The e-puck, a Robot Designed for Education in Engineering

Francesco Mondada1 , Michael Bonani1 , Xavier Raemy2 , James Pugh2 , Christopher Cianci2 ,
Adam Klaptocz3 , Stéphane Magnenat1 , Jean-Christophe Zufferey3 , Dario Floreano3 , Alcherio Martinoli2

Abstract— Mobile robots have the potential to become the

ideal tool to teach a broad range of engineering disciplines.
Indeed, mobile robots are getting increasingly complex and
accessible. They embed elements from diverse fields such as
mechanics, digital electronics, automatic control, signal pro-
cessing, embedded programming, and energy management.
Moreover, they are attractive for students which increases
their motivation to learn. However, the requirements of an
effective education tool bring new constraints to robotics. This
article presents the e-puck robot design, which specifically
targets engineering education at university level. Thanks to its
particular design, the e-puck can be used in a large spectrum
of teaching activities, not strictly related to robotics. Through
a systematic evaluation by the students, we show that the e-
puck fits this purpose and is appreciated by 90 percent of a
large sample of students.

I. I NTRODUCTION
Fig. 1. The e-puck robot.
Mobile robots are both fascinating objects and the result
of the fusion of multiple competences. This fascination
leads to the organization of plenty of robotics contests The Khepera II from K-Team is a redesign of the
worldwide annually [16]. From an engineering point of original Khepera robot [15]. With the same size of the
view, the design and control of mobile robots requires skills original Khepera, it is compatible with its extensions,
in many disciplines such as mechanics, electronics, energy software, and scientific tools. The Khepera II is interesting
management, computer science, signal processing, and because its size allows to use it on a desktop. It is expensive
automatic control. The combination of these two aspects (around 1500 e for a basic configuration) but is known for
(fascination and inter-disciplinarity) makes mobile robots being reliable and well supported.
an excellent educational platform that enables students to The Hemisson from K-Team is a cheap platform (225 e)
address a broad range of engineering fields. with a diameter of 120 mm. It only provides a limited
This paper presents, for the first time, the design ap- computational power and few sensors in its basic configu-
proach resulting in the e-puck1 , an educational desktop mo- ration, but is extensible. It is a robust platform well suited
bile robot developed at the École Polytechnique Fédérale for beginners.
de Lausanne (EPFL) for a broad exploitation in teaching
The IdMind’s circular GT kit is a similar platform
activities. The main objectives of this development were:
slightly cheaper (210 e) with a diameter of 150 mm. It
• The use of a common platform in all EPFL courses has less standard extensions than Hemisson, but has more
related to mobile robotics, to replace the different I/O available ports to connect self-made extensions. It is
robots previously in use. more suited for experimentation using custom self-made
• The use of a mobile robot in non-robotic courses, extensions.
for instance signal processing, automatic control, and Even cheaper (175 e), the platform Bot’n Roll is
embedded programming, in order to propose more representative of a set of simple robots with few sensors
project-based exercises. that are excellent starting kits for beginners. This and
• The introduction of mobile robots earlier in the
the previous kit improve their accessibility by providing
curriculum, which implies the deployment in larger graphic programming environments.
classes (50 to 100 students). The Lego Mindstorms RCX was the first robotic plat-
II. E XISTING ROBOTS FOR EDUCATION form from the Lego company. The RCX is built around a
small 8 bit processor and can manage only 3 inputs and
A wide range of mobile robots are available on the 3 outputs (typically DC motors without encoder); but the
market. In this section we survey the subset of them that combination with the Lego bricks makes it a fantastic tool
we think to be relevant as educational platforms. to discover new robots shapes. The RCX has been replaced
in 2006 by the Lego Mindstorms NXT. This newer version
This work was supported by the École Polytechnique Fédérale de Lau-
sanne, Switzerland (EPFL, http://www.epfl.ch) in the framework is equipped with more advanced sensors, including color
of a FIFO project (Fond pour l’Innovation dans la Formation). All authors and sound, and can drive motors equipped with encoders.
are associated with EPFL, in the following laboratories: 1. Laboratoire de It is a clear reference in the field because of its good
Systèmes Robotiques, 2. Distributed Intelligent Systems and Algorithms
Laboratory, 3. Laboratory of Intelligent Systems. computational power, its flexibility and interesting price
1 e-puck: http://www.e-puck.org (260 e).
 The Palm Pilot Robot Kit (PPRK) is a commercially cleaning module, this platform provides low cost mobil-
available platform2 from Carnegie Mellon University com- ity (100 e). Its sensors, designed for vacuum cleaning
bining a mobile base and a personnal digital assistant tasks, offer a good support for reactive navigation. The
(PDA), originally a Palm Pilot. The PDA provides the limited internal processor, dedicated to the low-level robot
computational power and the user interface and controls the control, is programmable by simple scripts. Any advanced
sensors and the actuators through a serial connection with programming or supplementary I/O requires an additional
a PIC processor. The result is a compact omnidirectional main processor.
platform with three distance sensors (in its basic config-
uration for 250 e). Furthermore, the availability and the III. ROBOT DESIGN FOR ENGINEERING EDUCATION
maturity of Palm development tools makes this platform Most of the aforementioned products are exclusively
an interesting starting kit. either efficient mobile robots or good educational tools.
