turbulent allows the ﬂow to have a non-chaotic

structure when it strikes the participant. While a one

meter distance allows the robot to provide a haptic

cue without entering the personal space of the user

but at the same time being sufﬁciently close to the

user that origin of the haptic air event is clear. Fur-

thermore, the width of the haptic air cue should be

sufﬁciently narrow that it can be targeted at the per-

son or persons blocking the robot’s path. Narrower

cues allows selectivity in terms of the target partici-

pant. Clearly, the desired width of the air stream de-

pends on the number of people that the robot needs

to contact simultaneously. However, we believe that a

ROBOVIS 2020 - International Conference on Robotics, Computer Vision and Intelligent Systems

108

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

0.000

0.500

1.000

1.500

2.000

2.500

3.000

3.500

4.000

4.500

No Funnel

Funnel 1

Funnel 2

Distance (m)

Air Velocity (m/s)

Figure 4: Generated air velocity measured at different dis-

tances for the haptic air device designs shown in Figure 3.

Air velocities were measured using an anemometer (AP-

846A AOPUTTRIVER) at distances of 0.1m to 1.5m from

the driving fan.

narrower stream is preferable over a wider one, since

it provides the robot the option of alerting a single

individual. Groups of people may still be contacted

by moving the air stream from side to side. Given

these observations, here we evaluate the efﬁcacy of

mechanical structures (funnels) in improving the per-

formance of a ﬁxed-speed fan in terms of:

1. The width of the air stream in which a lower

width means higher performance.

2. The maximum human detection distance in

which a larger distance means higher perfor-

mance.

3. Human reaction time to “air on” and “air off”

events in which shorter is better.

For a given air source (the fan) a funnel mounted on

the air source may increase performance in terms of

condition 1 (width) by blocking air ﬂow at high off-

normal angles. Furthermore, it may also improve per-

formance under conditions 2 (detection distance) and

3 (reaction time) by increasing wind velocity. In or-

der to explore this hypothesis, a number of different

funnel structures were constructed and evaluated us-

ing the three criteria given above. These funnel de-

signs were circularly symmetric and the radius de-

scribed by two cubic polynomials, with continuous

curvature at the matching point and zero ﬁrst order

derivatives at the inlet and outlet. Several researchers

have suggested that this design maximizes air veloc-

ity at the outlet by minimizing the adverse pressure

gradient along the walls of the funnel (Morel, 1975;

Cattafesta et al., 2010). Combining the aforemen-

tioned constraint with a length ratio, contraction ratio

and match point location fully deﬁnes the shape of the

funnel.

Figure 5: A view from above of the experimental setup for

measuring the width of the air stream. Orange paper stream-

ers are suspended from a bar and viewed from above. De-

tected deﬂection of the streamers indicates the presence of

an air signal. The effective width of the air signal and its

fall off with width can be estimated from the change in de-

ﬂection as a function of horizontal distance.

In our case, the diameter of the inlet is determined

by the size of the fan (105 mm). As shown is Fig. 3,

both Funnel 1 and Funnel 2 have the same length ratio

as 1:1 and match point x

as 1/2. They only differ

in terms of contraction ratio. The contraction ratio of

Funnel 1 is 4:1, while Funnel 2 has a higher contrac-

tion ratio of 16:1. Both funnels were 3D printed using

ABS with a layer height of 100 µm.

4 MEASUREMENT OF AIR FLOW

We performed measurements of maximum air speed

(Fig. 4) and air stream width (Tab. 1) at several dis-

tances from the fan for No Funnel, Funnel 1 and Fun-

nel 2 air sources. An anemometer (AP-846A AOP-

UTTRIVER) was used to measure air speed. Fig. 4

plots air speed by distance for the various designs. As

shown in the ﬁgure, an air speed associated with the

fan was measurable out to 1.5m. Furthermore, the

measured air velocity of Funnel 1 was higher than No

Funnel at the exit, but the air ﬂow did not reach as far

as the latter. There was a marked decrease in the per-

formance with Funnel 2, which can be explained by

reduced air throughput caused by the smaller outlet

aperture.

Stream width was measured using a device com-

posed of a contiguous array of thin paper strips. The

device was positioned at 0.5m or 1.0m from the fan

with the strips aligned orthogonal to the direction of

air ﬂow (Fig. 5). The width of the area where strips

were lifted by air ﬂow was then measured. Tbl. 1

summarizes the measured air stream widths at 0.5m

The Answer Is Blowing in the Wind: Directed Air Flow for Socially-acceptable Human-Robot Interaction

109

Table 1: Measured width of perceptible air ﬂow channel at

distance of 0.5m and 1.0m away from the fan.

