Seamless Multi-Robot Programming for the People:

A SEBA and the Wireless Thymio II Robot

Philippe Rétornaz∗ , Fanny Riedo∗ , Stéphane Magnenat† , Florian Vaussard∗ , Michael Bonani‡ , Francesco Mondada∗

∗ École
Polytechnique Fédérale de Lausanne † ETHZurich ‡ AssociationMobsya
Laboratoire de Systèmes Robotiques Autonomous Systems Lab Email: [email protected]
Email: [email protected] Email: [email protected]

Abstract—Robots are an ideal tool for introducing program-

ming to young generations. To be accessible to a large public,
educational robots must be affordable and easy to use. In a previ-
ous work, the authors have developed Thymio II, an educational
robot costing about 100 $. Thymio II is programmable using the
ASEBA framework, which provides an interactive development
experience through real-time compilation and inspection of the
internal variables of the robot. However, this solution currently
requires a USB cable connection between the robot and a
computer, impairing the robot’s mobility.
This paper presents a radio-based wireless interface, allowing
to program the Thymio II robot without the hassle of wires.
This solution is transparent to the user, and implements the
ASEBA protocol in a backward-compatible way. It is built on
top of IEEE 802.15.4, costs a fraction of the robot’s price, and
does not affect its battery life significantly. After discussing
the challenges and presenting the design of the interface, this
paper shows performance results assessing the suitability of this Fig. 1: The Thymio II robot (110x112x54mm).
interface for educational use.
The presented solution opens new perspectives for the use of
robotics in schools from the first graders to the universities.
behaviours through visual and text programming. Thymio II is
I. I NTRODUCTION an open-source/hardware project1 . One unit costs about 100 $
Robots are an ideal tool for introducing programming to and 2500 have been sold so far. This robot was successfully
young generations, because they are both objects of fascination used with children of different age groups to introduce robotics
and machines with a rich set of interaction possibilities. In the and programming. This success is strongly linked to the low
past few years, the Laboratory of Robotics Systems of EPFL, cost of the robot, which was achieved with some limitations
in collaboration with the École Cantonale d’Arts de Lausanne in its features. In its sold configuration, a Thymio II robot
(écal), developed the Thymio II robot (Fig. 1), a small and has no means of communication with other robots, and must
low-cost educational mobile robot [1], [2]. To be accessible always be connected to a computer to be programmed or
to a large public, Thymio II was designed to be affordable debugged. From a pedagogical point of view, it would be
while still providing many features including the ability to be beneficial that users have a connection with the robot while it
programmed by its users. Thymio II integrates a wide range is moving around in a test environment, without having any
of sensing capabilities (infrared distance and ground sensors, cable connected. Indeed, being able to see and understand
a three-axis accelerometer, a microphone, touch buttons, etc.) what the robot perceives while moving, and debugging its
and several actuators (two motors, about 40 LEDs, a speaker, behaviour directly in the problematic conditions, is a key
etc.) allowing many different possible behaviours. These point in the experimental learning process. In addition, this
range from pre-programmed behaviours directly usable with could open new perspectives of multi-robot experiments, for
the robot alone, such as obstacle avoidance, to user-defined instance in collaborative setups.
Thymio II is programmed via the ASEBA Studio integrated
This research was supported by the Swiss National Science Foundation
through the National Centre of Competence in Research Robotics. 1 http://www.thymio.org
development environment (IDE), which is part of the open- firmware, they improve the adaptability and, as shown by
source ASEBA framework2 . A SEBA consists of an event-based Levis et al. [8], the general performances of the network.
virtual machine running on microcontrollers, an IDE with an Some authors demonstrated that using a virtual machine
embedded debugger, real-time compilation and visualization can reduce the local processing performances because of the
of the internal variables, and an easy-to-use scripting language. bytecode interpretation overhead [9]. However, this can be
Compared to alternatives (such as Arduino [3] or ROS [4]), avoided by using a mechanism to access native code directly