The Cye platform3 is a medium-size robot However, being both implies the following criteria:
(40×28×13 cm) equipped with special wheels that • Desktop size. A robot that can evolve on the desk
ensure a good odometry. Its price is around 540 e. near the computer improves drastically the student
Designed for indoor domestic environments, Cye can efficiency during experimentation. We consider that
carry extensions such as a vacuum cleaner and can for a good mobility, the experimentation space should
navigate in indoor environments. be 10 times the diameter of the robot. On a table, this
The Khepera III from K-Team is a research oriented implies a robot diameter smaller than 80 mm.
platform much larger than the Khepera II (120 mm of • Wide range of possibilities from an engineering and
diameter). It is adaptable to specific research requirements educational point of view. To exploit this tool in
through extensions, for instance the korebot board which various fields of education such as signal processing,
provides an XScale processor. Flexible, efficient, and pow- automatic control, embedded programming, or dis-
erful with respect to its size, this robot is also quite tributed intelligent systems design, the robot should
expensive as an educational robot (around 2000 e in basic provide a wide set of functionalities in its basic
configuration). version.
The ER1 from Evolution Robotics4 is a simple alu- • User friendly. The user interface has to be simple,
minum frame kit supporting a laptop (not included) and efficient, and intuitive. This is an important point for
equipped with wheels. The laptop provides the computa- the acceptance of the system by the students.
tional hardware, which improves the performance / cost ra- • Low cost. The broad introduction in engineering
tio of the kit (the basic configuration costs around 230 e). classes requires a large number of robots. Knowing
The motor wheels controller has some free inputs/outputs that the budget of many schools is constant or de-
but provides limited computational power. This low-cost creasing, this is only feasible by reducing the cost of
kit comes with a sophisticated but expensive software an individual robot.
environment for navigation and vision. • Open information. This robot has to be shared among
The KHR-1 from Kondo 5 has been the first humanoid professors, laboratories, schools and universities. An
robot with good mobility capabilities (17 DOF) for a price open source hardware/software development model is
under 1000 e. an effective way to achieve this goal.
The Pioneer 3 (P3) is the latest version of the Pioneer
None of the platforms available on the market is re-
robot by ActivMedia. It is a large (44×38×22 cm) solid
specting these criteria. Most robots are large and thus need
platform on which the user can install custom processors,
to operate on the floor. The smallest robots are either
sensors, and actuators. The AmigoBot of the same com-
expensive or have limited functionalities. Very few are
pany is a cheaper version (1550 e) of the same concept.
open source.
The Garcia from Acroname6 is a small robot frame
This motivated us to create the e-puck robot in summer
(25×18×10 cm) designed to be controlled by a companion
2004. We tested a first set of prototypes with students
XScale board. The size of Garcia makes it suitable for
during the 2004–2005 academic year. Based on this ex-
experiments in compact environments. Its price is around
perience, we redesigned the robot and produced the final
1360 e in the basic configuration.
version in summer 2005 (Fig. 1). The following sections
Robotino from Festo7 is a modern mobile robotic plat-
present the robot design and a student evaluation based on
form for education. Robotino runs a real-time Linux kernel
its use during four semesters from 2005 to 2007.
and is built around industrial standards. These features
make this robot powerful in term of computational power IV. T HE E - PUCK DESKTOP MOBILE ROBOT
but also expensive (about 4500 e). Robotino is well
We based the design of the robot on the first two
suited for technical schools that want to approach technical
aforementioned criteria: desktop size and flexibility. Com-
problems using robotics.
bined, these two constraints imply the miniaturization of
Roomba Create from iRobot is an educational/research
a complex system. To achieve a low price, we opted
version of the roomba vacuum cleaner. Devoided of the
for the use of cheap components and mass production
2 PPRK : http://www.cs.cmu.edu/˜pprk/ manufacturing techniques. We took special care to make
3 Cye: http://www.personalrobots.com
4 Evolution Robotics: http://www.evolution.com
the robot as user-friendly and interactive as possible, in
5 Kondo: http://www.kondo-robot.com order to ensure that it would be well received by students.
6 Acroname: http://www.acroname.com In this design process, a central aspect is the choice
7 Festo: www.festo-didactic.com of the robot features. This particular choice is one of the
innovations of the e-puck design. The sensors, actuators, low battery detection 3.3V regul.
and interfaces of the e-puck are representatives of a wide to all
power- low battery devices
range of devices one can find in several engineering sub- on led indicator
2x stepper
domains: motors
extension connectors
• sensors in different modalities: audio, visual, dis-
body light
tances to objects, gravity, programming/
debug connector
• input devices with different bandwidths from 10 Hz
dsPIC 30F6014A 8x IR prox.
to 10 MHz (Figure 2, left), 16 Mips
RS232
• actuators with different actions on the environment connector 8kB RAM, 144kB FLASH
(Figure 2, right), 8x Red Leds
IR remote
• wired and wireless communication devices, control
Front Led
• two types of processors: general purpose and DSP.
bluetooth 14MHz
By exploiting this large set of possibilities, a teacher radio link clock CMOS
color
can present and the student can practice a broad set of RESET camera
engineering fields and skills. ON - 1.8V
OFF running regul.
mode 3D
selector accelero-
INPUTS PROCESSING OUTPUTS meter

