ArticlePDF Available

Abstract and Figures

This article considers the suitability of current robots designed to assist humans in accomplishing their daily domestic tasks. With several million units sold worldwide, robotic vacuum cleaners are currently the figurehead in this field. As such, we will use them to investigate the following key question: How does a service cleaning robot perform in a real household? One must consider not just how well a robot accomplishes its task, but also how well it integrates inside the user’s space and perception. We took a holistic approach to addressing these topics by combining two studies in order to build a common ground. In the first of these studies, we analyzed a sample of seven robots to identify the influence of key technologies, such as the navigation system, on technical performance. In the second study, we conducted an ethnographic study within nine households to identify users’ needs. This innovative approach enables us to recommend a number of concrete improvements aimed at fulfilling users’ needs by leveraging current technologies to reach new possibilities.
Content may be subject to copyright.
Lessons Learned from Robotic Vacuum Cleaners Entering in the Home Ecosystem
F. Vaussarda,
, J. Finkb, V. Bauwensb, P. R´
etornaza, D. Hamela, P. Dillenbourgb, F. Mondadaa
aEcole Polytechnique F´ed´erale de Lausanne (EPFL), Robotic Systems Laboratory (LSRO), Station 9, 1015 Lausanne, Switzerland
bEcole Polytechnique F´ed´erale de Lausanne (EPFL), Pedagogical Research and Support (CRAFT), Station 20, 1015 Lausanne, Switzerland
This article considers the suitability of current robots designed to assist humans in accomplishing their daily domestic tasks. With
several million units sold worldwide, robotic vacuum cleaners are currently the figurehead in this field. As such, we will use them
to investigate the following key question: Could a robot possibly replace the hand-operated vacuum cleaner? One must consider not
just how well a robot accomplishes its task, but also how well it integrates inside the user’s space and perception. We took a holistic
approach to addressing these topics by combining two studies in order to build a common ground. In the first of these studies, we
analyzed a sample of seven robots to identify the influence of key technologies, like the navigation system, on performance. In
the second study, we conducted an ethnographic study within nine households to identify users’ needs. This innovative approach
enables us to recommend a number of concrete improvements aimed at fulfilling users’ needs by leveraging current technologies to
reach new possibilities.
Domestic robotics, Cleaning robots, Energy eciency, Human factors in robotics, Human-robot interaction
1. Introduction
Since the early 1950s, futuristic scenarios of our daily lives
at home have included robots: robot maids, robot companions,
robot nannies, robot guards [2]. This vision has not substan-
tially changed and it was only a couple of years ago that Bill
Gates predicted in Scientific American that there will soon be
“a robot in every home” [3]. Where are we now, however? So
far, the only success of domestic robots can be noted in the
field of floor-cleaning robots; millions of these devices are used
to vacuum people’s homes today. We do not know much yet
about the long-term acceptance of domestic robots, but first ex-
ploratory studies carried out with robotic vacuum cleaners in
the United States [4–6] suggest that these devices have several
shortcomings that may restrict a broad user acceptance beyond
initial adoption. Also, there are strong novelty eects with in-
novative technologies such as robots [7, 8].
Today’s robotic technologies are mainly driven by the tech-
nical challenges arising when a mobile robot has to perform a
specific task in a loosely defined environment. However, some
other topics have long been neglected in the design of robots
for specific purposes, including the energy-use implications of
some technical choices or harmonious integration of the robot
into the user’s ecosystem [9]. With technological progress,
robotic vacuum cleaners (along with other domestic appliances)
are now becoming widely available. The time has come to take
A first version of this work was originally published as conference pro-
ceedings [1].
Corresponding author. Tel.: +41 21 693 78 39; fax: +41 21 693 78 07.
Email address: (F. Vaussard)
URL: (F. Vaussard)
these topics into account when considering the design of future
robots, as it appears essential to integrate the user into the de-
sign loop to advance these products further.
Our approach to these issues tries to be holistic and seeks
synergies between current technical research and design that is
acceptable for the user. For this, we integrate results from a
first technical study on several robotic vacuum cleaners with
findings from a second study conducted in people’s homes. In
doing so, we aim to advance personal robotics from both tech-
nical and user-oriented points of view. This collaborative ap-
proach brings together research from various fields.
In robotics today (in spite of a cross-disciplinary approach),
the main body of the current research addresses either technical
issues (perception, locomotion, or learning algorithms, just to
name a few) or social phenomena independently of each other.
The eectiveness of the research being performed across disci-
plines is muted by this process. The technical and user points
of view are seldom presented side by side. Our two-sided study
fills this gap.
Robotics is, by its nature, multi-disciplinary. With our pro-
posed approach, we aim to extend the borders of the robotic
community by showing how synergies can create meaningful
cross-disciplinary dialogue. Ultimately, the common goal is to
develop human-oriented domestic robots that enable meaning-
ful human-robot interaction (HRI) and have the potential to im-
prove people’s quality of life.
The remainder of the paper is organized as follows: in Sec.
2, we will state the main questions guiding both the technical
and user studies. Sec. 3 will list related work to determine the
state of the art in both fields, while Sec. 4 summarizes our dual
methodology. The results build the core of Sec. 5, and will be
Preprint submitted to Robotics and Autonomous Systems March 7, 2013
presented using a unified outline, raising the knowledge gained
up to a higher level. The conclusion of Sec. 6 will present
the analysis of current robotic vacuum cleaners in light of both
studies’ findings, and will summarize current shortcomings. In
this section, we also suggest research directions for leveraging
current technologies to enhance user acceptance with targeted
2. Motivation
Robotic vacuum cleaners have attained a fair degree of suc-
cess in the domestic robot market. The iRobot Corporation (one
of the main players in this market) claims to have sold 6 million
units of its “Roomba“ robot between 2002 (its first release) and
2010 [10]. According to the statistics of the International Fed-
eration of Robotics [11], about 2.5 million personal and service
robots were sold in 2011, an increase of 15 % in numbers (19%
in value) compared to 2010. The forecast for the period 2012–
2015 exceeds 10 million units. This trend clearly emphasizes
the growing impact of domestic robots in our homes, which cre-
ates new interaction paradigms. In parallel, the energy demand
for the operation of millions of new cleaning robots will follow
the same tendency.
Moreover, with the evolution of technologies, domestic
robots shifted from the simple “random walk” approach to-
wards more evolved navigation schemes, involving a localiza-
tion technology at an aordable price. Up to now, no scientific
study has analyzed the potential impact of these newer robots
in terms of user’s acceptance or energy consumption.
We have carried out two distinct but complementary studies
in the present work. The remainder of this section summarizes
the questions at the center of both studies, and the contributions
gained by linking them together.
2.1. Designing Ecient Domestic Robots
The primary part of this study analyzes the current state of
the art and level of achievement in domestic robotics, with a fo-
cus on robotic vacuum cleaner and energy-related topics. The
robot must have several capabilities in order to fulfill its task:
1) A navigation strategy inside the environment, 2) a cleaning
device, and 3) some kind of interaction with the user, at least to
start and stop the cleaning process. An energy storage and man-
agement unit powers these functions. This view is illustrated in
Fig. 1a. As the energy is located at the center of this robotic
system, we will refer to it as an energetic agent in the course of
this work.
Some research results and design choices for the various
functions will impact the energy consumption of the mobile
system, and thus aect its autonomy. Within this study, we
aim to highlight the influence of these choices on the energy
consumption, in order to design energy-wise agents that are
compatible with a sustainable growth of the number of robots.
As we will see, localization and navigation strategies are the
main energy savers, and also bring some other benefits, but
more could be achieved by adding better planning and learn-
ing. Minimization of energy consumption for robotic vacuum
cleaners (and other robots) is an important topic for consider-
ation, especially with the growth of mass-market demands and
society’s dependence on energy.
This paper presents an analysis of the performance of several
existing robots, assessing the impact of the embedded technolo-
gies on the system’s fulfillments. We present the results from
a three-month study performed on a sample of seven robots,
bringing together the main trends on the market. The focus is
on technical metrics and therefore concerns mostly short-term
issues, answering the question: “Does it work well in a domes-
tic environment?” The key findings of this study aim at improv-
ing future designs, by identifying key technologies that enable
robots to be more ecient in their environments while at the
same time increasing acceptance by the end user.
2.2. Robots in Homes Are More Than a Technical Artifact
When deploying robots in people’s homes, it is important to
also consider broad human factors, as well as aspects dealing
with user needs, acceptance, and social implications. Once it
begins to be used by lay people in their private spaces, a robot
no longer remains simply a technical artifact; rather it becomes
a “social agent” [5]. A cleaning robot can have strong impacts
on its direct environment (“the home”), the tasks that are related
to it (e.g., cleaning), and the people in contact with it (“social
actors”) [4]. Fig. 1b summarizes the relations between this
social agent and its surrounding environment. Ideally, these
aspects should be integrated in the development of the robot us-
ing techniques such as “interaction design” or “design research”
In spite of the fact that several million robotic vacuum clean-
ers have already been sold, not much is known about how peo-
ple use and experience the presence of a service robot in their
homes. Questions arise regarding the extent to which the avail-
able robotic vacuum cleaners meet user needs and expectations
and how people actually use these devices. Our user study ad-
dresses these questions. The motivation of the user study was
to understand users’ perceptions, needs, and personal use of
a robotic vacuum cleaner. We aim to identify the challenges
brought on by the real world and people’s unique ways of us-
ing a robotic vacuum cleaner, and to devise how design could
improve these points.
With this user-centered approach, we aim to advance domes-
tic robotic vacuum cleaners with respect to several aspects: us-
ability, perceived usefulness, and design. These are important
factors for the adoption of technology in homes [13, 14]. By
understanding people’s expectations and their ways of using a
robotic vacuum cleaner, we can better meet users’ true needs
and take them into account in the process of developing these
types of devices and future technologies.
To address these aspects in a holistic fashion, we conducted
a six-month ethnographic study with nine households that were
given an iRobot Roomba robotic vacuum cleaner. This social
study was carried out in parallel to the technical study. In con-
trast to the technical study, the user study was motivated by the
desire to shed light on the long-term implications that robotic
devices might have in people’s homes. In this paper, we fo-
cus on presenting results on usage and user needs of the robotic
(a) Energetic agent (b) Social agent
Figure 1: Our dual view of the domestic robotic agent. The agent in (a) spends its energy for an number of functions, in order to fulfill its task, and these functions
in turn influence the energy consumption. The agent in (b) interacts with several elements that compose its environment.
cleaner, to leverage technical insights and provide relevant de-
sign guidelines. The detailed design and results of the ethno-
graphic study, including long-term implications and impact of
the robot on its ecosystem, are presented in another publication
3. Related Work
In the current state of the art for domestic robotics, no other
study to date has attempted to provide such a close match be-
tween the scientific challenges and user acceptance of the tech-
nology. As our approach is somewhat unique, we present the
related work for each topic separately.