ASEBA allows the development flexibility of a virtual machine from inside the virtual machine for the computationally-
under the cost and energy-consumption constraints of a intensive algorithms without affecting the adaptability of the
microcontroller [5]. This brings programming to inexpensive virtual machine. Because a virtual machine can provide highly-
robots, which is of paramount importance in an educational optimized primitives for processing-intensive operations, it
context. As the ASEBA framework aims at safe operations can even increase the performances over naive C code in
of microcontroller-based research and educational robots microcontrollers [5]. An additional strong point of virtual
(software should not be able to harm the robot), it has a strict machine is their debugger support. As the bytecode is inter-
protocol. Furthermore, being already deployed on a number preted, it can easily be instrumented and remotely monitored
of target platforms, this protocol can only be updated in a to ease the debugging process. This is a decisive point when
backward-compatible way. Finally, because of its optimised inexperienced people are programming the device.
and tightly-integrated design, Thymio II has constraints of
its own. In this paper, we show how we took these into B. Declarative programming
account and developed a radio-frequency (RF) solution that Some sensor networks have a declarative programming
is flexible, affordable and compatible with existing ASEBA- (SQL-like) paradigm [10]. This enables the programming of
enabled robots. the sensor network as a unique entity and distributes the
A SEBA provides a solution for the programming and processing directly to the nodes. This is however best suited
debugging of a heterogeneous network of microcontrollers. for sensor acquisition and is quite limited for actuation. The
In particular, it distributes processing locally inside each main limitation with such an architecture is that one cannot
node, permits their dynamic on-the-fly re-programming, and easily program a specific behaviour on each node. Such a
provides a global view through Studio, its IDE. A SEBA network is designed to be programmed as a single instance
supports a wide range of physical transport protocols such of distributed sensors.
as CAN, UART, USB and Bluetooth. This paper presents the This is thus not entirely fitting our requirements, as each
porting of the ASEBA architecture to a wireless RF transport Thymio II robot within a network might demand a different
protocol. This is an non-trivial endeavour, because the ASEBA individual behaviour. Moreover, debuging systems using this
protocol currently makes assumptions, such as no loss of data, programming model is difficult for the inexperienced user,
that do not hold in wireless networks. This paper presents a because it is declarative rather than imperative [11], and does
robust and transparent implementation of the ASEBA protocol not support common debugging tools such as breakpoints.
over an RF network.
C. Native code generation
II. R ELATED W ORK
Some sensors network architectures focus on code re-
The main features of ASEBA are its ability to re-program usability. They are mainly template-based and output node-
nodes dynamically, its support of heterogeneous node types, specific code which is then compiled into native code [12].
and its real-time event-based programming paradigm. There- This model does not imply that the firmware in each node
fore, its wireless version lies in the field of sensor networks is fixed and not remotely updatable. Some dynamic linking
that can be dynamically re-programmed or re-configured. can take place in order to rewrite only some part of the
Hence, in this overview of related work, we focus on wireless firmware [13].
sensor networks with dynamic programming capabilities This is an interesting approach if the node behaviours do not
and limit ourselves to low-power systems that can fit on change often. However, when the behaviour needs to change
a microcontroller with a few kB of RAM and flash memory. often while operating in standard conditions, this approach
is sub-optimal: The re-programmability of most low-power
A. Virtual machine microcontrollers is limited (less than 1000 cycles for some)
Virtual machines are the most common solutions to dynami- and consumes a significant energy. Moreover, Lombriser et
cally re-program nodes in a deployed wireless network [6], [7]. al. [9] showed that using native code to remotely update a node
As these allow to change the running program by sending new uses more bandwidth than with a virtual machine, leading