5Wh Battery 3x micro

battery audio
10-100Hz displacement protection swappable and codec
IR proximity Generic, C prog. Motors rechargeable speaker
0.1-1kHz display
Accelerometer PIC core Lateral LEDs
10-40kHz audio Fig. 3. The outline of the electronic of the e-puck.
Microphones Speaker
Specific, ASM prog.
10k-10MHz directed
Camera Front LED
DSP core
1kHz serial aesthetic
IR remote
code
Body LED 2) Sensors and actuators: To ensure a broad range of
experimentation possibilities, the e-puck contains various
radio cable
sensors covering different modalities:
COMMUNICATION
• Eight infrared ( IR ) proximity sensors placed around

Bluetooth RS232
the body measure the closeness of obstacles or the
intensity of the ambient infrared light. These are
Fig. 2. The e-puck allows exploration of different engineering topics, typical sensors for simple navigation in cluttered
control options, and signal bandwidths. environments.
• A 3D accelerometer provides the acceleration vector
For applications where the basic features are not suffi- of the e-puck. This vector can be used to measure
cient, the e-puck can be extended with specific hardware. the inclination of the e-puck and the acceleration
We provide embedded software consisting of a library produced by its own movement. It can also detect
and several demo applications. We also provide an open collisions and if the robot falls. This sensor is rarely
source simulator and a monitoring tool to run on a desktop included in miniature low-cost mobile robots. We
computer. In this section we present the detailed hardware decided to include it because it allows a rich set of
and software design choices. experiments.
• Three microphones capture sound. Multiple micro-