3.1. Technical Study
Currently in the domestic environment, only a few types of
mobile robot have been mass-produced. The first successful
product, and now the most widespread, is the robotic vacuum
cleaner. The first research related to creation of a robotic vac-
uum cleaner dates back to the 1980s [16], while the first pro-
totype for domestic use can be dated back to 1991 [17]. Up to
now, studies have compared mobile domestic robots only from
an historical or purely technical point of view [18, 19]. They
do not take into account the most recent trends, like the use of
low-cost mapping technologies.
Other commercial applications of robotics to date have in-
cluded lawn-mowing, telepresence, and pool and gutter clean-
ing [20]. In the literature, other examples like assistive [21]
or rehabilitation robotics [22] can also be found. Most of the
research has focused on key aspects such as the navigation in
dynamic environments [23], or more broadly, the simultane-
ous localization and mapping problem (SLAM) [24–26]. Some
researchers have studied performance metrics, such as the cov-
erage of several domestic mobile robots performing a random
walk [27]. Again, this does not reflect the capabilities of the
latest technologies currently available. The question of energy
eciency for these kinds of appliances was only considered re-
cently, and only to point out the lack of regulations and stan-
dards compared to other home appliances [28].
Our study proposes to fill in the current gap in the state of the
art by studying a sample of the latest domestic robots, with a
special focus on the energy eciency of the overall system.
3.2. User Study
From a scientific viewpoint, surprisingly few evaluations of
domestic service robots in real households have been carried
out. For this study, we report on a pair of surveys, a set of inter-
views, and field studies that were carried out in people’s homes.
These dierent information-gathering techniques allowed us to
develop a strong user-centered view of the currently available
A pair of studies explored people’s general expectations of
robots and attitudes toward domestic robots [29, 30]. One im-
portant conclusion that could be drawn is that domestic robots
need to be evaluated separately from robots in general, as peo-
ple tend to hold dierent concepts of the two [6]. On one hand,
people overall have rather high expectations of robots and their
image of “the great mass of robotics” seems to be shaped by
what science fiction and novels present to them [29, 31]. On
the other hand, when people imagine a particular domestic ser-
vice robot, they have no clear idea of what it could do in their
household [29] and, accordingly, expectations of a robotic vac-
uum cleaner such as a Roomba are quite low [6]. Dautenhahn
et al. [32] described that people want to view home robots not
as friends but as machines, assistants, and servants that perform
various tasks for them. Furthermore, in terms of people’s per-
ception of robots, cultural background, gender, age, and other
personal factors seem to play a crucial role [29, 30].
Concerning the use of robotic floor cleaners in homes, Sung
et al. described the demographic profile of Roomba owners,
based on a survey among more than 350 Roomba users [33].
Against common expectations, they found that Roomba users
are equally likely to be men or women and tend to be younger
(<30 years) with higher levels of education or technical back-
grounds. Households with pets and children expressed greater
satisfaction with the robot, which implies that they might use
the robot dierently than do those who live alone. This eect
of household composition on how a robotic vacuum cleaner is
used has also been noted by Forlizzi et al. [5]. Besides in-
creasing their cleaning frequency with the Roomba, majority of
Roomba owners still performs extra cleaning with their manual
vacuum cleaners and most households had to make modifica-
tions to the physical layouts of their homes in order to make
the robot work well [4–6, 9, 15]. These findings point out that
domestic robotic vacuum cleaners could be improved to better
fit in people’s homes and thereby be more accepted.
Some concrete design suggestions come from Kim et al. [9]
and Sung et al. [34–36]. Kim et al. deployed four dierent vac-
uuming robots in Korean homes and identified a path-planning
behavior of the robot that met the assessed user needs [9]. These
researchers discovered a discrepancy between the cleaning path
people used when manually vacuuming and the paths chosen by
the robotic vacuum cleaners. Specifically, the actual user cleans
with methods unique to specific areas of the house, rather than
by following a technically optimal cleaning path. Based on this,
the authors suggest a robot’s path planning method should use
a layered map but also a cleaning area designation method re-
flecting each area’s characteristics. This goes along with peo-
ple’s wish to have an intelligent domestic robot that is able to
learn and adapt [6]. Furthermore, a domestic robot needs to
provide a certain amount of human control, be compatible with
the user’s domestic environment, and take gender into consider-
ation [34, 35]. Sung gives concrete suggestions for interaction
design with everyday robots in an unpublished document [36].
On the operation level, she suggests path planning, the robot
being able to learn, and reducing motor noise; on the communi-
cation level, she proposes allowing both autonomous and man-
ual control. On the engagement level, Sung’s guidelines include
customization and ensuring secure service.
Our user-centered study diers from previous work in several
aspects. Whereas Kim et al. [9] focused on the path-planning
of humans and robots when vacuuming, we were interested in
more generally exploring the usage and acceptance of vacuum
cleaning robots. Kim et al. let housewives use vacuum cleaning
robots for 10 days, whereas we followed up households over
6 months to be able to understand long-term eects, like user
acceptance factors. In this, our study is similar to Sung et al.’s
long-term study with Roombas in 30 households in the U.S.
[4]. Although it has methodological similarities [37], our work
diers in a sense that it was carried out several years later (and
technology had advanced meanwhile) and it was conducted in
a dierent cultural region. However, due to our small sample
size of 9 households, we are not able to generalize our results
and draw on cultural dierences. With our study, we are able to
show several similarities but also interesting dierences to the
findings of Sung et al. and contribute to a better understanding
of long-term eects of robots in homes.
4. Methodology
We used an orthogonal methodology for both of our ap-
proaches. On the one hand, seven samples of robots were tested
in a synthetic environment to bring out the main factors im-
pacting the energetic agent and its performances. On the other
hand, nine households took part in an ethnographic study with
a robotic vacuum cleaner to find out how people use and in-
tegrate the social agent in their cleaning routine according to
their needs.
4.1. Technical Study in Laboratory
The sample consisted of seven robot models, ranging from
the low-cost derivatives of the “Roomba” robot to recent prod-
ucts embedding more complex sensors and algorithms. These
products target the mass market with an aordable price (be-
tween $250 and $600). We classified them according to their
navigation strategy:
Robots 1 to 3 (Fig. 2a – 2c) follow a random walk using
some predefined behaviors (wall following, spirals, and
obstacle avoidance for example).
Robots 4 to 6 (Fig. 2d – 2f) perform Ceiling Visual SLAM
(CV-SLAM), implementing an algorithm similar to the
one described by [38].
Robot 7 (Fig. 2g) is fitted with a low-cost laser range scan-
ner performing 2D Laser SLAM [39].
Key Questions. First, we formulate a supporting equation to
help in our reasoning around the energetic agent. For a spe-
cific robot, let probot (t)be the instantaneous power drawn from
the battery, and Ttask be the time needed to complete the con-
sidered task. The total energy consumed to achieve the task is
Etotal =RTtask
0probot (t)dt. To represent the set of possible con-
figurations, let
αbe a set of generalized design parameters, like
the type of localization algorithm. These parameters will influ-
ence both probot (t)and Ttask . In addition, if we take into account
the eciency of the charging electronics, ηcharger, the parametric
total energy Etotal becomes
ηcharger ZTtask(
probot t,
One goal of this study is to help design more energy-aware
devices; that is, we explore the design space
αin order to min-
imize Etotal
α. In Eq. 1, two functions can be minimized
Random Navigation
(a) Robot 1
(Trisa Robo Clean)
(b) Robot 2
(Primotecq Mambo)
(c) Robot 3
(iRobot Roomba 770)
Ceiling Visual SLAM Laser SLAM
(d) Robot 4
(Samsung Navibot SR8855)
(e) Robot 5
(LG Hom-Bot VR5902)
(f) Robot 6
(Philips HomeRun FC9910)
(g) Robot 7
(Neato XV-11)
Figure 2: Robots used for the technical study, sorted by their localization technology. Copyright for (a) to (f): RTS /ABE; for (g): Neato Robotics.
by varying the design parameters: probot and Ttask. As we will
show, neither is independent, which makes the analysis of the
problem non-orthogonal.
The instantaneous power probot (t)comes from the “useful”
power on the one hand, and from the losses on the other hand.
The required power is minimized by removing useless functions
or fusing together several functions, leading, for instance, to a
decrease in the number of motors used. Losses are minimized
by increasing the robot’s eciency, for example by reducing
the numerous electrical and magnetic losses inside the motors,
as well as by reducing the Joule and switching losses inside the
electronics. For a mobile robot, the energy lost when braking
also accounts for a part of the total losses, and it can be par-
tially recovered by the addition of appropriate electronics [40].
The overall control, such as obstacle avoidance, is equally im-
portant, in order for the robot to follow a smooth trajectory and
avoid unnecessary braking. A modified trajectory will, in most
cases, influence the completion time.
The other function to be minimized is the completion time
Ttask. In this case, increasing the robot’s speed is often useless
because it will increase the instantaneous power accordingly.
Better planning and navigation are the keys for this strategy to
succeed. When complete coverage is desired or required, as
in cleaning, patrolling, or lawn-mowing tasks, a path planning
coupled with a localization strategy will cut down the coverage
time compared to a random walk approach. Recent develop-
ments in the semiconductor industry for mobile applications,
coupled with algorithmic and mechatronic advances such as
the laser scanner of [39], have made the simultaneous local-
ization and mapping (SLAM) aordable for the mass market.
This benefit comes at the price of extra sensors and computa-
tional power needed to achieve an ecient localization, which
conflicts with the reduction of the instantaneous power.
The concern is thus to choose the best technology in order to
reach a global trade-o. This article will assess, among other
factors, the eect of the navigation strategy on the total energy
based on measures performed with real mobile domestic robots.
Experimental Setups. During this study, we explored several
performance metrics for each of the fields emphasized in Fig.
1a, making a link with the energy consumption when possible.
The navigation (Sec. 5.2) is tested in the setup depicted in
Fig. 3a. It reproduces a two-room apartment with a total inter-
nal surface area of 15.5 m2made of a concrete floor. Robots are
started from their base station located in the upper right corner.
The evolution of the coverage as a function of time is measured
using the overhead camera (Fig. 3b) and a custom tracking soft-
ware. The experiment is stopped when the robot returns to its
base station or when it runs out of battery power. Between five
and eleven trials were conducted with each robot.
A simple setup (Fig. 3c) is used to measure the cleaning
capabilities (Sec. 5.3). It consists of a square surface (2 by 2
meters). Several surfacings were used: 1) smooth concrete; 2)
short-pile carpet, and 3) a crack with a cross-section measuring
14 mm by 6 mm. Dust was simulated using a mixture exhibiting
a broad granularity range: 5 g wheat flour, 5g of fine sand, and
5 g wood shavings. This was randomly spread by hand over
the central square (1 m2) to avoid interference with the edges.
The experiment was stopped when the robot returned to its base
station or when the elapsed time reached 7 minutes 30 seconds.
The collected material was weighed using a laboratory scale
with a resolution of 103g. Three trials were done for each
Concerning the figures for the energy consumption used
throughout this work, both the global and the intrinsic instanta-
neous powers were measured in the setup shown in Fig. 3a.