Distance No Funnel Funnel 1 Funnel 2

0.5m 49cm 15cm 20cm

1.0m 33cm 20cm 0cm

and 1.0m from the fan. The funnels both produced

a signiﬁcantly narrower air stream at 0.5m than No

Funnel. The stream width of Funnel 1 was also far

more consistent than No Funnel as the distance was

varied. The ﬂow produced by Funnel 2 was not de-

tectable by our device at a distance of 1.0m.

5 PERCEPTION OF AIR FLOW

A critical element for any haptic air delivery design

is its ability to deliver a perceivable airborne cue at

a socially acceptable distance. In other words, “did

you feel the robot tap your shoulder” using air ﬂow?

While the measurements performed in previous sec-

tions can help estimate the range of detectability of

a haptic air event, they do not provide much in-

sight about the ability of individuals to note the pres-

ence/absence of the haptic air stimulus and the corre-

sponding reaction time. This question was addressed

through a user study.

An experiment was performed to identify the max-

imum distance to which a subject could reliably detect

air ﬂow from the various designs. Fig. 6 shows the

basic setup. An observer sat in a chair at a measured

distance from the haptic air delivery device wearing

headphones that played music to obscure any acous-

tic information from the fan. The fan was left running

at all times and the air ﬂow was blocked/released us-

ing a large occlusion device. Observers were asked

whether they could feel the ﬂow, i.e. if the air stream

was ﬂowing or not at a range of distances. For each

test case, we used two measurements of 4 (Funnel 2)

or 5 (Funnel 1, no funnel) test subjects. Maximum

detectability was set to be the last measured distance

at which the participant was correct in identifying the

state of the haptic air device 80% of the time when

they were queried. As shown in Fig. 7, while air ﬂows

from No Funnel and Funnel 1 devices could be de-

tected reliably up to 2m, the air from Funnel 2 was

only detectable to 1m.

Once the maximum range at which observers

could reliably detect the presence/absence of the hap-

tic air stimulus was established, we explored the de-

tectability of ”air on” and ”air off” events and the de-

lay between the generation of an event (i.e. the inser-

tion or removal of the occlusion) and its perception by

the observer. Quantiﬁcation of this delay is of particu-

Figure 6: Quantifying air as a touch cue experimental setup.

Observers sat in a chair at a measured distance from the

haptic air device and estimated the time it took for them to

start and stop feeling the air signal.

No Funnel Funnel 1 Funnel 2

0.5

1.5

2.5

Max Sensi�vty

Distance (s)

Figure 7: Mean maximum sensitivity distance for the vari-

ous haptic air delivery device models. Error bars show stan-

dard errors.

lar importance in situations where the haptic air stim-

ulus must be synchronized with some other robot ac-

tion. The same subjects participated in this study as in

the sensitivity study reported above. As before, sub-

jects sat in a chair at different distances from the hap-

tic air device with the device positioned to blow air on

the back of their head/neck. Each subject completed

each condition twice and the mean of their responses

was used as their individual response. Subjects wore

their normal work clothes and made no particular ef-

fort to expose the back of their neck. In other words,

everyone’s neck was exposed. They were, however,

attentive to the stimulus and thus these results should

be regarded as the best-case results for detectability;

hence the need for the naturalistic user study in the

next section.

Figure 8 shows time to detection of the average

of on air and off air events for the No Funnel, Fun-

nel 1 and Funnel 2 designs for different distances. All

ROBOVIS 2020 - International Conference on Robotics, Computer Vision and Intelligent Systems

110

0.25 0.75 1.25 1.75

0.50

1.00

1.50

2.00

2.50

3.00

Reaction Time (Air On Event)

No Funnel

Funnel 1

Funnel 2

Distance (m)

Time (s)

0.25 0.75 1.25 1.75

0.50

1.00

1.50

2.00

2.50

3.00

Reaction Time (Air Off Event)

No Funnel

Funnel 1

Funnel 2

Distance (m)

Time (s)

(a) (b)

Figure 8: Time to detection of air on (a) and off (b) events for No Funnel, Funnel 1 and Funnel 2. Error bars show standard

errors.

of the curves are quite well ﬁt by straight lines. This

suggests that it should be relatively straightforward

to properly schedule haptic air on and haptic air off

events synchronized with other events on the robot.

As shown in Fig. 8, the design of Funnel 1 was able

to attract people’s attention more effectively than No

Funnel at distances within 1.0m, which is in the range

of personal distance zone described by Hall (1966),

in which someone might tap another on the shoulder.

The air off event showed a similar time course as the

air on event, as shown in Fig. 8. Perception of air

ﬂow from Funnel 2 was signiﬁcantly delayed com-

pared with the other two.