A. e-puck hardware (basic configuration) phones allow the e-puck to localize the source of
the sound by triangulation. The bandwidth of this
1) Microcontroller: The electronic structure of the e- signal is much larger than the one of the accelerometer
puck (Fig. 3) is built around a Microchip dsPIC mi- or of the infrared sensors, making the microphones,
crocontroller. This microcontroller complies with the ed- because of their larger computational demands, the
ucational criteria of flexibility because it embeds both ideal tools to learn how to use the DSP unit.
a 16 bit processor with a 16 entry register file and a • A color CMOS camera with a resolution of 640×480
digital signal processor (DSP) unit. This CPU runs at 64 pixels in front of the e-puck enables experimentation
MHz and provides 16 MIPS of peak processing power. in vision. Only a sub-part of the image can be
The instruction set is mostly orthogonal8 and rich; in grabbed: the size of acquisition is limited by the
particular, it contains multiply-acccumulate and hardware- memory size of the dsPIC and the rate is limited
repeat instructions suitable to drive the DSP unit, for by its processing power. Any format of sub-image
instance to efficiently compute scalar products and fast is acceptable, providing these two constraints are
fourier transforms. Finally, this processor is supported by fulfilled. For instance, the e-puck can grab a color
a custom tailored version of the GCC C compiler. For the image of 40×40 pixels at 4 frames per second; the
e-puck, we chose a microcontroller version with 8 kB of frame rate is doubled in gray-scale. This limitation
RAM and 144 kB of flash memory. shows to the students the impact of high bandwidth
sensors such as cameras.
8 An instruction set is orthogonal if any instruction can use data of
any type via any addressing mode (http://en.wikipedia.org/
The e-puck provides the following actuators:
wiki/Orthogonal_instruction_set) • Two stepper motors. They control the movement of
 the wheels with a resolution of 1000 steps per wheel
revolution.
• A speaker, connected to an audio codec. Combined
with the microphones, the speaker can create a com-
light diffusion
munication network with the ability to detect the ring
direction of peers. It is also an excellent output device
for human interaction.
• Eight red light emitting diodes (LED) placed all default top
around the e-puck. These LEDs are covered by a extension turret
translucent plastic and the e-puck can modulate their
extension
intensities. They provide a visual interface with the main CPU board connectors
user; furthermore, another e-puck can observe them
with its camera which allows mutual visual interac-
left wheel
tions.
• A set of green LEDs placed in the transparent body. slot for bottom
By lighting the body, they improve the interactions extension
with the user.
motors
• A red front LED placed beside the camera. This LED
generates a focused beam that projects a red spot
on objects in front of the e-puck. Combined with
the camera, this spot allows distant measurements at main mechanical
longer range than the infrared proximity sensors. body
right wheel
3) User interface: The e-puck also contains several
devices to interact with the user and to communicate with LiIon battery
other equipments:
• Two LEDs show the status of the battery: One indi- Fig. 4. The mechanical structure of the e-puck in an exploded view.
cates whether the robot is powered on, while the other
indicates a low battery condition.
• A connector to interface to an in-circuit debugger, to B. e-puck extensions
program the flash memory and to debug code. To address the needs of specific teaching scenarios that
• An infrared remote control receiver, to control the e- require different mechatronics, the user can connect phys-
puck with standard television remote controls. ical extensions to provide additional sensors, actuators,
• A classic RS232 serial interface to communicate with or computational power. Extensions physically connect
a desktop computer. through an extension bus which routes a two wires inter-
• A Bluetooth radio link to connect to a desktop com- processor communication bus (I2 C) as well as the connec-
puter or to communicate with up to 7 other e-pucks. tions to most sensors.
• A reset button. There are three physically different types of extensions:
• A 16 positions rotary switch to specify a 4 bit number, “top”, “bottom”, and “sandwich”.
which can be used, for instance, to select among pre- Top extensions do not allow other extensions above
programmed behaviors or parameters. them. The basic e-puck includes a default extension board
4) Mechanics: The robot has a diameter of 75 mm of this type which provides the speaker, the 16 positions
and a height which depends on the connected extensions. rotary switch, the infrared remote control receiver, the
The mechanical structure of the e-puck consists of injected RS232 connector, and a reset button (Fig. 1, right). There
plastic parts. We have chosen this manufacturing technique are several other extensions of this type available, for
because it reduces the unit price of the robot for sufficient instance:
quantities. The robot structure is simple, being made of • A rotating scanner. It is equipped with infrared trian-