For the instantaneous power, a wireless datalogger working
at 1 kHz was hooked to the battery and used to measure the
robot’s in situ power probot (t)during operation. As for the
(a) (b) (c)
Figure 3: The experimental setups for the technical study. (a) Arena reproducing an apartment, side view. (b) Arena reproducing an apartment, from the overhead
camera. The white ellipse marks an area of dicult access. (c) Simple arena for cleaning tests.
global power, a power analysis bench1was used to measure the
overall energy drawn by the charging station during a complete
recharge of the battery. Dividing the overall energy drawn by
the run time gives the average power consumption.
4.2. User Study of Robot Usage “In the Wild”
In parallel with the technical study, we deployed nine iRobot
Roombas for six months in dierent households and studied
how people used and lived with these robots in their homes.
The main motivation of this long-lasting ethnographic study
was to understand the process of adoption, social implications
and usage patterns, and to find factors that impact the long-term
acceptance of these types of devices. The identified usage pat-
terns and user needs can be closely related to the findings from
the technical study.
Course of the Study. The study was conducted from March to
October 2011 and consisted of five home visits over the six
months at each household in which we deployed a robot, as
follows: (1) one week prior to handing out the iRobot Roomba,
(2) the initial introduction, (3) two weeks after deployment of
the robot, (4) two months after deployment, and (5) six months
after. Several qualitative and quantitative measurements were
used to capture users’ feedback in order to derive usage pat-
terns and needs: at each visit, semi-structured interviews (1 –
1.5 hours) were conducted in which participants were asked to
describe how they used the robot, their satisfaction with it, and
perceived benefits and constraints. In addition, we collected
field notes, photos, and videos from the on-site observations.
During the ten days before each visit, each household filled out
a daily diary to document cleaning activities and Roomba usage
(starting with the second visit). The methodological setup was
inspired by Sung et al. [37]. We were assisted by an ethnogra-
pher during the entirety of the study.
Participants. We recruited nine households from the area of
Lausanne in the French-speaking part of Switzerland. The
1Alciom PowerSpy:
sample consists of two single-person households (a 40-year-
old male and a 71-year-old woman with a dog), a couple in
their early 60s with two cats, and six families. The families
had between one and four children; these children were aged
6 months to 18 years. Some of the families had a cat or dog.
Households had culturally dierent backgrounds. As previous
studies showed that the physical layout of the home as well as
the composition of the household can impact a family’s use of
a robotic vacuum cleaner, we recruited a range of households
living in studios, apartments, or houses, with and without chil-
dren or pets, with working mothers and housewives, and with
and without a cleaning service. Those households without a pet
received a Roomba 520 (Fig. 4a) whereas pet-owners received
a 563 PET version (Fig. 4b). We did not expect to see dier-
ences in how these two dierent models were used and also did
not look for cultural dierences impacting use, as the broader
picture was of more interest to us.
Data Analysis. We collected a comprehensive and diverse set
of data from our 44 household visits (five were planned at each
of the nine households; one of these had to be canceled). The
data considered here consist of audio transcripts from the in-
terviews, photos, videos, written field notes, and paper-pencil
diaries that people kept about their cleaning routines (descrip-
tions about who cleans what, when, where, how, and why), as
shown in Fig. 4c. The analysis of these mostly qualitative data
is based on “Grounded Theory” [41, 42] and the “Method to
Analyze User Behavior in Home Environment” [43]. An exam-
ple of how Grounded Theory can be applied to an ethnographic
study on human-robot interaction can be found in [44]. The di-
ary entries provided quantitative data that were subjected to a
descriptive statistical analysis. In total, we were able to con-
sider n=634 distinct cleaning activities from the diaries, of
which 193 were activities that involved the robotic vacuum and
65 involved the traditional vacuum cleaner. Our findings are
based on these quantitative data and are enhanced using partic-
ipants’ qualitative statements.
(a) Roomba 520 (b) Roomba 563 PET (c)
Figure 4: The experimental methodology for the user study. (a) and (b) are the robots used for the user study. Copyright: iRobot Corporation. (c) Sample of a field
diary describing the cleaning routines.
5. Results
Results from both studies will be presented side-by-side to
better highlight synergies. Sec. 5.1 will specifically focus on
some of the power consumption issues. The remaining results
are grouped by the research fields identified in Fig. 1a, which
are the navigation (5.2), the cleaning eciency (5.3), and the
interactions with the user (5.4).
5.1. Anatomy of the Power
Ecient analysis of the energy issues related to a robotic sys-
tem requires that one first knows the relative impact of each sub-
system inside the total budget. We want to understand where
the power is being used from a systemic point of view. Simple,
yet informative analysis methods are used to reveal which sub-
systems correlate with specific power usage. The relevance of
energy eciency for the user is likewise considered.
5.1.1. Technical Study
An in situ analysis of the consumed power was first per-
formed by placing an embedded, single-channel datalogger
module between the battery and the robot. This module records
the power consumed during the whole experiment. It gives
a good, but somewhat indirect insight into the system. This
method is easy to put into practice, as the battery is easily re-
movable. Two informative plots are drawn in Fig. 5. On the
other end, a multichannel datalogger would give direct figures
for each subsystem, but considerable eort would be required
to hack the robot’s electronics.
On Fig. 5b, the startup sequence of the cleaning process for
Robot 7 (Laser SLAM) is clearly visible. Starting from the
idle state, the following phases can be identified: 1) the laser’s
spinning motor starts and stabilizes; 2) the powerful suction
fan starts; 3) the main brush starts to rotate (no side brush);
and 4) finally, the robot starts the driving motors and begins to
clean. It can be deduced that the Laser SLAM itself consumes
only 1.9 W (6.3 % of the total cleaning power), compared to
the cleaning subsystem, which takes 23.8 W (78.8 %). The mo-
bility accounts for 2.5 W (8.3 %). For this specific robot, the
power used for the navigation functions is a marginal addition
compared to the cleaning device.
Such a clear breakdown, however, is not always possible with
our simple method. One example of this is in the case of Robot
5 (Fig. 5a). The first small increase in power (black circle) is
devised to be due to the Visual SLAM subsystem (camera and
algorithm). It takes about 1.1 W of extra power, or 8.4 % of the
total power that is typically consumed by an embedded proces-
sor. All the motors start together and thus cannot be evaluated
The influence of the navigation subsystem on the power bud-
get can be further studied. Fig. 7 plots the distribution of the in
situ power consumption of each robot in two working cases. Let
us first consider the idle case, when the robot is turned on, but
not moving. The three robots performing CV-SLAM are, not
surprisingly, among the top consumers, as the additional em-
bedded processor will need between 0.5 to 1 W of extra power,
even when not processing any data.
When considering the cleaning case, things are completely
dierent. The previous increase, due to the extra processing
power, is largely overwhelmed by the dierence due to the driv-
ing and cleaning motors. Thus, the addition of the SLAM repre-
sents only a small part of the total consumption when compared
to the energy required for moving and cleaning. As we will see
in Sec. 5.2, SLAM-enabled robots benefit from the acceleration
of coverage, saving energy during the overall process.
A power analysis was also performed directly at the plug of
the recharge station. One initial observation is the high idle
power of the sole base station, with the worst result noted in
the case of Robot 2 (up to 3.5 W). When the charged robot is
left connected to its base station, the result further deteriorates,
as power consumption increases to between 3.2 and 8.1 W de-
pending on the robot. Unfortunately, this type of appliance is
not, at present, bound by any regulations similar to the Euro-
pean regulation 1275/2008 [45], which limits the standby mode
to 2 W. This represents a serious concern for these types of
mass-produced electrical appliances.
The eciency of the recharge station ηcharger was also com-
puted as
ηcharger =1
Etotal ZTtask
probot (t)dt ,(2)
where Etotal is the energy consumed at the plug to recharge the
robot, and probot (t)is the in situ power measured on the robot
during the whole process.
This eciency figure includes the intrinsic quality of the
charger as well as the storage eciency of the battery subsys-
tem, and varies between 0.33 and 0.84 in our study. The best
two robots are the ones that use Li-ion batteries, while the oth-
(a) Robot 5: Cleaning on concrete. The black circle pinpoints
the startup of the CV-SLAM process, just before the robot starts
moving and cleaning.
(b) Robot 7: Carpet cleaningstartupprocesses. 1: laser 2: suction
3: brushes 4: driving motors.
Figure 5: Two plots from the in situ power measures.
Figure 7: Task-related in situ power measured for each robot. Around 1000
samples have been used for each dataset.
ers use the Ni-MH technology. In the case of the two worst
robots, more than 50 % of the recharge power is already lost at
the plug.
5.1.2. User Study
The user study confirmed the importance of the topics we
addressed in our technical study. All nine households surveyed
considered the energy usage of the robot, sometimes related to
more practical reasons like the autonomy and the charging time.
Interestingly, we did not have to bring up this topic; all of the
participants independently asked for more information concern-
ing the robot’s energy use, or told us that they were concerned
about the fact that the robot’s charging stations needed to be
plugged in at all times.
Our findings indicate that users wish to have more trans-
parency with respect to energy usage. Some other home tech-
nologies (such as washing machines or dishwashers) already
indicate their level of electricity consumption according to Eu-
ropean standards. The reasons given by the participants for de-
siring reasonably low energy consumption on the part of the
robot ranged from the financial importance to factors related to
the environment and healthy living.
Most households completely switched othe robot and un-
plugged its charging station when the device was not in use to
avoid overly high energy consumption. However, this hinders
a programmable robot from starting a cleaning session when
scheduled and thus undermines one of its most valuable facets:
the ability to clean autonomously. For robotics in general, this
shows how closely the aspect of reducing the energy consump-
tion is related to user satisfaction and the acceptance of techni-
cal devices in homes.
5.1.3. Joint Outcomes
The user study indicated that the end-users do care about the
energy consumption. This fact even leads users to overreact by
unpluging the charging station to avoid idle losses. This ac-
tion hinders the robot from working in an autonomous way, as
a researcher would expects in the first place. On the other hand,
measures performed on real hardware show poor energy perfor-
mances in idle mode, with no incentives for better performance,
as the current products are not bound by any regulations.
This can be fought using several tools. Better electronics, of
course, is the first key to success, but the user should not be
ignored. A first step goes through a restrictive regulation, but
energy-awareness should also be advertised to the user using
adequate feedback. The system should be able to provide infor-
mation regarding past and present operations of the robot. As
feedback appears to be a key issue in our study, we will discuss
this point further in the coming sections.
Finally, the energy autonomy will also be of prime impor-
tance for the end-user, especially in the case of wide spaces.
A robot should be able to cover the space under considera-
tion in a single session, without requiring intervention from
the user. This is directly related to the energy consumption
at the heart of this work, and an improvement in the field of
the energy consumption will thus directly benefit the end-users.
Laser-based and vision-based mapping technologies induce an
(a) Evolution of the average coverage over one hour. (b) Completion time for each robot. (c) The specific energy, in Jm2, for each robot.
(d) Impact rate for each robot.
Figure 6: Analysis of navigation performance for the seven robots.
increased power consumption by 1 to 3 W, according to our
measurements. The question remains whether they will be able
to save energy on the overall process.
5.2. Navigation Eciency
A cleaning robot, like some other domestic robots, should
not only be low-power and energy-ecient, but should also ef-
fectively cover the area of importance. Overall planning, such
as the rooms to cover, the order in which the rooms should be
covered, and the cleaning tasks’ frequency are important topics
for the user.
