Smartphone Teleoperation for Self-balancing Telepresence Robots

Antti E. Ainasoja

, Said Pertuz

and Joni-Kristian K¨am¨ar¨ainen

Laboratory of Signal Processing, Tampere University of Technology, Finland

Keywords:

Teleoperation, Telepresence, User Interface.

Abstract:

Self-balancing mobile platforms have recently been adopted in many applications thanks to their light-weight

and slim build. However, inherent instability in their behaviour makes both manual and autonomous operation

more challenging as compared to traditional self-standing platforms. In this work, we experimentally evaluate

three teleoperation user interface approaches to remotely control a self-balancing telepresence platform: 1)

touchscreen button user interface, 2) tilt user interface and 3) hybrid touchscreen-tilt user interface. We provide

evaluation in quantitative terms based on user trajectories and recorded control data, and qualitative ﬁndings

from user surveys. Both quantitative and qualitative results support our ﬁnding that the hybrid user interface

(a speed slider with tilt turn) is a suitable approach for smartphone-based teleoperation of self-balancing

telepresence robots. We also introduce a client-server based multi-user telepresence architecture using open

source tools.

1 INTRODUCTION

Telepresence can provide substantial aid in certain

applications of welfare services, such as remotely

visiting homes of self-living elderly (Apostolopoulos

et al., 2012; Leeb et al., 2015). In this scope, it is de-

sirable that these telepresence-based welfare services

allow for a teleoperation via mobile devices since

their users, e.g. nurses and health care providers,

should have access to quick and effective interaction.

Due to the ﬂexibility and ease of access of mobile de-

vices, the combination of novel user interfaces and

smartphone-based teleoperation arise as a potential

alternative for these applications (Fong et al., 2001).

In the literature, several researchers have proposed

different approaches for the teleoperation of telepres-

ence platforms, such as haptic interfaces, or touch

screens (Ogata et al., 2015), and tilt user interface

(Findlater et al., 2013; Baldauf et al., 2015). However,

conclusive results on identifying the advantages and

disadvantages of each approach in the context of self-

balancing platforms are lacking and, therefore, fur-

ther research is warranted. In this work, we ﬁrst de-

velop an architecture for the remote control of a self-

balancing “Segway-type” telepresence robot. Subse-

quently, we compare three different smartphone tele-

operation strategies: touchscreen button user inter-

face, tilt user interface, and hybrid touchscreen-tilt

https://orcid.org/0000-0003-2016-3870

https://orcid.org/0000-0001-8498-9917

https://orcid.org/0000-0002-5801-4371

user interface.

Telepresence platforms have shown their poten-

tial for human assistance in healthcare environments

(Boissy et al., 2007; Tiberio et al., 2012). In turn, self-

balancing robotic platforms are gaining momentum

in the community due to their slim and light-weight

build (Tsai et al., 2010). However, these platforms

pose certain challenges in term of stability and con-

trol. The teleoperation through wireless networks in-

troduce delays in robot response to commands. Fur-

thermore, the self-balancing mechanism prevents the

robot for speeding up, slowing down or stopping in-

stantaneously. Teleoperation user interfaces must be

comfortable to the operators and feel responsive even

with delayed execution of commands. This work pro-

vides an experimental baseline for the assessment of

the effectivenessof teleoperation user interface strate-

gies for self-balancing telepresence platforms.

The main contributions of this work can be sum-

marized as follows:

• We develop a client-server architecture for teleop-

eration using open source tools.

• We develop three intuitive but distinctly differ-

ent user interfaces for operating a self-balancing

telepresence robot based on three different strate-

gies: 1) touchscreen button user interface, 2) tilt

user interface and, 3) hybrid touchscreen-tilt user

interface.

• We propose a navigation task and evaluation crite-

ria to compare the implemented teleoperation user

interface strategies.

Ainasoja, A., Pertuz, S. and Kämäräinen, J.

Smartphone Teleoperation for Self-balancing Telepresence Robots.

DOI: 10.5220/0007406405610568

In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2019), pages 561-568

ISBN: 978-989-758-354-4

561

Experimental results from both quantitative and

qualitative evaluation with random users indicate that

the hybrid interface provides best performance in

terms of steadiness of driving and user preference.