One complication with the use of haptic air as a

cue is the potential for protective clothing and/or long

hair to interfere with the delivery of the haptic cue. In

order to consider this we conducted a pilot study with

a single individual with long hair using the same pro-

cedure as before, but with the hair either being held up

or blocking the back of the participant’s neck. Fig. 9

shows the sensitivity for this one individual. As ex-

pected, having long hair that impedes air ﬂow reduces

the efﬁciency of the cue, but does not eliminate it. For

most of the conﬁgurations tested, the air cue remained

detectable out to 0.75m. Note that this is not the case

of Funnel 2 where the subject in the hair down condi-

tion was unable to perceive the haptic cue. This sug-

gests that a robot vehicle with an appropriate haptic

air generator could potentially infer the appropriate

range or air ﬂow for a speciﬁc individual based on

their neck coverage.

6 DISCUSSION

A critical challenge for any robot that is to be de-

ployed in a social setting is maneuvering in coordi-

nation with other users in the space. The problem of

No Funnel Funnel 1 Funnel 2

0.5

1.5

2.5

Max Sensitivity

Hair Up

Hair Down

Distance (m)

Figure 9: Haptic air sensitivity for one individual with her

long hair up and down. Even though there is a considerable

decrease in sensitivity, for most of the haptic air devices the

cue was perceived reliably out to 0.75m. Note that for the

hair down condition for Funnel 2 the subject was unable to

perceive the air cue.

moving one agent through a crowd of dynamic obsta-

cles is a classic one, and indeed research in this space

can be traced back to Kant and Zucker (1986) . As the

space becomes busier, however, it may become im-

possible to ﬁnd a path through the space and it then

becomes necessary to develop strategies to interact

with people in the space to create a path through the

space. This requires the robot gaining the attention of

people and communicating that the robot needs to get

through.

In some domains, the process of gaining the at-

tention of other people in the vicinity can be accom-

plished by intrusive methods such as loud noises and

bright lights. Indeed this is exactly the approach

used by emergency vehicles where efﬁcient passage is

vastly more important than politeness. In more social

settings, such approaches are unlikely to be consid-

ered acceptable, hence an autonomous system needs

to be able to get the attention of other people in the

The Answer Is Blowing in the Wind: Directed Air Flow for Socially-acceptable Human-Robot Interaction

111

space in a socially appropriate manner. We have ex-

plored the use of fan-generated air ﬂows to acquire

this attention and have shown that these ﬂows are ca-

pable of delivering a haptic air cue within a range of

2m. We found that the delay between the generation

of a haptic air event and a human’s perception of it

is a predictable function of distance. Therefore, hap-

tic air cuing could be easily integrated into a multi-

modal attention seeking approach. We also demon-

strated that the addition of an appropriately shaped

funnel can create a narrower air stream with a more

consistent width than a fan alone. A narrow air stream

allows the robot to be selective about which humans

to contact, allowing it to be less disruptive overall.

Funnels can also produce a higher air speed near the

fan, resulting in slightly shorter cuing delays at short

distances (<1m).

Although we explored the use of fans to generate

air ﬂow this is not the only potential technology that

can be applied to the problem. One promising alter-

native technology relies on the use of arrays of sub-

woofer speakers to drive air in a controlled manner.

Air cannons driven by such arrays can provide con-

trolled haptic air events. See Sodhi et al. (2013) and

Mowaﬁ et al. (2015) for examples of this approach. It

is also possible to use stored compressed air to gener-

ate easily controlled haptic air events. See Tsalamlal

et al. (2014) for an example. Another promising ap-

proach is the use of a haptic ultrasound (Long et al.,

2014).

In a related study Friedman (2020a, b) we inte-

grated the haptic air cue with an audio cue and visual

instructions in a robot interaction experiment. After

gaining the attention of the person, in that experiment,

the robot communicated to the person its requirement

that the the person move off of the robot’s path (i.e.,

to get out of the way). The use of this multimodal

cue integrated with explicit instructions was found to

be more effective than the robot just waiting for hu-

mans to move out of the way. In this earlier study

we used a simple fan to provide the haptic air cues.

Results presented here demonstrate that even more ef-

fective haptic air cues can be delivered through simple

augmentation of the air source. If robots are to oper-

ate well in social settings, they require mechanisms to

gain the attention of other people and to communicate

their requirements to them. Haptic air cues provides a

safe and socially acceptable mechanism for capturing

that attention.

REFERENCES

Burgoon, J. K., Johnson, M. L., and Koch, P. T. (1998). The

nature and measurement of interpersonal dominance.

Communications Monographs, 65(4):308–335.