bytecodes over the air without flashing the microcontrollers’ to lower performances. While it is possible to debug such
code, it is quite difficult with microcontrollers to remotely
2 http://aseba.wikidot.com single steps code or debug memory access. Thus, for the
inexperienced programmer, the learning curve is harder than
I²C
with the solution based on virtual machines. Sensors
Accelerometer
III. A SEBA WIRELESS C HALLENGES
I²C Bus Thymio II
The usefulness of ASEBA to script the behaviour of a swarm Main µC
Actuators
of robots has been demonstrated in a previous work [14].
This work was using E-Puck robots [15], connected through
CC2533 USB
Bluetooth. The main drawbacks of that work were the Radio-µC Slave
robot itself, as the E-Puck costs more than 800 $, and the
Bluetooth protocol, which limits the number of robots in a Fig. 2: A diagram of the Thymio II electronics architecture.
network to a maximum of 7. Furthermore, all data transited
through a central computer, doubling the delay compared to a
broadcasting-based approach. Although the main target of the other frameworks such as ROS. Because the wireless layer
current work is the Thymio II robot, the proposed protocol must be compatible with all these, and because ASEBA is
shall be generic for any ASEBA network. The microcontroller already employed in several robots [14], [16], [17], [18], [19],
of the Thymio II can only handle simple serial protocols and the protocol can only change in a backward-compatible way.
no complex packets re-transmission or routing because of
memory constraints. The added electronics must encapsulate D. Low power
all the RF protocol and provide a simple interface using the The ASEBA protocol is meant to be used on robots with
native ASEBA protocol. actuators and sensors. The power consumption while the
A. Cost robot is running is thus not a critical factor as long as it
stays moderate. For instance, the power consumption of the
In the educational context, the cost is a critical factor Thymio II robot is about 1 W while switched on. A wireless
to have a wide acceptance in schools and families. The module consuming less than 100 mW is therefore perfectly
wireless addition to the Thymio II robot must therefore be acceptable.
as low cost as possible. As the targeted production volume
is moderate (starting with batches of 1000 units), the cost IV. T HYMIO II WIRELESS
computation is not limited to the different electronic chips The Thymio II robot has an internal extension connector
but must include the industrialization, the tooling of the providing an I 2 C bus and battery power. Any additional
printed circuit board (PCB) as well as the certification and module needs to use this connector. A schematic view
needed licenses. Furthermore, this analysis must include any of the hardware connection between the Thymio II main
necessary additional hardware to have a functional setup, such microcontroller and its various peripherals is shown in Fig. 2.
as a USB dongle on the computer side. The physical constraints inside the robot are tight, which
B. Ease of use limits the PCB area available for the wireless module. One
additional constraint is the powered-off consumption. As the
As the Thymio II robot is used by children, the ease of use
Thymio II electronics cannot power down the wireless module,
is very important. A simple configuration step is too much
an efficient software power down must be implemented. The
to handle and would discourage the user. Thus the wireless
wireless module must also be able to wake up through an I 2 C
setup must be truly plug and play, without any intermediate
bus access (as shown in Fig. 2).
steps. If any complex network configuration is needed, such
as having different separate networks in the same area, this A. Hardware
configuration should be manageable by people unfamiliar
We selected the radio integrated circuit (IC) among the
with technology. Moreover, as users might use Thymio II both
vast choice offered by almost all major semiconductor
wirelessly and through the USB cable, these two interfaces
manufacturers based on the following needs:
must behave the same when accessed from a computer. In
• a robust modulation in the 2.4 GHz worldwide ISM band,
addition, the user should be able to switch from one to the
• a low total production cost, including bill of materials
other during the same programming session, without loosing
her work. This use case is realistic, as the user may need to (BOM), licenses and needed certification,
• a native USB interface (computer side),
recharge the robot through USB. 2
• a native I C interface (robot side),
C. Compatibility • an ultra-low powered-down consumption,