Our source code and examples will be made publicly

available to facilitate further development of new fea-

tures and applications of telepresence for supporting

self-living elderly.

2 RELATED WORK

The use of smartphone-based user interface for telep-

resence robots can be traced back to (Fong et al.,

2001). In that work, they focused on the motivation

and design criteria for different user interface strate-

gies. More recently, improved user interface by robot

autonomy was proposed by Vaughan et al. (Vaughan

et al., 2016), who adopted the “reaching through the

screen” concept (Foote et al., 2004). In their system,

the telepresence bot is equipped with a ﬁsheye lens

that provides broader view where a remote user clicks

the path and indicates the goal. ROS-based visual

navigation controls the movement towards the goal

location using the path points. Bohren et al. (Bohren

et al., 2016) take autonomy to a higher level by adopt-

ing a prediction model where robot operation is based

on predicted user controls. A formal deﬁnition of

prediction and user assistance in teleoperation was

given by Dragan and Srinivasa (Dragan and Srinivasa,

2012). This strategy is particularly important in the

presence of long network delays and a similar simu-

lation model has been proposed for multi-player game

client-server systems (Bernier, 2001). The aforemen-

tioned works have focused on the development of

novel user interface strategies without a direct com-

parison of different approaches.

In the literature, a lot of efforts have been devoted

to the improvement of the user experience in the op-

eration of telepresence platforms either by incorporat-

ing new features or by providing autonomous capabil-

ities. Kisevel et al. (Kiselev et al., 2015) studied the

usage of semi-autonomous features by novice telep-

resence users. Tee et al. (Tee et al., 2014) showed

improved telepresence experience in meetings by us-

ing audio-visual gesture and attention recognition.

Schwarz et al. (Schwarz et al., 2014) provide ad-

justable autonomy for teleoperation where manual

operation is requested when autonomy fails. Cosgun

et al. (Cosgun et al., 2013) incorporate autonomous

person following as a feature for telepresence plat-

form.

Most efforts in the comparison of user interface

strategies have been performed in the context of gam-

ing or virtual reality. In this scope, the results by Find-

later et al. (Findlater et al., 2013) indicate that modern

touchscreen interfaces are easier to use for non-expert

users than traditional “gaming setups” with mouse

and keyboard for simple pointing and clicking and

tracking task. Baldauf et al. (Baldauf et al., 2015)

did an analysis on different mobile phone game con-

troller interfaces and found that, while navigation but-

tons and virtual joystick are easier to use and more

accurate than tilt interfaces, they also needed more at-

tention from users in order to touch the controls pre-

cisely. These ﬁndings were based on 2D games on

large screen where the view point was third person or

from the bird eye perspective and suggested that tilt

user interface may be more suitable for certain ﬁrst

person perspective games, such as ﬂight simulators.

Most strategies considered in the state-of-the-art

are mainly based on touchscreen user interfaces. In

this work, we decided to include the tilt-based user

interfaces, as suggested by Baldauf et al. (Baldauf

et al., 2015) for ﬁrst-person perspective games due to

their similarities with telepresence. Remarkably, self-

balancing telepresence is distinctly more challeng-

ing to operate than the traditional standing (three/four

wheel) platforms (Shimada and Hatakeyama, 2008).

3 SELF-BALANCING

TELEPRESENCE SYSTEM

The motivation for this work is the development of

a social robot platform with adjustable autonomy to

support self-living elderly. This particular application

poses speciﬁc requirements from both the user and

system perspective. Speciﬁcally, the platform must be

based on affordable commodity hardware and open

source software to allow agile development of new

features and services by the open source community.

In addition, the platform should guarantee that ad-

justable autonomy is available.

Adjustable autonomy in our case means that con-

trol can always be given to human users either when

requested by active peers or when the platform cannot

perform requested tasks autonomously. The lowest

level of autonomy is teleoperation and, for the needs

of different peers (nurses, relatives, elderly them-

selves and possible third parties), the teleoperation in-

terface must be easy-to-use, affordable and intuitive.

The teleoperation system developed for this purpose

in this work is illustrated in Fig. 1. The system is

comprised of four main components:

• Android client

• Communication server

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

562

Figure 1: The most important parts of the telepresence sys-

tem. The system includes a self-balancing Double telepres-

ence robot controlled by a system that comprises an iPad,

an Android tablet device and a Communication server.