5.2.1. Technical Study
Coverage Analysis. The trajectories were recorded using an
overhead video tracking system when the robot was engaged
in cleaning the apartment space shown in Fig. 3a. A sample
of each trajectory is shown in Fig. 8. It clearly shows distinct
strategies for each robot.
To estimate the surface covered, the image analysis integrates
over time the surface hidden by the robot’s shape. For this pur-
pose, the Gaussian mixture-based background segmentation of
[46] is applied to the calibrated pictures. This estimation does
not take into account the side brushes used by most robots, or
the width of the main brush under the robot. In the results, hid-
den areas (like under the sofa) are not taken into account. The
cleaning eciency per surface unit is considered separately in
Sec. 5.3.
The evolution of the coverage, as a function of time and av-
eraged between all the trials, is plotted in Fig. 6a. The SLAM-
enabled robots seem much faster than the others. This is con-
firmed by the computed completion times shown in Fig. 6b.
Localization-less robots have no robust way to compute the
achieved coverage. Consequently, they do not return to their
base stations when the coverage reaches a steady state, and most
of the time they dock to their stations after an extended period
of time. On the contrary, robot 6, which is the slowest among
the robots performing SLAM, is three times faster on average
than are the random walk robots.
Regarding the achieved coverage, random walk robots take
time but achieve a robust coverage. On the other hand, robot
4, and to a lesser degree Robot 5 (both relying on vision), un-
derperform compared to the others. Looking back at the im-
age analysis, it appears that some places are harder to reach for
them. One of these places is between the sofa, the intermediate
wall, and the bin (white ellipse in Fig. 3b). In half of the runs,
Robots 4 and 5 were unable to reach this place, thereby losing
part of the coverage. On the contrary, Robot 6 was successful
on all of its 11 runs, as its path planning uses thinner bands, as
one can see in Fig. 8f. While some time is lost by this strategy,
it gains greatly in robustness and coverage.
Specific Energy. One of our key question is the influence of de-
sign parameters on the energy consumption, and especially the
navigation strategy. To answer this question, we now compare
the coverage strategy, with respect to the energy. For this, we
define the specific energy, which is the energy needed to cover
Random navigation
(a) Robot 1 (b) Robot 2 (c) Robot 3
Ceiling Visual SLAM
(d) Robot 4 (e) Robot 5 (f) Robot 6
Laser SLAM
(g) Robot 7
Figure 8: Sample of the trajectories for each robot, grouped by localization strategy.
(a) (b)
(c) (d)
Figure 9: Sample of the expected trajectories drawn by several users.
1 m2of floor. It equals
Especific =1
Aeective ZTtask
probot (t)dt ,(3)
where Aeective is the surface eectively covered, as deduced
from the previous coverage analysis. Fig. 6c shows the re-
sults, which clearly demonstrate the eectiveness of the SLAM-
enabled robots over those employing random-walk methods,
counterbalancing the increase of power by a drastically reduced
Ttask. However, no clear conclusion can be drawn between CV-
SLAM and Laser SLAM robots, even if Robot 6 (CV-SLAM)
is two times more energy ecient than are the other robots.
Impact Rate. The mitigation of the eects from the robot im-
pacting or colliding with objects is closely related to the user’s
wish regarding preservation of potentially fragile objects. Dur-
ing the course of our experiments, an accelerometer was placed
inside a ballasted dustbin to record the number of hits. This
setup was located in a corner of the room (bottom right of Fig.
3b). The number of impacts per hour is plotted for each robot
in Fig. 6d.
Obstacle avoidance relies primarily on the proximity sen-
sors. Most robots use infrared (IR) proximity sensors with the
number of sensors employed varied between two (Robot 2) and
seven (Robot 4 and Robot 6). Robot 5 was the only robot that
used five ultrasound sensors (with an addition of two long-range
IR sensors on the sides); Robot 7 relied on its laser scanner with
a short-range IR sensor on the side to allow it to accurately fol-
low walls.
If we take robots using only IR and ultrasound sensors
(Robots 1 to 6), no sharp claim can be made regarding better
performance when using a mapping technology. Robots 4 and
5 (CV-SLAM) perform comparably to Robot 1 (random walk),
even if Robots 2 and 3 (random walk) perform far worse. Robot
6 (CV-SLAM), however, is the best performer in our sample.
One can also notice the slower coverage in this case (Fig. 6b),
as the robot clearly slows down when arriving near an obsta-
cle, where others maintain their speed up to the point of con-
tact. Robot 7 (Laser SLAM) also performs well, since its sensor
readily provides a dense map of obstacles.
5.2.2. User Study
Coverage. The robot’s cleaning path is closely related to the
users’ perception of how eciently the robot cleans. In agree-
ment with the results of Kim et al. [9], we found that users de-
sire the robot to plan its path intelligently according to specific
aspects such as area /sub-area of a room, floor material, and
level of dirt. People generally want a cleaning robot to cover
the whole floor. Participants were particularly annoyed when
the robot left some spots uncleaned even in the middle of the
room. For this reason, good coverage of the entire space to be
cleaned is very important.
Other specifics emerged about the satisfaction (or lack
thereof) of the users with the Roomba. In one instance, a mother
expected the robot to clean the whole room in a short amount
of time. She left it for only five minutes in the spacious liv-
ing room, and then she switched it oto carry it to another
room, only to find that the living room had not been completely
cleaned. The problem, in this instance, is that the robot is being
used in a way that is technically dierent from how it is in-
tended to work, coupled with insucient feedback. The robot
required patience from the user side, which was cited as a di-
culty for some of the households in the user study. Thus, swift
coverage is also of prime importance.
Cleaning Habits. Another aspect related to the navigation of
the robot is how people wish to use the robot. In agreement
with Kim et al. [9], we also found that users want to use the
robot according to specific spaces (for example, in the hallway,
kitchen, or living room), and that users often have specific ideas
concerning how a robot should progress through the cleaning of
these spaces. The robot, however, does not follow such a path.
Some participants also wished that the robot would clean areas
in a specific order; one example of this would be to vacuum
dirty spots such as under the bed at the end (also in accordance
with [9]).
We analyzed participants’ cleaning diaries with respect to
where vacuuming was performed (Table 1). Interestingly, the
robot is used in some rooms much more often than in others,
and the distribution was not similar to where the manual vac-
uum cleaner was used. Whereas the hand-operated vacuum
cleaner was most often used in the bathroom (35 %), this was
the room where Roomba was used least often (2 %). Asked
why, participants answered that they would be afraid of let-
ting the robot pass over wet spots. Participants also indicated
that bathrooms were tiny rooms and conveyed the belief that
the robot was not adept at cleaning these spaces. Conversely,
35 % of all Roomba cleanings took place in the kitchen, while
it comprised only 17 % of all hand-operated vacuum cleanings.
An interpretation of this user behavior is that the kitchen is a
place where frequent quick spot cleaning is required and that
the Roomba is preferred over the manual vacuum cleaner for
this kind of cleaning. A possible explanation is that it takes
longer to have the manual vacuum cleaner ready for vacuum-
ing: it needs to be taken from the closet, plugged in, adjusted,
plugged out, and re-stored, whereas a robot can be used right
Planning Transparency. The user’s viewpoint of how the
robot’s path planning is perceived is important for the accep-
tance of the robot. A robotic vacuum cleaner needs to be able
to trace a smart, smooth, and ecient path that is comprehen-
sible to some extent to humans. Before participants received a
Roomba, we asked them to draw a path of how they expected
the robot to move around their home (Fig. 9). Most people
imagined the robot would go room by room and would spend
more time cleaning in areas where more dirt was likely to be
found; for example, around the dining room (crumbs from eat-
ing, etc.) (Fig. 9b and 9c). One user was convinced that the
robot would go back to where she put it for the start of its clean-
ing path, namely at the entrance of the house (Fig. 9a). One
Table 1: Cleaning frequencies by places (in %), when using the vacuum cleaner
and the robotic cleaner, analyzed from the cleaning diaries; not all possible
places are listed.
Vacuum Robotic
Where? Cleaner Cleaner
(%) (%)
Bathroom 35 2
Kitchen 17 35
Children room 17 17
Bedroom 14 10
Living room 10 28
family in which the father was a computer scientist imagined
rectangular movements (Fig. 9d). Overall, people had no clear
idea about how the robot would plan its path.
When people used the Roomba for the first couple of times,
members of all nine households carefully watched how it
moved around their homes. Most participants attempted to
understand how and why the robot “decided” its path. Most
households were skeptical about the Roomba’s random path
and one mother expressed her disappointment: “How does it
decide where it goes? It’s stupid, it does not see where the
dirt is, it always moves away from it!” One family enjoyed the
robot’s unpredictability but all remaining eight households de-
scribed the behavior as “stupid” and “not understandable. This
unpredictability made the robot appear uncontrollable to them
and six of the nine households did not want to rely on it; they
were afraid to let it clean when they were not at home. The ver-
dict was that the robot could not be trusted enough, especially
in the presence of fragile objects on the ground.
This works against an early acceptance of the robot. Robotic
vacuum cleaners are not entertainment robots; they are there
there to fulfill a routine task. As we found in the user study,
people’s first impression about the robot had a strong impact
on the long-term acceptance of it. Thus, preferably, a domes-
tic service robot convinces people from the very beginning by
making them understand how it works, by being transparent in
its path-planning so it is “understandable,” and users have fewer
worries about its random path. This goes along with the aspect
of “transparent robot behavior” [47]. People wish to understand
the robot’s navigation as this leads to the impression of having
the system under control.
5.2.3. Joint Outcomes
The user study makes it clear that the current robots do not
match users’ preconceptions. Often, the user does not under-
stand what the robot is doing, and sometimes experiences frus-
tration or loses patience. On this point, robots using SLAM
have a clear advantage by being systematic, and thus more pre-
dictable, in addition to being faster and sparing energy. From
the user perspective, SLAM can play a key role to enhance user
acceptance, and support the robot to provide services matching
users’ expectations.
Nonetheless, the feedback given by the robot could be greatly
Figure 10: Analysis of the cleaning task: Surface-related in situ power measured for each robot. More than 1000 samples have been used for each dataset.
extended to improve user’s awareness and to make the device
more trustworthy. With the house’s map at hand, the robot
could communicate the estimated completion time, cleaned and
uncleaned areas, or locations where it often encounters trouble
and requires user’s assistance. The communication medium,
for example, could be an embedded screen, or perhaps even an
application for a smartphone.
The user study also reveals how the cleaning habits are seg-
mented both in time and space. Not all the areas have the same
functional and emotional importance, nor do they require clean-
ing with the same regularity. Again, SLAM robots could reach
a higher level by learning these patterns (with some feedback
from the user), and fuse them with the map in order to adap-
tively clean. Areas with a higher density of dust can be cleaned
first, and more often. Another benefit is an increased energy
eciency, as only the necessary amount of work is performed.
Finally, we have the necessary elements to answer this ques-
tion: compared to a manual vacuum cleaner, does a robot per-
form better in term of energy? We will focus on the setup of Fig.