This requirement is linked to the previous one, but with • at least 4 kB of RAM and 64 kB of Flash memory,

deeper consequences in the protocol implementation. A SEBA • a 3.3 V power supply,

provides multiple software tools, including integration with • a C compiler available.

 ✁✂✄☎✆✝ ✗ ✘✙✚✛✜✢✜✙✢✣✙✚✛✜ ✤✚✛✥✦✧ side. The USB dongle transparently emulates the ASEBA
★✩✙✪✫✬✪✦✜ ✤✚✛✥✦✧ serial protocol used by the physical USB connection on the
☎✟✁ ☎✟✁ Thymio II. It encapsulates the full RF stack and performs
packet reassembly when needed. Therefore, all existing
☎✟✁ installations of ASEBA can directly use the RF module. This
✁✂✄☎✆✝ ✞ ✁✂✄☎✆✝ ✌ decreases the maintenance burden, especially in schools where
users do not have administrative rights on the computers to
☎✟✁ install additional software, drivers or change network settings.
☎✟✁ ☎✟✁ ☎✟✁ Moreover, our solution provides a smooth transition from
✍✎✏ ✍✎✏ programming through a USB connection, then switching to
✑✒✓✔✕✖ ✑✒✓✔✕✖
the RF one, and finally going back to the USB one for
recharging the battery. These transitions are done by plugging
✠☎✡☛☞✂✁✆ ✠☎✡☛☞✂✁✆ and unplugging the dongle or the robot’s USB connection,
and do not require any configuration change.
Fig. 5: Different ASEBA networks configurations
V. E XPERIMENTAL VALIDATION
This section shows the performances of the ASEBA wireless
2) Multiple co-existing networks: One main target use of interface. These are mainly limited by the current imple-
the Thymio II RF interface is a classroom, with several distinct mentation on the Thymio II robot. Indeed, the 400 kHz I 2 C
networks in the same room (Fig. 5). We implement separation communication bus between the robot’s main microcontroller
between different networks using the IEEE 802.15.4 personal and the RF module is the weakest part in terms of bandwidth.
area network identifier (PAN ID) in addition to a separate radio This communication bus is shared with others sensors, further
channel. Currently only three different channels are used to limiting the bandwidth dedicated to the radio communication.
minimize cross-talk with main Wi-Fi channels, but our system This I 2 C bus is based on a master/slave architecture and does
can be extended to the full sixteen channels supported by not provide any interrupt line, forcing the main microcontroller
IEEE 802.15.4. In the example of Fig. 5, a first network is
to constantly poll the radio microcontroller to check if some
completely standalone, with multiple nodes exchanging events data are available. The chosen polling frequency is 10 ms
without any central authority. A second network is formed by because the main use case of the Thymio II robot is interacting
just one node and a computer debugging/programming this with humans. As the next section shows, this has a strong
node. A third network is a mixture of the first and second influence on the latency of the exchanged messages.
cases, where the computer is programming a whole set of
nodes. The ASEBA framework enables the user to switch A. Latency
on-the-fly between the different network topologies just by Fig. 6 shows an histogram over 600’000 measurements of
powering on and off the different elements. This network is the latency between two Thymio II robots at a distance of
based on unslotted IEEE 802.15.4, a detailed performance 30 cm. Since the robot is the limiting factor we performed
analysis of such networks can be found in [23]. the evaluation on the worst case scenario of a communication
The user can change the PAN ID and the radio channel of between two robots. The latency between a PC and a robot,
the USB dongle using a simple configuration running on a while not experimentally evaluated, should be better while
desktop computer. This tool puts the dongle in a presence- staying in the same magnitude order. One robot was emitting
beacon–broadcasting mode. The user can then put the robot a “Ping” event while the second robot was emitting a “Pong”
in a “pairing mode” by approaching it to the dongle and by event immediately after receiving the “Ping” event. This
holding two buttons for 5 seconds. Doing so will make the experiment measures the latency of the whole communication
robot scan every IEEE 802.15.4 channel for presence beacons. stack: from the main microcontroller down to the RF layer and
Using the received strength of the beacon signal, the robot will back. The achieved latency is 20 ms, which is fully expected
automatically re-configure itself to join the closest network. because the main delay in the system is the 10 ms polling
Two LEDs blinking in a synchronized way on the USB dongle latency on each robot. Therefore, this relatively large latency
and the Thymio II robot give a feedback to the user. This is specific to the Thymio II robot and would be lower in a
procedure has been designed to be easily performed by people robot with an available interrupt channel.
unfamiliar with technology, such as first grade teachers.
3) Compatibility: As the goal of the presented extension B. Events rate
is to give wireless communication capabilities to existing Fig. 7 shows the event throughput given the event’s payload
ASEBA robots, its compatibility with the ASEBA framework size, between two robots. The performance between one
was a pre-requisite. The developed solution is fully compatible robot and a PC would be the same as the weakest link limit
and does not require any interface change on the computer the performance of the whole communication chain. The
 150000