• Telepresence robot

• iPad control platform

Each one of the components of the system are de-

tailed below.

Android Client. All the interactions between the

user and the system are managed through a tablet run-

ning the Android client. In order to allow for a ﬂexible

operation of the telepresence robot, an Nvidia Shield

K1 Android tablet was used. The table is equipped

with a capacitive multi-touch screen, accelerometer

and gyroscope. An application with several graphical

user interfaces (GUI’s) was developed in order to im-

plement the operation strategies under study (section

4).

The Android client is connected to Internet

through an LTE network. In order to have access

to the video and audio streams, as well as for send-

ing commands to the robot, the Android client is

connected to the Communication server (see below)

through an HTTP application running on a Linux

host.

Communication Server. The job of the communi-

cation server (CS) is establishing the connection be-

tween the Android client (user) and the iPad used to

control the Double robot upon request by the user.

Once the connection is established, the CS manages

the communication between the user and the iPad.

Depending on the type of signal, the communi-

cation goes through two different paths. On the one

side, status signals reported by robot, as well as com-

mands sent by the user from the Android client, go

through the CS. The CS logs and sends the teleoper-

ation commands received from the user. On the other

https://www.doublerobotics.com/

Figure 2: A sequence diagram of iOS and Android clients

making connection through signaling server (CS).

side, video and audio signals from the iPad are trans-

ferred to the Android server using the OpenWebRTC

implementation of WebRTC

standard. The WebRTC

protocol was chosen due to its ﬂexibility, modular-

ity and cross-platform interoperability. A sequence

diagram illustrating connection initialization between

clients and server is presented in Figure 2.

iPad Control Platform. An iPad tablet is used for

direct control of the Double robot and for capturing

the video and audio signal on the robot end. In or-

der to facilitate navigation, the camera was equipped

with a Black Eye full-frame ﬁsheye lens with 180-

degree ﬁeld of view

. This was found to be a good

compromise between viewing the surroundings of the

robot and being able to see the objects in front of the

robot in detail. As stated previously, the video and

audio signals are sent to the user using WebRTC. For

motion control of the telepresence platform, a custom

application was developed based on the Double SDK

for the iPad to delegate the commands to the robot.

4 SMARTPHONE USER

INTERFACES

For comparison purposes, we implemented three user

interfaces for the self-balancing two-wheel telepres-

ence navigation system described in previous section:

1) touchscreen user interface (found best in (Baldauf

et al., 2015) for third person interfaces in 2D games),

2) Tilt user interface (simplest interface requiring the

https://www.openwebrtc.org

https://www.w3.org/TR/webrtc/

https://blackeyelens.com/product/full-frame-ﬁsheye/

Smartphone Teleoperation for Self-balancing Telepresence Robots

563

least amount of user attention), and 3) Hybrid (tilt

for turning and a slider for forward/backward). In all

of our interfaces it is assumed that a user holds the

tablet horizontally with both hands. The interfaces

have a view of the remote video stream at the center

and the control interfaces are on left and right sides of

the stream view. Examples of the user interfaces are

shown in Fig. 3. A more detailed explanation of the

features of each implemented user interface strategy

can be found below.

4.1 Touchscreen Button UI (B-GUI)

The navigation buttons interface consists of four but-

tons marked with arrows on each side of the remote

video view. By pressing the buttons the robot moves

forward or backward and turns left or right. Both

the left and right hand side buttons worked identically

and the intention was that they could be used together

to accelerate/decelerate with one hand and steer with

other, or use one hand for both turning and speeding.

The buttons only have binary states so when the but-

ton is pressed the robot moves according to the direc-

tion at full speed. When no buttons are pressed, the

robot stops. Due to the self-balancing nature of the

robot, the acceleration is never immediate. Using lin-

ear or exponential acceleration over a short period of

time after the button is pressed made the interface feel

more sluggish and delayed.

4.2 Tilt Interface (T-GUI)

The tilt user interface uses the pitch and roll of the

tablet relative to real world coordinates to operate the

robot. The pitch is used to control the speed of the

robot; tilting away (forward) makes the robot move

forward and tilting back backwards. Likewise, the roll

turns the robot left and right, respectively.