3a. With a robot consuming roughly 20W, it will take around
15 minutes for the fastest one, equaling an expense of 18 kJ.
On the contrary, a human will take about 5 minutes with a vac-
uum cleaner consuming 1000 W, equaling an expense of 300 kJ.
The robot takes about 15 times less energy. Even if people use
the robot every day, they will still save energy compared to a
weekly vacuuming.
5.3. Cleaning
Vacuum cleaning may seem to be an easy task at first, but in
reality it is quite challenging, as the robot will evolve on a broad
diversity of surfaces and face heterogeneity in the material to
collect. The design of the embedded system must also take into
account the limited power at the robot’s disposal, as well as the
limited space available to integrate the cleaning system. In this
section, we will study the influence of the environment on the
cleaning task and describe people’s dierent cleaning strategies
to ascertain how a robotic vacuum cleaner could meet them.
5.3.1. Technical Study
Fig. 10 shows the averaged in situ power measured for the
robots when cleaning two type of surfaces, namely concrete
and a short-pile carpet2. The power distribution clearly shifts
upward for all the robots when cleaning the carpet. This is ex-
plained by the increased current due to the additional frictional
resistance on the cleaning brushes. Cleaning a rough surface
requires more power.
From the user’s perspective, cleaning eciency is one of
the most important factors that determines the usefulness of a
robotic vacuum. Eciency figures were measured on three dif-
ferent surfaces, according to the setup described in Sec. 4.1,
and averaged on three trials. These figures were then compared
to the averaged in situ power previously measured. Plots are
given, as a function of the surface’s type, in Figs. 11a to 11c.
In the case of the concrete surface (Fig. 11a), no relationship
between the cleaning eciency and the robot’s power can be
seen (using a linear fit, goodness of fit R2is only 0.018). Most
robots scored about 90 % with respect to the amount of material
collected during cleaning. For comparison, the same test con-
ducted with a manual vacuum cleaner (Dyson DC05) showed
an eciency above 98 %.
In the case of the carpet (Fig. 11b), cleaning eciency
ηcleaning,carpet exhibits a moderate dependency on the robot’s
power, as shown by the linear regression
ηcleaning,carpet =0.007 ·probot 0.092 (R2=0.56) ,
where probot is the averaged in situ power. The overall eciency
remains poor (below 35 %) in all cases.
Finally, for the cleaning of the 14 mm by 6 mm crack (Fig.
11c), data are quite well explained by a linear regression
ηcleaning,crack =0.041 ·probot 0.47 (R2=0.74) .
In conclusion, the suction power does not really help when
dealing with a flat and smooth surface. The design of the
brushes is the primary concern in this case. However, suction
power does become the main tool on a hard and uneven surface.
5.3.2. User Study
Each household cleans dierently, according to the physical
characteristics of their home and their personal preferences. In
2All robots were unable to clean a long-pile carpet due to the small distance
between the ground and the robot’s frame.
(a) (b) (c)
Figure 11: Analysis of the cleaning task. (a) – (c): Cleaning eciency as a function of the robot’s averaged in situ power when cleaning on concrete (a), on a carpet
(b), or when cleaning a 6-mm-deep crack (c). The green line shows the linear regression performed on the data. Three trials were done each time.
cleaning, there is no “right” or “wrong.” In turn, one cannot
make general statements about how people clean. We would
like to give a rough idea of how the households in our case
study used their manual and robotic vacuum cleaners. We de-
scribe how usage patterns are determined by the housekeeper’s
personal conviction of cleanliness and reflected in a specific
cleaning strategy.
On average, participants vacuumed their homes once per
week or once every other week. Only one household with a
dog that shed a lot of hair used to vacuum three times per week
and one single person household only once every two months.
The three households that integrated the robot in their clean-
ing routine used the Roomba on a daily basis, much more fre-
quently than they used their manual vacuum cleaner. Still with
the robot, the manual vacuum cleaner was used from time to
time to clean well in corners. Whereas the manual vacuum
cleaner was used most often on weekends (41 %), the robot was
used equally often on every day of the week, with a small peak
on Wednesdays (20 %). Detailed results are plotted in Fig. 12.
This shows how the robot changes people’s cleaning frequency,
by using the robot more regularly all over the week. How-
ever, this change could only be observed in the households that
adopted the robot. A household was considered as “adopter”
when they kept using the robotic vacuum cleaner during the six
months of the study and expressed interest in buying it if we
were to take it from them.
When considering various floor surfaces, people had dier-
ent opinions about how well the robot cleaned on carpet, tiles
or parquet floor. However, participants consistently found the
robot had diculties with the transitions between dierent floor
types and lost dirt it had already collected when moving onto a
carpet, for instance.
Users prefer that a robotic vacuum cleaner would meet their
standard of cleanliness as well as their expectations of “how
cleaning is done.” However, people have dierent attitudes to-
wards cleaning and cleanliness. To understand better how a
robot could meet a user’s needs, we derived four types of clean-
ing strategies based on the motivation that a household shows
to keep the home clean and the eorts made and time spent for
cleaning [15]:
Figure 12: Frequency of use in % for the vacuum cleaner (VC) and the robotic
cleaner, as a function of the days of the week. nvalues indicate size of each
Spartan cleaners barely notice dirt and do very little about it.
They have no or low motivation to clean, and hardly use
the few cleaning tools they do possess. Vacuuming once
every two months might be considered a typical behavior.
However, in spite of this, they feel comfortable in their en-
vironments because cleaning is not of great importance to
them. In our sample, one of the spartan households always
set an alert on a cell phone to be reminded to vacuum; it
would have been forgotten otherwise.
A robotic vacuum cleaner could more easily suit the clean-
ing standard of the spartan cleaners as it tends to be not
very high. Given that the robot would clean eciently, it
could even replace the vacuum cleaner. Further, a clean-
ing robot could help spartan cleaners by automatically
starting a cleaning session from time to time. A robotic
vacuum cleaner for spartan cleaners could be fairly au-
tonomous and clean while they are not at home.
Minimalistic cleaners notice dirt around the house that makes
them feel a little uncomfortable (which creates some in-
trinsic motivation to clean). They do what is necessary but
not more. Vacuuming is done only when they have time to
do so. Cleanliness is not a priority.
Since minimalistic cleaners do not like to spend a lot of
time tidying and cleaning, a robotic vacuum that meets
their needs would have to be always ready for use (“op-
portunistic cleaning” [5]), small, and time-ecient. It also
would have to work somewhat silently and discreetly to
not make the user think about cleaning.
Caring cleaners really care to have a clean and nice-looking
home to show guests that they have a well-working “home
ecosystem” (which creates some extrinsic motivation).
They like to keep the home clean (or engage a cleaning
service) and enjoy cleanliness and order.
Ensuring a healthy environment for their families is cen-
tral to caring cleaners. A robotic vacuum based on their
needs could have a more meticulous cleaning path, with
a detailed report of the collected dust (where, how much)
and a visible energy-saving function.
Manic cleaners clean almost obsessively. They are very picky,
notice every little piece of dirt and every blemish, and
probably constantly feel the pressure to clean or tidy up.
Accordingly, they permanently spend a lot of time tidy-
ing up and cleaning tasks are a priority for them. They
would not engage a cleaning service because then the state
of their homes would not be under their control anymore.
It is dicult to meet the standards and needs of manic
cleaners as they feel a strong need to control their sur-
roundings. For this reason, a cleaning robot would be
more of an extra cleaning tool than a partial replacement
for their vacuum cleaner. A less autonomous robot is
suggested, which could be precisely scheduled (time and
space), possibly remote controlled, and would also need
to be discreet while providing perceivable value in how it
cleaned (as manic cleaners would tend to observe the robot
and not leave it “alone”) by eventually moving slower and
remaining longer in edges and corners.
This classification is based on our ethnographic study in only
nine homes and is therefore eventually not generalizable. How-
ever, what is important here is to shed light on people’s dierent
cleaning strategies and also dierent expectations of a robotic
vacuum cleaner.
5.3.3. Joint Outcomes
People have dierent expectations regarding the cleaning
task. We were able to classify our households into four cate-
gories in terms of the frequency of cleanings and the level of
expectations. The needs of spartan and minimalistic cleaners
are already addressed by current robots. This is not the case for
caring cleaners, not to mention manic ones. To address these
shortcomings, the robot should be configurable, in order for the
robot’s design to reflect people’s dierent needs of cleanliness
and patterns of using the robot. This implies a varying level of
autonomous decision (when to clean), configurable and adap-
tive path planning (where to clean), and the feedback given to
the user (how it was cleaned).
Robots with a low power consumption clean equally well on
smooth surfaces, but still have a cleaning eciency slightly be-
low a vacuum cleaner. On uneven surfaces, the suction power
becomes important, whereas the eciency on carpets is low in
all cases. Our recommendation is to adapt the cleaning power
to the type of surface using a classification algorithm.
5.4. User Interactions
In the following, we describe how the user and the respective
robotic vacuum cleaner interact with each other. The interac-
tion can be direct or indirect and be split into three main parts:
how users operate and give commands to the robot; how the
system gives feedback to the user; and other more indirect in-
teraction with the user’s and robot’s shared environment, such
as the noise the robot makes and its visibility during the clean-
ing process.
Operating the robot. The control of the robot is usually kept
very simple, with one or several push buttons for immediate
operation, and with a daily or weekly timer for scheduled tasks.
The only operating option is to clean the whole surface, which
is not the most energy-ecient, as not all the rooms would need
to be cleaned with the same schedule. A rich user-friendly in-
terface, like a touch screen, would provide the user with more
control over task planning. To ease the burden on the user, a
map combined with a capacitive dust sensor would enable the
robot to learn where dirt was more likely to accumulate in the
cleaning area. This would allow the robot a higher degree of
autonomy in being able to choose the high-priority places to
clean. This is related to the work of Kim et al. [9].
Feedback from the robot. The feedback is limited to a few
LEDs, or a screen with very limited information. With the ad-
vent of mapping technologies, information could cover broader
aspects, like where the robot is intending to go. This has the
goal of reassuring the user. The robot could also provide a kind
of “cleaning report,” which would give more transparency to the
user with respect to what the system achieved, what has to be
done, and so forth. Ultimately, this could enhance acceptance
and adoption of the robotic vacuum cleaner.
With respect to human-robot interaction, participants appre-
ciated the audio-feedback (dierent sounds) as well as the spo-
ken verbal feedback of the Roomba, as no other technical prod-
uct they possessed would use this kind of feedback. Children,
in particular, reacted very positively to the variety of sounds
that the Roomba played. However, some parents were afraid
this would be too engaging for the children and that their chil-
dren would not stop playing with the robot. Thus, sounds could
probably be more functional and less entertaining but we also
found that parent’s role of introducing the robot to the children
influences how they approach it. Several participants suggested
the robot could understand verbal commands or gestures to help
it find the dirty spots in the apartment. The idea of using verbal
feedback was suggested by some of the users but strongly re-
jected by others, as they found it intimidating. Spoken feedback
can probably be integrated as an option and it is up to the user
to choose it.