only four injected plastic parts: the main body, the light gulation distance sensors with a range of 40 cm (Fig.
ring, and the two wheels (Fig. 4). The main body is the 5, left). For exercises involving robot localization,
core of the mechanical structure and encloses the battery. the short range (2–3 cm) of the proximity sensors
The user can extract the battery from the bottom of the e- available on the basic version of the e-puck is not
puck. The two motors are simply laterally screwed onto the sufficient.
main body, with the wheels directly attached to the motor • A turret with three linear cameras. It provides a very

axis. The main printed circuit board (PCB), containing most large field of view for measuring optical flow [8], [21]
of the electronics, is screwed on top of the main body. (Fig. 5, right) still requiring few memory.
A light diffusion ring and a default extension board are Bottom extensions are internal PCBs, enclosed in a
mounted over this main PCB; the user can replace the vertical slot in the front part of the main body. These
default extension board with application specific boards, extensions are close to the ground and are connected to
as illustrated by some examples in the next section. All the main PCB by an I2 C bus and a power connection. For
mechanical parts are transparent and allow to observe all instance, an extension of this type provides three analog
components. sensors which measure the ground color (Fig. 6, left top).
 1) Bootloader: To program and debug a microcon-
troller, one typically uses an in-circuit debugger, which
is a specific piece of hardware. To reduce cost and remove
this requirement, the e-puck comes with a bootloader that
allows the user to re-program the flash of the microcon-
troller through the Bluetooth or the serial port. At boot,
the bootloader listens a small amount of time for activity
on these ports, and if none is detected, launches the user
application. During this time, the user has the opportunity
to send a special command in order to flash a new program.
Fig. 5. Two “top” extensions for the e-puck robot: an infrared distance 2) Low-level library: To facilitate the use of the e-
scanner (left) and a large field of view linear camera (right).
puck hardware, which requires specific code with precise
timings, we provide a collection of functions called the
“Low-Level Library”. This library contains functions such
as “move the right motor at that speed”, “read 40×40 pixel
image”, or “send this message through Bluetooth”. It is
statically linked with the user application at compile time.
3) BTcom protocol: When developing for a robot, it
is often useful to be able to control it from a desktop
computer. To that end, the e-puck comes with a moni-
tor implementing a remote control protocol through the
Bluetooth or the serial port, called the “BTcom protocol”.
This protocol provides full remote control of the e-puck,
allowing the desktop computer to set the speed of the
motors, read the image of the camera, specify the state of
the LEDs, read the accelerometer, and so on. This allows
the user to develop applications on a desktop computer,
in a comfortable environment with a rich set of develop-
ment tools. Moreover, with this approach, applications can
exceed the computational capabilities of the dsPIC. The
same strategy can be used to control the e-puck from a
simulator, such as Enki or Webots.
Fig. 6. Left top: A bottom extension to measure the ground color. Left
bottom: a “sandwich” extension implementing a Zigbee radio link. Right:
A complete visual communication system consisting of two “sandwich”
extensions. D. e-puck simulation