During the early experiments we noticed that

some mechanism was needed to tell when the user

interface was used – active – and when not. For this

reason we placed dead man’s switch-type buttons on

the interface. In the initial setting, pressing only one

of the buttons made the system active, but this was not

sufﬁcient since users, when distracted, kept holding

the tablet touching a switch with one hand and subse-

quently lost the control of the robot. In the ﬁnal in-

terface, both dead man’s switches need to be pressed

simultaneously.

The tilt interface measured the angular displace-

ment of the tablet from the neutral position. Displace-

ment angles were mapped to the velocity (pitch) and

turning speed (roll) of the robot. In order to make it

easier to hold the robot still, displacement angles pro-

portionalto the full calibration scales and less than the

threshold τ

= 10 % from the neutral position were

mapped to zero. It was noticed that, in order to have

better control of the robot at lower velocities, linear

mapping was not sufﬁcient. We squared the angles

to have more ﬁne control at the lower velocities. For

turning, linear mapping seemed more comfortable.

4.3 Hybrid Interface (BT-GUI)

The third implemented user interface was a hybrid of

the two above. Steering of the robot is the same as

in T-GUI and the velocity was controlled similar to

B-GUI, but the button was replaced with a “spring

slider” (right hand side of the GUI). A user drags the

slider up and down to move the robot forward and

backwards. Due to the spring effect the switch returns

immediately to the neutral position when released.

The position of the switch was linearly mapped to the

robot velocity. One dead man’s switch was placed

on the left hand side to allow turning the robot with-

out moving it backwards or forwards. Turning works

only when either the dead man’s switch or the slider

switch are touched.

5 ASSESSMENT

METHODOLOGY

In order to assess the performance of each operating

strategy, users were asked to perform a simple task us-

ing one or more approach in random order and differ-

ent performance indicators were measured. The de-

tails about the performance indicators are presented

in Section 5.1 and Section 5.2. A summary of the per-

formance indicators can be found in Fig. 4 The de-

scription of the task and testing environment assigned

to users are presented in Section 5.3.

5.1 Quantitative Performance

Indicators

Quantitative performance indicators were divided in

two groups: indicators aimed at measuring the effec-

tiveness of the assigned task, and indicators aimed at

measuring the driving style and comfort of the user.

Driving Effectiveness. The driving effectiveness is

aimed at assessing how effective the teleoperation

strategies were in the completion of the assigned task,

regardless of the driving style or perceived comfort.

Indicators aimed at measuring the effectiveness were

three:

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

564

Figure 3: Developed interfaces for the implemented teleoperation strategies. (a) touchscreen interface with navigation buttons,

(b) tilt interface and, (b) Hybrid touchscreen-tilt interface. The two tilt-based interfaces (b)-(c) include a dead man’s switch

(“Drive” button in the GUI’s) that turns the tilt interface on/off.

Figure 4: Performance indicators. Quantitative indicators

were measured by analyzing the logged signals from the

telepresence platform. Qualitative indicators were assessed

by means of user surveys.

• Completion rate

• Number of collisions

• Completion time

From these indicators, the most important one is

the completion rate; whether or not the user could

maneuver the robot safely, or at all. The amount of

collisions was measured by observing the amount of

crashes during each test drive. The completion time

indicates how fast the operators are able to complete

the task.

Driving Comfort. The indicators for driving com-

fort are aimed at assessing how smooth and steady is

the operation of the platform. For this purpose, we

deﬁned three performance indicators:

• Average angular speed

• Smoothness of the trajectory

• Number of full stops

Both the average angular speed and the smooth-

ness of the trajectory are aimed at measuring how sta-

ble is the trajectory of the platform. Ideally, the robot

should move following a smooth trajectory without

sudden changes in direction or speed. For this pur-

pose, given a parametric trajectory {x(t), y(t)} as a

function of time t, the instantaneous steering angle is

given by:

θ(t) ≈ tan

−1



x(t + ∆t) − x(t)

y(t + ∆t) − y(t)



, (1)

where ∆t is the time step used when recording the data

of the telepresence platform.