Cattafesta, L., Bahr, C., and Mathew, J. (2010). Fundamen-

tals of Wind-Tunnel Design. Wiley, Chichester, UK.

Dakopolos, A. J. and Jahromi, L. B. (2019). Differences in

sensory responses among children with autism spec-

trum disorder and typical development: Links to joint

attention and social competence. Infant and Child De-

velopment, 28(1):e2117. e2117 ICD-17-0123.R1.

Friedman, N., Goedicke, D., Jenkin, M., Degutyte, Z.,

Zhang, V., Astell, A., and Dudek, G. (2020a). Ex-

ploring modalities to ask, “excuse me, could i get

through?”. In Proc. Workshop on Integrating Multi-

disciplinary Approaches to Advance Physical Human-

Robot Interaction, Paris, France. Held in conjunction

with IEEE ICRA.

Friedman, N., Goedicke, D., Zhang, V., Jenkin, M.,

Degutyte, Z., Astell, A., Liu, S., and Dudek, G.

(2020b). Out of my way! exploring different modali-

ties for robots to ask people to move out of the way. In

Workshop on Active Vision and Perception in Human(-

Robot) Collaboration, Held Online. Held in conjunc-

tion with IEEE Int. Conf. on Robot and Human Inter-

active Communication.

Gupta, S., Morris, D., Patel, S., and Tan, D. (2013). Air-

Wave: Non-contact haptic feedback using air vor-

tex rings. In ACM Joint Conference on Pervasive

and Ubiquitous Computing, pages 419–428, Zurich,

Switzerland.

Hall, E. T. (1966). The Hidden Dimension, volume 609.

Doubleday, Garden City, NY.

Hasegawa, K., Qiu, L., Noda, A., Inoue, S., and Shinoda, H.

(2017). Electrically steerable ultrasound-driven long

narrow air ﬂow. Applied Physics Letters, 111:06104.

Hoshi, T., Iwamoto, T., and Shinoda, H. (2009). Non-

contact tactile sensation synthesized by ultrasound

transducers. In 3rd Joint Eurohaptics Conference and

Symposium on Haptic Interfaces, pages 256–260, Salt

Lake City, UT.

Kant, K. and Zucker, S. (1986). Toward efﬁcient trajectory

planning: The path-velocity decomposition. Interna-

tional Journal of Robotic Research - IJRR, 5:72–89.

Long, B., Seah, S. A., Carter, T., and Subramanian, S.

(2014). Rendering volumetric haptic shapes in mid-

air using ultrasound. In ACM SIGGRAPH Asia, Shen-

zhen, China.

Morel, T. (1975). Comprehensive design of axisymmetric

wind tunnel contractions. Journal of Fluids Engineer-

ing, Trans. ASME, 97(2):255.

Mowaﬁ, O., Khamiodhi, M., and Abouelsaadat, W. (2015).

Airdisplay: experimenting with air ﬂow as a com-

munication medium. In INTERACT, pages 316–323,

Bamberg, Germany.

Sodhi, R., Poupyrev, I., Glisson, M., and Israr, A. (2013).

AIREAL: Interactive tactile experiences in free air. In

ACM SIGGRAPH, Anaheim, CA.

ROBOVIS 2020 - International Conference on Robotics, Computer Vision and Intelligent Systems

112

Suzuki, Y. and Kobayashi, M. (2005). Air jet driven force

feedback in virtual reality. IEEE Computer Graphics

and Applications, Jan/Feb:44–47.

Tolley, D., Nguyen, T., Tang, A., Ranasinghe, N.,

Kawauchi, K., and Yen, C.-C. (2019). Windywall:

Exploring create wind simulation. In TEI 2019: Inter-

national Conference on Tangible and Embedded In-

teraction, pages 635–644, Tempe, AZ.

Tsalamlal, M. Y., Issart, P., Ouarti, N., and Ammi, M.

(2014). Hair: Haptic feedback with a mobile air jet. In

2014 IEEE International Conference on Robotics and

Automation (ICRA), pages 2699–2706, Hong Kong,

China. IEEE.

Verlinden, J., Mulder, F. A., Vergeest, J. S. M., de Jonge, B.,

Krutly, D., Nagy, Z. E., Logeman, B. J., and Schouten,

P. (2013). Enhancement of presence in a virtual sailing

environment through localized wind simulation. Pro-

cedia Engineering, 60:435–441.

Wogollter, M. S. (2006). Attention switch and maintenance.

In Mahwah, N., editor, Handbook of Warnings, pages

245–265. Lawrence Erlbaum Associates.

The Answer Is Blowing in the Wind: Directed Air Flow for Socially-acceptable Human-Robot Interaction

113