Several simulators support the e-puck. Among them,

Extensions can also be “sandwich” boards that replicate we use Webots and Enki. Webots [13] is commercial
the connector on their top, for instance: and supports three-dimensional physics through the ODE10
library. Enki11 is open source and provides fast 2D physics,
• a Zigbee radio link with adjustable radio communica- which, for instance, makes it suitable for evolutionary
tion range does not require mechanical access to the robotics experiments [7].
top of the e-puck [4] (Fig. 6, left bottom);
• a color RGB LED ring (Fig. 6 right);
• an omnidirectional vision turret with onboard micro- E. Availability
controller dedicated to vision (Fig. 6 right)
All our software, design, and production documents are
Several extensions of this type can be stacked, and a top
available under an open source hardware/software license,
turret placed above (Fig. 6 right).
meaning that anyone can use and modify them as long as
they comply with the license.
C. e-puck embedded software Currently, two companies (GCtronic and AAI Japan)
To develop software for the e-puck, we provide stan- produce the e-puck and about ten companies distribute it
dard components: a bootloader to program the e-puck under the term of the open hardware license. A growing
over Bluetooth, a low-level library to drive the hardware, community is using the e-puck both for research and
and a monitor to communicate with a desktop computer. education (see for instance [1], [20], [18], [12], [11], [5],
Within the community of e-puck users, standard software [2], [10], [14], [9], [17]).
modules are also developed. For instance a Player server is The production price of the e-puck basic version is
developed by VeRLab9 . These components are all released around 250 e. The selling price by most companies is
under the e-puck open source license. around 550 e.

9 Player 10 Open Dynamics Engine: http://www.ode.org

driver for the e-puck robot: http://code.google.com/
p/epuck-player-driver/ 11 Enki: http://home.gna.org/enki/
 V. E NGINEERING EDUCATION USING THE E - PUCK their awareness of the differences between various
ROBOT AT EPFL types of implementation levels; for example realistic
At EPFL, several courses exploit the e-puck robot as ex- simulations and real experiments.
perimentation platform. In particular we organize practical • Position estimation and path finding of a mobile robot.
exercices in the following teaching areas: In a 10 hours practical session we use the e-puck
to explore robot localization and path finding. These
• Signal processing. We use the e-puck to explore signal
techniques are active fields of research and begin
processing on sound, which is an optimal context for to see deployment in industry [6]. They are usually
education because of the signal low frequency, low implemented using expensive sensors such as laser
propagation speed, ease of generation, acquisition, range finder [19], which, because of limited education
and direct perception by humans. In our embedded budgets, often restricts their teaching to simulation.
programming course we explore signal processing ap- Yet simulation is not always sufficient to fully un-
plications during two sessions of five hours each. We derstand what is critical for correct functionality on
analyze the sound using a fast fourier transform (FFT), physical robots. The e-puck, with the distance scanner
with a close attention to its efficient implementation extension (see Section IV-B and Fig. 5, left), allows
on the DSP unit. In each session, we first present students to explore these techniques in a physical
the theory; then the students verify it on MATLAB setup.
with data acquired from the robot, and finally they
implement it on the dsPIC of the e-puck. VI. E VALUATION OF E - PUCK BY THE STUDENTS
• Automatic control. In our embedded programming
Our embedded programming course using the e-puck
course, we apply automatic control concepts on the
has been evaluated on a regular basis since the introduction
e-puck. The goal of the five hours session is to build
of the e-puck in 2005. The goal of this course is the
a regulator that controls the e-puck and maintains a
understanding and practice of embedded programming.
constant distance with respect to a visual target (Fig.
Most of the course focuses on C programming, with links
7, left). The input of a proportional-integral regulator
to assembler and some concepts of C++. Practicals explore
is the estimated distance to the target, extracted from
signal processing and control problems. The students who
the camera looking at a vertical black stripe on a white
attend this course are not specialized in robotics. All the
background (Fig. 7, right). The output of the regulator
programming exercises of this course are performed using
is assigned to the speed of both wheels, reducing the
the e-puck as a motivation tool.
problem to one degree of freedom.
Over the first three years of e-puck use, students were
asked to give feedback about the use of the e-puck to
illustrate the concepts of the course. Fig. 8 summarizes
the results of this analysis and shows that more than 90
percent of the students agree that the e-puck is a good
tool to illustrate the concepts of the course. The score has
slightly improved over the years, with the fine-tuning of
the course.
Fig. 7. Model of the problem (left) and scheme of the regulator (right)
for the e-puck automatic control practical.
The e-puck robot is a good tool to illustrate the concepts of the course
June 2007, 48 students,
on ASM, C and C++
• Behaviour-based robotics. In a robotics course, we programming