The average angular speed, ω

mean

, is then mea-

sured as:

mean

∑

n=1

dθ(t)

dt, (2)

where N is the number of time stamps.

The smoothness of the trajectory is measured by

estimating the Laplacian in the x and y directions, as:

Smoothness = |∇

x(t)| + |∇

y(t)| (3)

Finally the number of full-stops was measured in

order to record how conﬁdently the users could per-

form the task. It was assumed that stopping was done

when the operators felt they were about to lose con-

trol of the robot. A full stop was assumed when the

operator sent no commands to the robot and robot was

not moving. Speciﬁcally, the amount of robot motion

was computed as:

Robot motion = |

max

| + |

max

|, (4)

where v and ω are the instantaneous velocity and an-

gular velocity, v

max

is the maximum velocity (0.5 m/s)

and ω

max

is the maximum angular velocity (0.586

rad/s) of the robot.

Due to the self-balancing mechanism, the robot is

constantly swaying back and forth, even when sta-

tionary. This was accounted by setting a threshold

level on the motion. The robot was considered fully

stopped if the amount of motion was below the thresh-

old of 0.05. The robot motion and the mechanism for

the detection of full stops is illustrated in Figure 5.

5.2 Qualitative Performance Indicators

An important performance indicator is how the opera-

tors subjectively felt about each operation strategy. A

short survey was created for this purpose in order to

assess three different aspects:

Smartphone Teleoperation for Self-balancing Telepresence Robots

565

Figure 5: An example of sent commands in relation to the robot motion over time. The robot is considered to be in full stop

when the yellow line is below the threshold level.

• Ease of use

• Perceived speed

• User preference

On the survey, operators were asked to give feed-

back on how they felt about the ease of use of each

user interface approach. Each approach was assigned

a qualiﬁcation of 0, 1, 2 or 3, corresponding to four

levels of difﬁculty: very easy, easy, difﬁcult and very

difﬁcult. Users were also surveyed on how fast they

perceived that they could perform the given task. This

was to assess the consistency between objective quan-

titative performance indicators, such as the comple-

tion time, and the perceived user experience. Finally,

operators who tested more than one user interface ap-

proach were asked which one they preferred. If a n

user interface feels too difﬁcult and stressful, it would

not be used by the operators even if it was safe and

fast.

5.3 Navigation Task and Environment

In order to collect quantitative and qualitative indi-

cators, users were to maneuver the robot around an

approximately 50 m long route in well-lit, populated

lobby area (Fig. 6). Due to people having different

ways of holding a tablet, the tilt and hybrid interfaces

had initially a calibration tool for conﬁguring the neu-

tral (no movement) and extreme positions (full for-

ward, full backward and extreme left and right) for

each user. In preliminary tests, this was found to be

too time consuming and complicated. Therefore the

experiments reported in this work were done with a

common pre-calibration. For controlling speed, the

neutral position was set to 30

◦

with 0

◦

and 60

◦

extreme

positions. For turning, the neutral position was set to

◦

with -45

◦

and 45

◦

extreme positions. During initial

experiments users reported the fast acceleration and

speed to be difﬁcult when using the touchscreen but-

ton interface. For the touchscreen button interface,

velocity was therefore limited to 75% of the robot’s

maximum velocity.

Figure 6: Map of the experiment area. Yellow denotes

working areas with non-ﬁxed furniture. Green dot is the

start/ﬁnish location and green line is the ideal route of the

task.

6 RESULTS AND DISCUSSION

We recruited 21 participants (18 males, 3 females)

with ages ranging from 21 to 47 (mean 27.1). 13

participants completed the navigation task described

in section 5.3 using all the three user interfaces. Re-

ported quantitative indicators were collected from the

logged navigation signals and robots internal mea-

surement unit (received by the platform), whereas

qualitativeresults were obtained from user surveys af-

ter completing the task.

6.1 Quantitative Indicators

The average indicators for the different interfaces are

shown in Table 1 (Driving effectiveness) and Table 2

(Driving comfort). Tests that failed to be completed

due to network failures (1 for T-GUI, 5 for BT-GUI)

were omitted from quantitative data. Cases where

users fatally crashed the robot (task could not be com-

pleted) are included in the completion rate, but not

omitted from other qualitative data.