Figure 13: A-weighted sound power as a function of the averaged in situ power
for each robot. The green line shows the linear regression performed on the
Noise. The noise emitted by the robot is also an important fac-
tor behind long-term acceptance. A noisy device will be less
attractive, especially if the user wants to keep an eye on it to
monitor the cleaning process. One of the families in the user
study wanted to use their Roomba overnight. Although the bed-
rooms were located on the second floor, they still woke up due
to the noise the robot made in the kitchen on the first floor. Con-
versely, noise is not a problem when the system works while
no one is at home, which is also the intended use of a robotic
Furthermore, people’s dierent cleaning strategies and ways
of using the robot require the system to be as quiet as possible.
Some users, especially caring cleaners, would also tend to keep
an eye on the robot when it was in use but would still wish to
do “multi-tasking.”
The total sound power level LWwas measured according to
the norm NF EN ISO 3743-1 in a reverberating room, using
the mid-band frequencies between 125 Hz and 8 kHz. The A-
weighted power level LWA for each robot, as a function of the
averaged in situ power, is shown in Fig. 13.
The noise will depend on a broad variety of parameters, like
the rotation speed of each motor, the quality of mechanics, iso-
lation materials, brushes, and so on. But as one can see in Fig.
13, the data can be partially modeled by the linear regression
LWA =0.64 ·probot +61.4 (R2=0.61) .
As was shown during the crack test, the dierence in power be-
tween the robots can be mainly explained by the suction power
produced by the cleaning system. A low-power robot will be
consequently easier to silence, making it more pleasant for the
user, but reducing the cleaning eciency of the robot on rough
surfaces. This is a great opportunity for an adaptive system,
which would be able to minimize the noise most of the time,
while still retaining its cleaning eciency on diverse types of
Robot’s Visibility. In contrast to a manual vacuum cleaner, a
robotic vacuum cleaner is visible to the household members
most of the time, even when the system is not in use. For this
reason, the visible design of the robot requires special attention.
Optimally, the design would fit the style of the home and likely
not resemble a cleaning tool so much as some kind of home ac-
cessory. One participant explained: “I need the vacuum cleaner
every single day to clean up around the table. Sure, it would be
easier to have it just right next to the table. But it would not look
nice. I do not want my vacuum cleaner or any other cleaning
tool to be that present. The importance of a good visible de-
sign becomes increasingly important when people are hosting
guests. Whereas five of the households proudly presented their
cleaning robot to guests, four of the households unplugged their
Roomba’s charging station and put it, together with the robot,
in a closet or under the bed to hide it, as they found it did not
fit with the rest of the house and looked rather ugly. It is there-
fore necessary to put together an optimal functional shape of
the robot with a good design to make the robot less intrusive,
permitting it to merge with its surroundings.
6. Conclusion
In this paper, we presented results from an evaluation of sev-
eral domestic robotic vacuum cleaners with respect to several
main topics: power consumption, navigation, cleaning perfor-
mance, and human-robot interaction. We did not only carry
out a formal comparison of seven dierent robots but we also
conducted a long-term user study with nine households. This
holistic approach was chosen to advance further personal ser-
vice robots and enhance their acceptance by merging together
both sides of the robot’s lifecycles: technical design and final
Our methodology is based on a number of assumptions. First
of all, some of the results from the user study are based on peo-
ple’s self-reported data. We carefully verified these data during
the interviews and our on-site observations to capture “reality”
as good as possible. However, these qualitative data remain
subjective. Therefore, the results can only be taken as being
reflective of individual cases but cannot be generalized.
Second, we only used random navigation robots for the user
study, whereas the technical study also included systematic nav-
igation robots. To investigate the impact of user’s perception of
the robot’s path planning, another type of robot with planned
navigation could have been used and complemented the results.
With our study in hands, we have strong evidences that this type
of robot would improve the long-term acceptance.
Finally, concerning the technical study, the environment was
simulated. Although modeled after a real scenario, it still lacked
some challenges encountered by robotic vacuum cleaners in
people’s homes. Our environment was static and constrained,
while a house is generally dynamic and unconstrained. The ro-
bustness of the navigation consequently remains untested.
Nevertheless, we were able to combine the outcomes of both
studies in a meaningful way. We identified concrete bene-
fits and drawbacks of mapping technologies for domestic floor
cleaning robots, on one hand regarding the amount of energy
consumed by the system and on the other hand taking into ac-
count users’ needs and perceptions. We found that the user’s
perception highly depends on how users actually use the robot,
and that this perception in turn influences the design require-
ments. People wish to understand how their robotic vacuum
cleaner is working (“transparency”), which is not the case with
random navigation. Currently, the robot does not provide ade-
quate feedback to the user, decreasing the chances of long-term
acceptance. This could be improved by providing the user with
a map of the environment, and fusing inside it information com-
ing from the sensors like dust, energy consumption, and type of
surface. The user could in turn give information to plan more
precisely the tasks.
Based on people’s dierent cleaning strategies and attitudes
towards cleanliness and robots, there is no single best solution.
Our evaluation shows that the manual vacuum cleaner is still
more ecient for cleaning. However, this comes with a much
higher energy consumption, whereas a robot is more energy ef-
ficient. When comparing dierent navigation strategies, the ad-
dition of a SLAM-based solution enables reduction of overall
energy consumption by speeding the completion time.
Our initial question — “can a robot be a drop-in replace-
ment to accomplish domestic tasks?” — cannot be answered
out of the box. It mostly depends on how the respective product
is used and thus always involves both the user and the system
and their shared environment. It has to be clear that as soon as
robotics tries to enter people’s homes, the human will be at the
center of it. Therefore, success first and foremost depends on
providing solutions that match real needs.
Participants in our study wished that the robotic vacuum
cleaner would solve the shortcomings of their manual vacuum
cleaner, and consequently decrease the amount of work for the
user. However, six of the nine households stopped using the
robot after a while. Although people were at first enthusiastic
and interested in trying out a robotic vacuum cleaner, the ma-
jority became disappointed as they actually assessed the robot’s
relevance within their own ecosystem. In this case, the rejection
of robotics is not motivated by some underlying fear or nega-
tive preconceptions, but is an issue of how functional the robot
is within people’s ecosystem. After novelty eects had worn
o, the robot became another cleaning tool with its own flaws.
We also noticed two big hurdles from the human side: a
lack of trust in the robot to do its work autonomously while
the homeowner is not around, and the willingness to adapt and
make physical alterations to the home itself for the robot. We
believe that domestic robots should be designed to minimize the
need for these types of eorts.
We can suggest several practical improvements aimed at in-
creasing the synergies between the user’s needs and the robot’s
capabilities. It principally comes to enhancing the perceived en-
ergy eciency, the navigation inside the user’s space, and the
feedback coming from the robot. These improvements lever-
age the available mapping technology, by pushing forward the
learning and reasoning algorithms, as they are currently at a
very rudimentary level.
The surfaces to be cleaned can be classified regarding their
roughness, opening doors for an adaptive suction power.
Advantages to this are a reduced power consumption on
flat surfaces, with a reduced noise power level. The suction
power can be increased when cleaning an uneven terrain
for increased eciency.
The dirty spots can be learned over several runs. This
enables adaptive planning to be put in place, where dirty
places (kitchen, living room) are cleaned with a dierent
frequency to meet the user’s needs. Energy is also spared,
as only the necessary amount of cleaning is performed.
The map enables the robot to clean by unit area, as de-
scribed by [9]. This way, certain parts of the surface can
be cleaned using a more refined schedule. The participants
in our study expressed this aspect either as a wish for a fu-
ture version of the robot, or instead as a shortcoming of
the current version.
The feedback to the user can be enhanced. On the map of
the house, several layers of information could be added:
what is the intended planning, what has already been
cleaned, where is the dirt, are there any unreachable points,
and so forth. This map can be provided using a rich inter-
face, for example through a smartphone application. This
feedback could also include estimates of time remaining
until completion. This will enhance user acceptance, espe-
cially for more demanding users, by creating transparency
and a feeling that the user “knows” what the robot is doing.
These improvements will be at the center of our future work,
in order to provide more helpful robots for daily life. More-
over, this study allowed us to identify key parameters to re-
duce the energy consumption of domestic robots. By leverag-
ing this knowledge, we are aiming at robots able to operate au-
tonomously indoors, without having to rely on the power grid.
This improvement goes towards the expectations of users and
the needs of our society.
This advance implies the necessity of embedding some kind
of energy harvesting into the mobile robot and/or on a charging
station, providing it with energy extracted directly from the sur-
rounding environment, as previously discussed in [48]. Light,
heat, or mechanical work produced by humans could act as the
primary source of energy. In any case, the available energy level
will be low and highly fluctuating over time, driving the need
to spare energy at the level of the complete system and to gain
more information concerning the surrounding environment.
This research was supported by the Swiss National Science
Foundation through the National Centre of Competence in Re-
search Robotics. Most of the robots have been provided by
Franc¸ois Jean-Richard from Swiss National Television (TSR).
We thank Daniel Burnier (EPFL–STI–LSRO) for the datalog-
ger used during the in situ analysis, as well as C´
edric Monchˆ
and Herv´
e Lissek (EPFL–STI–LEMA) for the acoustic mea-
[1] F. Vaussard, P. R´
etornaz, D. Hamel, F. Mondada, Cutting Down the
Energy Consumed by Domestic Robots: Insights from Robotic Vacuum
Cleaners., in: G. Herrmann, M. Studley, M. J. Pearson, A. T. Conn,
C. Melhuish, M. Witkowski, J.-H. Kim, P. Vadakkepat (Eds.), TAROS,
volume 7429 of Lecture Notes in Computer Science, Springer, 2012, pp.
[2] F. Kaplan, Everyday robotics: robots as everyday objects, in: Proceedings
of the 2005 joint conference on Smart objects and ambient intelligence:
innovative context-aware services: usages and technologies, sOc-EUSAI
’05, ACM, New York, NY, USA, 2005, p. 59–64.
[3] B. Gates, A robot in every home, Scientific American 296 (2007) 58–65.
[4] J. Sung, R. E. Grinter, H. I. Christensen, Domestic robot ecology, Inter-
national Journal of Social Robotics 2 (2010) 417–429.
[5] J. Forlizzi, How robotic products become social products: an ethno-
graphic study of cleaning in the home, in: Proceedings of the ACM/IEEE
international conference on Human-robot interaction, HRI ’07, ACM,
New York, NY, USA, 2007, p. 129–136.
[6] J. Forlizzi, C. DiSalvo, Service robots in the domestic environment:
a study of the roomba vacuum in the home, in: Proceedings of the
1st ACM SIGCHI/SIGART conference on Human-robot interaction, HRI
’06, ACM, New York, NY, USA, 2006, p. 258–265.
[7] T. Kanda, T. Hirano, D. Eaton, H. Ishiguro, Interactive robots as social
partners and peer tutors for children: a field trial, Hum.-Comput. Interact.
19 (2004) 61–84.
[8] T. Kanda, H. Ishiguro, Communication Robots for Elementary Schools,
Technical Report, CiteSeerX, 2005.