use the e-puck to provide the students with a better June 2006, 79 students,
on C++ programming
understanding of the subsumption architecture [3], February 2006, 90 students,
its advantages and its shortcomings. The goal of the on ASM, C programming
February 2005, 56 students,
exercise is to implement a behavior-based controller on ASM, C programming
for a simplified industrial application consisting of the totally agree partially agree no opinion partially disagree totally disagree
transport of goods throughout a warehouse. The e-
puck (the carrier) has to follow a line painted on the Fig. 8. Comparison between the evaluations of the e-puck robot as
ground (the track), while avoiding obstacles and going teaching tool.
back to the line after disruption. To that end, the e-
puck is equipped with the ground sensor extension When asked about the quality of the e-puck, students
(see Section IV-B and Fig. 6, left top). gave the feedback shown in Fig. 9. These results show that
• Distributed intelligent systems. Our course on dis- the latest version of e-puck is considered as performing
tributed intelligent systems (formerly focusing on well by more than 90 percent of the students. We observe
swarm intelligence) includes weekly laboratory ex- a clear improvement since the first version evaluated in
ercises in which the students use a combination of February 2005.
real e-puck robots and realistic simulations using
Webots [13] to test and verify the topics and the- VII. C ONCLUSION
ories presented in lectures. This help the students The e-puck robot is an innovative education tool for
to assimilate theoretical concepts around multi-robot engineering in a broad sense. For its size and price, it is a
coordination and networking, and to understand the complex systems that we exploit to teach a wide range of
difficulties of implementing them. This also enhances topics. By integrating the latest technologies in a compact
 The e-puck robot is performing well [12] Mattias Jacobsson, Sara Ljungblad, Johan Bodin, Jeffrey Knurek,
June 2007, and Lars Erik Holmquist. Glowbots: robots that evolve rela-
48 students, on V2 tionships. In SIGGRAPH ’07: ACM SIGGRAPH 2007 emerging
June 2006, technologies, page 7, New York, NY, USA, 2007. ACM.
79 students, on V2
[13] Olivier Michel. Webots: Professional Mobile Robot Simulation.
February 2006, International Journal of Advanced Robotic Systems, 1(1):39–42,
90 students, on V2
2004.
February 2005, [14] Yasser F. O. Mohammad and Toyoaki Nishida. TalkBack: Feedback
56 students, on V1
from a Miniature Robot, pages 357–366. Springer, 2007.
totally agree partially agree no opinion partially disagree totally disagree [15] F. Mondada, E. Franzi, and P. Ienne. Mobile robot miniaturization:
A tool for investigation in control algorithms. In T. Yoshikawa
Fig. 9. Comparison between the evaluation of the quality of the e-puck and F. Miyazaki, editors, Proceedings of the Third International
robot and its on-board software, following its evolution over the years Symposium on Simulation on Experimental Robotics (ISER-93),
and versions. volume 200 of Lecture Notes in Control and Information Sciences,
pages 501–513. Springer, 1993.
[16] R.R. Murphy. “competing” for a robotics education. Robotics and
Automation Magazine, IEEE, 8(2):44–55, Jun 2001.
design, it repositions robotics in the engineering education [17] F. Rastegar and M.N. Ahmadabadi. Grounding abstraction in sen-
landscape. More than 200 students over 3 years have sory experience. Advanced intelligent mechatronics, 2007 ieee/asme
international conference on, pages 1–8, 2007.
validated this concept and shown their satisfaction using [18] P. Roduit, A. Martinoli, and J. Jacot. A quantitative method
the e-puck. The open-source nature of this robot improves for comparing trajectories of mobile robots using point distribu-
the quality of the support to the students by providing tion models. Intelligent Robots and Systems, 2007. IROS 2007.
IEEE/RSJ International Conference on, pages 2441–2448, 2007.
full access to knowledge at every level. For teachers this [19] Roland Siegwart and Illah R. Nourbakhsh. Introduction to Au-
simplifies maintenance and opens new experimentation tonomous Mobile Robots. MIT Press, 2004.
possibilities. Finally, the rapid diffusion of the e-puck in [20] Vlad M. Trifa, Christopher M. Cianci, and Dominique Guinard.
Dynamic Control of a Robotic Swarm using a Service-Oriented
the research community shows its versatility as a scientific Architecture. In 13th International Symposium on Artificial Life
experimentation tool. and Robotics (AROB 2008), 2008.
[21] Jean-Christophe Zufferey and Dario Floreano. Toward 30-gram
Autonomous Indoor Aircraft: Vision-based Obstacle Avoidance and
VIII. ACKNOWLEDGMENT Altitude Control. In IEEE International Conference on Robotics and
We thank EPFL for funding the development of the e- Automation (ICRA’2005), pages 2594– 2599, 2005.
puck robot through its Funding Program for Teaching and
Learning. We also thank the Pedagogical Research and
Support team of EPFL for the detailed evaluation of the
e-puck educational performances. We finally thank our
students for their constructive feedback.