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

566

Table 1: Quantitative indicators for driving effectiveness.

UI Completion rate Avg. collisions Avg.time (std) N

B-GUI 0.94 0.13 220 (42) s 15

T-GUI 1.0 0.23 198 (55) s 13

BT-GUI 0.91 0.33 211 (71) s 10

Table 2: Quantitative indicators for driving comfort.

UI stops (std) angular speed (std) Smoothness (std) N

B-GUI 8.2 (6.8) 0.1641 (0.1035) 0.7668 (1.1819) 15

T-GUI 7.7 (6.5) 0.1385 (0.0180) 0.4907 (0.1105) 13

BT-GUI 7.8 (5.1) 0.1337 (0.0158) 0.3596 (0.0876) 10

Table 3: The mean ease of use and perceived speeds re-

ported for each user interface on a scale from very easy/very

slow = 0 to very difﬁcult/very fast = 3.

UI Ease of use Perceived speed

B-GUI 0.93 1.64

T-GUI 1.00 1.16

BT-GUI 0.96 1.21

Table 4: Teleoperation interface preferences of people who

tested all three interfaces.

B-GUI T-GUI BT-GUI

15% 35% 50%

The tilt interface (T-GUI) is the most effective in

the average task completion time. In driving com-

fort, however, the hybrid interface is clearly superior

in smoothness and also the best in average angular

speed (the lower values in table 2 indicate a better per-

formance). The number of full stops are similar for all

methods. The quantitative indicators do not provide a

clear winner, but the tilt and hybrid interfaces seem to

perform better than the plain touchscreen interface.

6.2 Qualitative Indicators

Results based on subjective feedback are summarized

in Tables 3 and 4. Users found controlling the robot

on all of the user interfaces to be easy. The two inter-

faces using tilt (T-GUI and BT-GUI) had a perceived

speed faster than the touchscreen-only interface (Ta-

ble 3). The user preference in Table 4 support this

observation.

Slide interface was preferred by users with tilt be-

ing the second favorite. Some users found it difﬁcult

to use tilting for both turning and acceleration. This

was partly due to the interface’s sensitivity to cali-

bration and the position in which the table was held.

Some users found it difﬁcult to hold the tablet in a po-

sition they were not accustomed to. Those preferring

the buttons reported that it was easier to perceive how

the robot moves using them.

Haptic feedback and better indication on turning

speed was suggested for the tilt and slide interfaces.

Not directly related to the interface schemes, the user

generally foundestimating the robot dimensions in re-

lation to the surroundings difﬁcult. This was due to

the robot not being visible to the operator. An aug-

mentation to cast a shadow on the robot’s projected

path could make it easier to avoid obstacles and go

through tight spots.

6.3 Other Findings

The communication with the telepresence platform

relied on good wireless networks. The delay between

giving a command and robot reaction was usually

around 300 ms, but occasionally it could go up to sev-

eral seconds or connection was lost altogether.

7 CONCLUSIONS

In this paper we developed and implemented an archi-

tecture for the teleoperation of self-balancing telep-

resence platform using open source tools. The goal of

our architecture is to foster open community develop-

ment of social robotics applications for independent

living elderly.

The ﬁrst obviousrequirement is user-friendly tele-

operation user interface for the cases an operator is

needed. To study teleoperation of a self-balancing bot

we implemented three different user interface strate-

gies: touchscreen, tilt and hybrid (a combination of

tilt and touchscreen). Quantitative and qualitative

measurements on a real navigation task suggest that

the hybrid interface yields the best performance in

terms of both driving comfort and user preference, tilt

Smartphone Teleoperation for Self-balancing Telepresence Robots

567

being the second best with small margin and touch-

screen buttons being the worst with clear margin.

ACKNOWLEDGEMENTS

This work was funded by Academy of Finland

project: “Robots and the Future of Welfare Services”,

decision number: 314180.

REFERENCES

Apostolopoulos, J. G., Chou, P. A., Culbertson, B., Kalker,

T., Trott, M. D., and Wee, S. (2012). The road to

immersive communication. Proceedings of the IEEE,

100(4):974–990.

Baldauf, M., Fr¨ohlich, P., Adegeye, F., and Suette, S.