[9] H. Kim, H. Lee, S. Chung, C. Kim, User-centered approach to path plan-
ning of cleaning robots: analyzing user’s cleaning behavior, in: Proceed-
ings of the ACM/IEEE international conference on Human-robot interac-
tion, HRI ’07, ACM, New York, NY, USA, 2007, p. 373–380.
[10] iRobot CEO Discusses Q4 2010 Results,
results-earnings-call-transcript, 2010.
[11] International Federation of Robotics: World Robotics 2012 Service
Robots,, 2012.
[12] Y. Rogers, H. Sharp, J. Preece, Interaction Design: Beyond Human -
Computer Interaction, John Wiley & Sons, 2011.
[13] F. D. Davis, Perceived usefulness, perceived ease of use, and user accep-
tance of information technology, MIS Quarterly 13 (1989) 319–340.
[14] V. Venkatesh, H. Bala, Technology acceptance model 3 and a research
agenda on interventions, Decision Sciences 39 (2008) 273–315.
[15] J. Fink, V. Bauwens, F. Kaplan, P. Dillenbourg, Living with a vacuum-
cleaning robot. a 6-months ethnographic study., International Journal of
Social Robotics (to appear).
[16] F. Yasutomi, M. Yamada, K. Tsukamoto, Cleaning robot control, in:
Robotics and Automation, 1988. Proceedings., 1988 IEEE International
Conference on, IEEE, pp. 1839–1841.
[17] E. Prassler, A. Ritter, C. Schaeer, P. Fiorini, A Short History of Cleaning
Robots., Auton. Robots 9 (2000) 211–226.
[18] P. Fiorini, E. Prassler, Cleaning and Household Robots: A Technology
Survey., Auton. Robots 9 (2000) 227–235.
[19] E. Prassler, K. Kosuge, Domestic Robotics., in: B. Siciliano, O. Khatib
(Eds.), Springer Handbook of Robotics, Springer, 2008, pp. 1253–1281.
[20] I. Alonso, Service robotics, Service Robotics within the Digital Home
(2011) 89–114.
[21] T. Breuer, G. R. G. Macedo, R. Hartanto, N. Hochgeschwender, D. Holz,
F. Hegger, Z. Jin, C. M ¨
uller, J. Paulus, M. Reckhaus, J. A. ´
A. Ruiz, P.-G.
oger, G. K. Kraetzschmar, Johnny: An Autonomous Service Robot for
Domestic Environments., Journal of Intelligent and Robotic Systems 66
(2012) 245–272.
[22] I. Carrera, H. A. Moreno, R. J. Saltar ´
en, C. P´
erez, L. Puglisi, C. E. G.
Cena, ROAD: domestic assistant and rehabilitation robot., Med. Biol.
Engineering and Computing 49 (2011) 1201–1211.
[23] F. Yuan, L. Twardon, M. Hanheide, Dynamic path planning adopting hu-
man navigation strategies for a domestic mobile robot., in: IROS, IEEE,
2010, pp. 3275–3281.
[24] J. Leonard, Challenges for Autonomous Mobile Robots, in: Machine Vi-
sion and Image Processing Conference, 2007. IMVIP 2007. International,
p. 4.
[25] S. Thrun, Simultaneous Localization and Mapping, in: M. Jeeries, W.-
K. Yeap (Eds.), Robotics and Cognitive Approaches to Spatial Mapping,
volume 38 of Springer Tracts in Advanced Robotics, Springer Berlin Hei-
delberg, 2008, pp. 13–41.
[26] H. Durrant-Whyte, T. Bailey, Simultaneous localization and mapping:
part I, Robotics Automation Magazine, IEEE 13 (2006) 99–110.
[27] T. Palleja, M. Tresanchez, M. Teixido, J. Palac´
ın, Modeling floor-cleaning
coverage performances of some domestic mobile robots in a reduced sce-
nario., Robotics and Autonomous Systems 58 (2010) 37–45.
[28] A. De Almeida, J. Fong, Domestic Service Robots [TC Spotlight],
Robotics & Automation Magazine, IEEE 18 (2011) 18–20.
[29] M. Scopelliti, M. V. Giuliani, F. Fornara, Robots in a domestic setting: a
psychological approach, Universal Access in the Information Society 4
(2005) 146–155.
[30] C. Ray, F. Mondada, R. Siegwart, What do people expect from robots?,
in: Intelligent Robots and Systems, 2008. IROS 2008. IEEE/RSJ Interna-
tional Conference on, pp. 3816 –3821.
[31] J. E. Young, R. Hawkins, E. Sharlin, T. Igarashi, Toward acceptable do-
mestic robots: Applying insights from social psychology, International
Journal of Social Robotics 1 (2008) 95–108.
[32] K. Dautenhahn, S. Woods, C. Kaouri, M. L. Walters, K. L. Koay, I. Werry,
What is a robot companion - friend, assistant or butler?, in: 2005
IEEE/RSJ International Conference on Intelligent Robots and Systems,
2005. (IROS 2005), IEEE, 2005, pp. 1192– 1197.
[33] J.-Y. Sung, R. E. Grinter, H. I. Christensen, L. Guo, Housewives or
technophiles?: understanding domestic robot owners, in: Proceedings of
the 3rd ACM/IEEE international conference on Human robot interaction,
HRI ’08, ACM, New York, NY, USA, 2008, p. 129–136.
[34] J. Sung, H. I. Christensen, R. E. Grinter, Sketching the future: Assessing
user needs for domestic robots, in: The 18th IEEE International Sympo-
sium on Robot and Human Interactive Communication, 2009. RO-MAN
2009, IEEE, 2009, pp. 153–158.
[35] J. Sung, R. E. Grinter, H. I. Christensen, ”Pimp my roomba”: designing
for personalization, in: Proceedings of the 27th international conference
on Human factors in computing systems, CHI ’09, ACM, New York, NY,
USA, 2009, p. 193–196.
[36] J. Sung, Design guidelines for everyday robots,, 2012. Unpublished document.
[37] J. Sung, H. I. Christensen, R. E. Grinter, Robots in the wild: understand-
ing long-term use, in: Proceedings of the 4th ACM/IEEE international
conference on Human robot interaction, HRI ’09, ACM, New York, NY,
USA, 2009, p. 45–52.
[38] W. Jeong, K. Lee, CV-SLAM: A new ceiling vision-based SLAM tech-
nique, in: Intelligent Robots and Systems, 2005.(IROS 2005). 2005
IEEE/RSJ International Conference on, IEEE, pp. 3195–3200.
[39] K. Konolige, J. Augenbraun, N. Donaldson, C. Fiebig, P. Shah, A low-
cost laser distance sensor., in: ICRA, IEEE, 2008, pp. 3002–3008.
[40] H. Seki, K. Ishihara, S. Tadakuma, Novel Regenerative Braking Control
of Electric Power-Assisted Wheelchair for Safety Downhill Road Driv-
ing, Industrial Electronics, IEEE Transactions on 56 (2009) 1393–1400.
[41] B. G. Glaser, A. L. Strauss, The Discovery of Grounded Theory: Strate-
gies for Qualitative Research, Transaction Publishers, 1967.
[42] J. M. Corbin, A. Strauss, Grounded theory research: Procedures, canons,
and evaluative criteria, Qualitative Sociology 13 (1990) 3–21.
[43] T. S. Ha, J. H. Jung, S. Y. Oh, Method to analyze user behavior in home
environment, Personal and Ubiquitous Computing 10 (2005) 110–121.
[44] A. Sabelli, T. Kanda, N. Hagita, A conversational robot in an elderly care
center: An ethnographic study, in: 2011 6th ACM/IEEE International
Conference on Human-Robot Interaction (HRI), pp. 37 –44.
[45] Commission Regulation (EC) No 1275/2008, http://eur-
EN:PDF=rja, 2008.
[46] P. KaewTraKulPong, R. Bowden, An improved adaptive background mix-
ture model for real-time tracking with shadow detection, in: Proc. 2nd
European Workshop on Advanced Video Based Surveillance Systems,
volume 25, pp. 1–5.
[47] T. Kim, P. Hinds, Who should i blame? eects of autonomy and trans-
parency on attributions in human-robot interaction, in: The 15th IEEE
International Symposium on Robot and Human Interactive Communica-
tion, 2006. ROMAN 2006, IEEE, 2006, pp. 80–85.
[48] F. Vaussard, M. Bonani, P. R´
etornaz, A. Martinoli, F. Mondada, Towards
Autonomous Energy-Wise RObjects., in: R. Groß, L. Alboul, C. Mel-
huish, M. Witkowski, T. J. Prescott, J. Penders (Eds.), TAROS, volume
6856 of Lecture Notes in Computer Science, Springer, 2011, pp. 311–
... For example, Sung et al. conducted a sixmonth long-term field study on how 30 households accepted a new vacuuming robot and established a domestic robot ecology framework including four temporal stages: re-adoption, adoption, adaptation and use/retention [19]. Vaussard et al. identified the influence of key technologies by comparing seven robots and conducted an ethnographic study within nine households to uncover users' needs [18]. Their studies suggest two aspects of cleaning robot design, how well a robot accomplishes its task and how well it integrates inside the user's space and perception [18]. ...
... Vaussard et al. identified the influence of key technologies by comparing seven robots and conducted an ethnographic study within nine households to uncover users' needs [18]. Their studies suggest two aspects of cleaning robot design, how well a robot accomplishes its task and how well it integrates inside the user's space and perception [18]. Also, a lack of trust in the robot and the unwillingness to make a physical change at home hinder the acceptance of a cleaning robot [18]. ...
... Their studies suggest two aspects of cleaning robot design, how well a robot accomplishes its task and how well it integrates inside the user's space and perception [18]. Also, a lack of trust in the robot and the unwillingness to make a physical change at home hinder the acceptance of a cleaning robot [18]. ...
Full-text available
INTRODUCTION: In the interwoven trend of the experience economy and advanced information technology, user experience becomes the substantial value of an interactive system. As one of the early innovations of a smart home, the current design of household cleaning robots is still driven by technology with a focus on pragmatic quality rather than the experiential value of a robotic system.OBJECTIVES: This paper aims to uplift the design vision of a cleaning robot from an automatic household appliance towards a meaningful robotic system engaging users in happy housework.METHODS: Theoretically, experience design and scenario-based design methods were combined into a specific design framework for domestic cleaning robotic systems. Based on the user study and technology trend analysis, we first set three experience goals (immersion, trust, and inspiration) to drive the design process, then chose 3D point cloud and AI recognition as backup technologies and afterwards extracted three main design scenarios (scanning and mapping, intelligent cleaning, and live control).RESULTS: The design features multi-view switching, a combination of animation rendering and real scene, fixed-point cleaning, map management, lens control and flexible remote, and shooting modes are proposed. Seventy-one participants evaluated the concept with online AttrakDiff questionnaires. The results indicate the targeted experience is fulfilled in the design concept.CONCLUSION: By integrating experience design and scenario-based design methods with technology trend analysis, designers can envision experiential scenarios of meaningful life and potentially expand the design opportunity space of interactive systems.