R EFERENCES
[1] Alberto Acerbi, Davide Marocco, and Stefano Nolfi. Social facil-
itation on the development of foraging behaviors in a population
of autonomous robots. Advances in Artificial Life, 4648:625–634,
2007.
[2] Saeed Amizadeh, Majid Nili Ahmadabadi, Babak N. Araabi, and
Roland Siegwart. A bayesian approach to conceptualization using
reinforcement learning. Advanced intelligent mechatronics, 2007
ieee/asme international conference on, pages 1–7, 2007.
[3] Rodney A. Brooks. Cambrian Intelligence. MIT Press, 1999.
[4] Christopher M. Cianci, Xavier Raemy, Jim Pugh, and Alcherio
Martinoli. Communication in a Swarm of Miniature Robots: The e-
Puck as an Educational Tool for Swarm Robotics. In Simulation of
Adaptive Behavior (SAB-2006), Swarm Robotics Workshop, pages
103–115, 2006, Lecture Notes in Computer Science (2007), vol.
4433.
[5] Roozbeh Daneshvar, Abdolhossein Sadeghi Marascht, Hossein Am-
inaiee, and Caro Lucas. A Quantitative Investigation into Distribu-
tion of Memory and Learning in Multi Agent Systems with Implicit
Communications, volume 4850, pages 124–133. Springer, 2007.
[6] H. Durrant-Whyte and T. Bailey. Simultaneous localization and
mapping: part i. Robotics and Automation Magazine, IEEE,
13(2):99–110, June 2006.
[7] Dario Floreano, Sara Mitri, Stéphane Magnenat, and Laurent Keller.
Evolutionary Conditions for the Emergence of Communication in
Robots. Current Biology, 17:514–519, 2007.
[8] N. Franceschini, J. M. Pichon, C. Blanes, and J. M. Brady. From
insect vision to robot vision [and discussion]. Philosophical
Transactions: Biological Sciences, 337(1281):283–294, 1992.
[9] O. Gigliotta and S. Nolfi. Formation of spatial representations in
evolving autonomous robots. Artificial Life, 2007. ALIFE ’07. IEEE
Symposium on, pages 171–178, 2007.
[10] Babak N. Araabi Hadi Firouzi, Majid Nili Ahmadabadi. A
probabilistic reinforcement-based approach to conceptualization.
International Journal of Intelligent Technology, 3:48–55, 2008.
[11] Daisuke Kurabayashi Herianto, Toshiki Sakakibara. Artificial
pheromone system using rfid for navigation of autonomous robots.
Journal of Bionic Engineering, 4:245–253, 2007.