(2015). Investigating on-screen gamepad designs for

smartphone-controlled video games. ACM Trans.

Multimedia Comput. Commun. Appl., 12(1s):22:1–

22:21.

Bernier, Y. (2001). Latency compensating methods in

client/server in-game protocol design and optimiza-

tion. In Game Developers Conference.

Bohren, J., Paxton, C., Howarth, R., Hager, G., and Whit-

comb, L. (2016). Semi-autonomous telerobotic as-

sembly over high-latency networks. In ICRA.

Boissy, P., Corriveau, H., Michaud, F., Labont, D., and

Royer, M.-P. (2007). A qualitative study of in-home

robotic telepresence for home care of community-

living elderly subjects. Journal of Telemedicine and

Telecare, 13(2):79–84. PMID: 17359571.

Cosgun, A., Florencio, D., and Christensen, H. (2013). Au-

tonomous person following for telepresence robots. In

ICRA.

Dragan, A. and Srinivasa, S. (2012). Formalizing assistive

teleoperation. In RSS.

Findlater, L., Froehlich, J. E., Fattal, K., Wobbrock, J. O.,

and Dastyar, T. (2013). Age-related differences in per-

formance with touchscreens compared to traditional

mouse input. In Proceedings of the SIGCHI Confer-

ence on Human Factors in Computing Systems, CHI

’13, pages 343–346, New York, NY, USA. ACM.

Fong, T., Conti, F., Grange, S., and Baur, C. (2001). Novel

interfaces for remote driving: gesture, haptic, and pda.

In Proc. SPIE 4195, Mobile Robots XV and Telema-

nipulator and Telepresence Technologies VII.

Foote, J., Liu, Q., Kimber, D., Chiu, P., and Zhao, F. (2004).

Reach-through-the-screen: A new metaphor for re-

mote collaboration. In Advances in Multimedia In-

formation Processing (PCM).

Kiselev, A., Kristoffersson, A., Melendez, F., Galindo, C.,

Loutﬁ, A., Gonzalez-Jimenez, J., and Coradeschi, S.

(2015). Evaluation of using semi-autonomy features

in mobile robotic telepresence systems. In Robotics,

Automation and Mechatronics (RAM).

Leeb, R., Tonin, L., Rohm, M., Desideri, L., Carlson, T.,

and d. R. Milln, J. (2015). Towards independence:

A bci telepresence robot for people with severe mo-

tor disabilities. Proceedings of the IEEE, 103(6):969–

982.

Ogata, M., Teramura, R., and Imai, M. (2015). Attractive

telepresence communication with movable and touch-

able display robot. In 2015 24th IEEE International

Symposium on Robot and Human Interactive Commu-

nication (RO-MAN), pages 179–184.

Schwarz, M., Stuckler, J., and Behnke, S. (2014). Mobile

teleoperation interfaces with adjustable autonomy for

personal service robots. In ACM/IEEE International

Conference on Human-Robot Interaction.

Shimada, A. and Hatakeyama, N. (2008). Movement con-

trol of two-wheeled inverted pendulum robots consid-

ering robustness. In 2008 SICE Annual Conference,

pages 3361–3366.

Tee, K., Yan, R., Chua, Y., Huang, Z., and Liemhetcharat, S.

(2014). Gesture-based attention direction for a telep-

resence robot: Design and experimental study. In

ROS.

Tiberio, L., Cesta, A., Cortellessa, G., Padua, L., and Pel-

legrino, A. R. (2012). Assessing affective response of

older users to a telepresence robot using a combina-

tion of psychophysiological measures. In 2012 IEEE

RO-MAN: The 21st IEEE International Symposium on

Robot and Human Interactive Communication, pages

833–838.

Tsai, C. C., Huang, H. C., and Lin, S. C. (2010). Adap-

tive neural network control of a self-balancing two-

wheeled scooter. IEEE Transactions on Industrial

Electronics, 57(4):1420–1428.

Vaughan, J., Kratz, S., and Kimber, D. (2016). Look where

youre going: Visual interfaces for robot teleoperation.

In Int. Symp. on Robot and Human Interactive Com-

munication (RO-MAN).

VISAPP 2019 - 14th International Conference on Computer Vision Theory and Applications

568