... Personal service robots can assist humans and provide entertainment. Robotic devices for the home and yard, like vacuum cleaners and lawn mowers, are well-established market segments (Blackman, 2013;Sung et al., 2008;Vaussard et al., 2014). Robots in hotel lobbys were proposed as an experimental alternative to digital signage in a recent Japanese study (Pan et al., 2015). ...
... Industrial robots may have a significant impact on the creation of robot-friendly kitchens. Similar to how redesigning hotels and restaurants for efficient cleaning and customer service might be required by service robots (Vaussard et al., 2014). ...
The study from 1984 that pushed hospitality educators to incorporate robots into their classes and research is revisited, updated, and summarised in this conceptual research note. Organizations across the world are beginning to use robots, artificial intelligence, and service automation in a variety of ways. The tourism and hospitality sectors are embracing cutting-edge technology. Travel firms have the ideal chance to improve marketing, customer service, customer experience, and retention thanks to AI. Artificial intelligence (AI) has the potential to provide travel agencies and hoteliers with priceless insights they have never had before. This thesis contrasts the existing situation with the possibilities for the future through an update of service design techniques. It does this by outlining the application of RAISA in hotel SPD through a literature study. It is advised that there may be benefits and drawbacks to adopting robots in the hotel sector. Key words : Robotics, Hospitality, Tourism industry
... Examples of autonomous agents popular in recent times range from robotic vacuum cleaners (e.g. [30]) and self-driving cars (e.g. [2]) to robots making lives easier from behind the scenes, such as warehouse optimization robots (e.g. ...
There is a lack of standard benchmarks for Multi-Agent Reinforcement Learning (MARL) algorithms. The Starcraft Multi-Agent Challenge (SMAC) has been widely used in MARL research, but is built on top of a heavy, closed-source computer game, StarCraft II. Thus, SMAC is computationally expensive and requires knowledge and the use of proprietary tools specific to the game for any meaningful alteration or contribution to the environment. We introduce SMAClite -- a challenge based on SMAC that is both decoupled from Starcraft II and open-source, along with a framework which makes it possible to create new content for SMAClite without any special knowledge. We conduct experiments to show that SMAClite is equivalent to SMAC, by training MARL algorithms on SMAClite and reproducing SMAC results. We then show that SMAClite outperforms SMAC in both runtime speed and memory.
... The test area is a square of 4m 2 delimited by small walls. A time of 8 minutes is estimated as sufficient to clean this area (Vaussard et al., 2013;Palleja et al., 2010). In order to assess cleaning efficiency, a Java-based image processing program (Goldstein, 2011) is used to automatically calculate the percentage of clean area. ...
... For example, a floor cleaning robot could not access a narrow space resulting from the furniture arrangements of the workspace. The work [26], [27] found out that the area coverage performance of commercially available robots is limited in workspaces cluttered with obstacles. Reconfigurable robots and mechanisms have been introduced for floor cleaning to overcome the limitations of cleaning robots with fixed shapes [28], [29]. ...
Complex optimization problems are often associated to large search spaces and consequent prohibitive execution times in finding the optimal results. This is especially relevant when dealing with dynamic real problems, such as those in the field of power and energy systems. Solving this type of problems requires new models that are able to find near-optimal solutions in acceptable times, such as metaheuristic optimization algorithms. The performance of these algorithms is, however, hugely dependent on their correct tuning, including their configuration and parametrization. This is an arduous task, usually done through exhaustive experimentation. This paper contributes to overcome this challenge by proposing the application of sequential model algorithm configuration using Bayesian optimization with Gaussian process and Monte Carlo Markov Chain for the automatic configuration of a genetic algorithm. Results from the application of this model to an electricity market participation optimization problem show that the genetic algorithm automatic configuration enables identifying the ideal tuning of the model, reaching better results when compared to a manual configuration, in similar execution times.KeywordsAutomatic algorithm configurationElectricity marketsGenetic algorithmMetaheuristic optimizationPortfolio optimization
This paper describes a full navigation architecture which includes a set of available Deep Learning-based modules, focused on “speech to text” and “text to speech” translation, and face recognition, for enabling natural interaction between a smart mobile assistant robot and its human users, in a context of autonomous navigation. The system is novel because it allows complex spoken commands to be syntactically analyzed in Spanish and transformed into motion plans, ready to be executed by the robot, by using the well-known Navigation stack included in the ROS ecosystem. A novel computationally efficient approach (to semantically label the free space from raw data provided by a low cost laser scanner device), enables the generation of a labelled polygonal map, which enhances fixed and mobile obstacle avoidance and local planning in real indoor environments. Robot configuration, learned maps, and Deep Learning models adapted to each scenario are safely and privately stored in Google Cloud, allowing the robot to adapt its behavior to different users and settings. The tests which demonstrate the performance of the system, both in simulation and real environments, are also described.
Modern homes are moving to automation to provide convenience and reduce the time spent on household chores Vacuum cleaners make cleaning easier, but they are big, noisy, and bulky in everyday use.Therefore, vacuum cleaner technology needs to be improved to reduce these shortcomings. As of 2021, the Covid-19 pandemic continues, so keeping our living space clean and disinfecting is a top priority. People's workload is shifting from offline work in the office to online work at home, and the workload of many people has increased significantly, so some people may neglect to clean the environment. there is. In today's scenario, the whole family is busy with work and is supposed to clean the house. The cleaning robot helps you clean and disinfect the floor with the push of a switch. This also reduces the extra effort of housekeeping for cleaning purposes. Furthermore, we are enthusiastic about designing and developing an automatic cleaning robot "household automatic vacuum cleaner robot" that performs all cleaning and disinfection work with the push of a button.
Full-text available
Understanding the attachment owners can feel to their robot vacuums, which also map and collect data about their homes, is key to understanding the ambivalences involved in the integration of automated visualities in the home. Drawing on qualitative video interviews and observations of people interacting with their robot vacuums, this article identifies three key factors in understanding how cohabitation with a robot vacuum and its particular form of automated sensoria is experienced by its user: firstly, the robot assists with work that we would otherwise do ourselves with the aid of a broom or traditional vacuum cleaner; it is thus often regarded as an extension of ourselves, the equivalent of a cleaning assistant, or even a kind of pet with which you can interact. Secondly, its ability to move autonomously increases the inclination to anthropomorphize the robot as a being with some level of agency and intelligence. Thirdly, the robot vacuum cleaner is a very visible part of the intimate sphere. It has its charging station in the home; it cannot be hidden away in a cupboard like an ordinary vacuum cleaner; more often than not, furniture needs to be moved around for it to run smoothly. This article argues that these three factors are important for understanding people's difficulty in perceiving the robot as an entity that potentially participates in surveillance practices, and to understand the nature of this form of surveillance that emanates from the leaky home.
Conference Paper
Full-text available
The market of domestic service robots, and especially vacuum cleaners, has kept growing during the past decade. According to the International Federation of Robotics, more than one million units were sold worldwide in 2010. Currently, there is no in-depth analysis of the energetic impact of the introduction of this technology on the mass market. This topic is of prime importance in our energy-dependant society. This study aims at identifying key technologies leading to the reduction of the energy consumption of a domestic mobile robot, by exploring the design space using technologies issued from the robotic research field, such as the various localization and navigation strategies. This approach is validated through an in-depth analysis of seven vacuum-cleaning robots. These results are used to build a global assessment of the influential parameters. The major outcome is the assessment of the positive impact of both the ceiling-based visual localization and the laser-based localization approaches.
Conference Paper
Full-text available
It has long been recognized that novelty effects exist in the interaction with technologies. Despite this recognition, we still know little about the novelty effects associated with domestic robotic appliances and more importantly, what occurs after the novelty wears off. To address this gap, we undertook a longitudinal field study with 30 households to which we gave Roomba vacuuming robots and then observed use over six months. During this study, which spans over 149 home visits, we encountered methodological challenges in understanding households' usage patterns. In this paper we report on our longitudinal research, focusing particularly on the methods that we used 1) to understand human-robot interaction over time despite the constraints of privacy and temporality in the home, and 2) to uncover information when routines became less conscious to the participants themselves.
Full-text available
A study developed and validated new scales for perceived usefulness and perceived ease of use, which were hypothesized to be fundamental determinants of user acceptance. The definitions of the 2 variables were used to develop scale items that were pretested for content validity. The items were then tested for reliability and construct validity in 2 studies involving a total of 152 users and 4 application programs. After refining and streamlining the measures, the resulting 2 scales of 6 items each demonstrated reliabilities of .98 for usefulness and .94 for ease of use. The scales also exhibited high convergent, discriminant, and factorial validity. In both studies, usefulness had a greater correlation with usage behavior than did ease of use, though both were significantly correlated with current usage and future usage. Regression analyses suggest that perceived ease of use may actually be a casual antecedent to perceived usefulness, as opposed to a direct determinant of system usage.
Full-text available
Little is known about the usage, adoption process and long-term effects of domestic service robots in people’s homes. We investigated the usage, acceptance and process of adoption of a vacuum cleaning robot in nine households by means of a six month ethnographic study. Our major goals were to explore how the robot was used and integrated into daily practices, whether it was adopted in a durable way, and how it impacted its environment. We studied people’s perception of the robot and how it evolved over time, kept track of daily routines, the usage patterns of cleaning tools, and social activities related to the robot. We integrated our results in an existing framework for domestic robot adoption and outlined similarities and differences to it. Finally, we identified several factors that promote or hinder the process of adopting a domestic service robot and make suggestions to further improve human-robot interactions and the design of functional home robots toward long-term acceptance.
In this article, the RObject concept is first introduced. This is followed by a survey of applicable energy scavenging technologies. Energy is a key issue for the large scale deployment of robotics in daily life, as recharging the batteries places a considerable burden on the end-user and is a waste of energy which has an overall negative impact on the limited resources of our planet. We show how the energy obtained from light, water flow, and human work, could be promising sources of energy for powering low-duty devices. To assess the feasibility of powering future RObjects with technologies, tests were conducted on commonly available robotic vacuum cleaners. These tests established an upper-bound on the power requirements for RObjects. Finally, based on these results, the feasibility of powering RObjects using scavenged energy is discussed.
This article provides a comprehensive introduction into the simultaneous localization and mapping problem, better known in its abbreviated form as SLAM. SLAM addresses the problem of a robot navigating an unknown environment. While navigating the environment, the robot seeks to acquire a map thereof, and at the same time it wishes to localize itself using its map. The use of SLAM problems can be motivated in two different ways. One might be interested in detailed environment models, or one might seek to maintain an accurate sense of a mobile robot's location. SLAM servers both of these purposes. We review three major paradigms of algorithms from which a huge number of recently published methods are derived. First comes the traditional approach, which relies in the extended Kalman filter (EKF) for representing the robot's best estimate. The second paradigm draws its intuition from the fact that the SLAM problem can be viewed as a sparse graph of constraints, and it applies nonlinear optimization for recovering the map and the robot's locations. Finally, we survey the particle filter paradigm, which applies non-parametric density estimation and efficient factorization methods to the SLAM problem. This article discusses extensions of these basic methods. It elucidates variants of the SLAM problem and poses a taxonomy for the field. Relevant research is referenced extensively, and open research problems are discussed.