4
Packet error rate (%)

3
Event Count

2
50000

1
0

0
0 5 10 15 20 25 30 35 5 10 20 30 40 50 60 70

Round Trip Time [ms] Distance [m]

Fig. 6: The latency between two Thymio II robots. Fig. 8: The packet error rate in function of distance.

C. Range
Fig. 8 shows the dropped event rate in function of the
300

distance. The measurement was performed in an obstacle-free

Events/s

environment between two robots over 10’000 events. We see

that the rate is below 1 ‰ up to 35 meters, which is beyond
200

the target range of a classroom. The dropped packet count was

in the measurement noise for the first 30 meters. Therefore,
the measured range completely fulfils the requirements.
100

In addition, we performed a test with a Thymio II enclosed

0 5 10 15 20 25 30 inside a freezer. Indeed, the robot is fitted with a temperature
Event Size [16−bit word] sensor, allowing an interesting educational experiment in
which the robot measures the temperature in different places.
Fig. 7: The event throughput in function of payload size. We verified that the robot was still able to communicate
smoothly with the USB dongle in this situation.
D. Ease of use
maximum size that a Thymio II robot can send is 32 16-bit To deploy the wireless interface, the only adaptation needed
words. The minimum achieved throughput is 100 events/s, to the existing tools is to update the Thymio II’s firmware.
which given that the fastest internal event-generation rate in All other tools will automatically be compatible, validating
the robot is 100 Hz, is sufficient. Moreover, in most situations that the presented solution is truly plug and play. In summary,
fewer than 32 words are required. As we demonstrated in [5], the system is usable in a wide variety of configurations:
the bandwidth and the bus utilization can be dramatically
• computer to one robot (Fig. 5, network 2), when
lowered by adopting an event-based control policy, therefore
developing a behaviour for a single robot;
the throughput is not a limiting factor in an ASEBA network.
• computer to multiple robots (Fig. 5, network 3), for
Thus the network is limited by the events rate and not by the
instance when several robots are used to animate a LEGO
number of robots.
structure;
While Fig. 7 shows results with a single network of two • many to many robots, no computer (Fig. 5, network
robots, but several will be used concurrently in a classroom. 1), for instance to demonstrate bio-inspired collective
Because of the limited speed of the I 2 C bus, the current behaviour, such as flocking [24].
system uses only about 25 % of the physical bandwidth
(250 kbps) of the IEEE 802.15.4 physical layer. Therefore, VI. C ONCLUSION
3 to 4 robots can be used on a single channel without much Using the presented solution, a user unfamiliar with
effect on the usable bandwidth. Hence, using three different technology can program, debug, and monitor a network of
channels accommodates up to 12 robots without any loss of wireless robots easily, without any prior technical knowledge.
performance. With more robots, performances will decrease Thymio II and ASEBA are open-hardware/source projects,
but will most-probably still be sufficient for educational use. and ASEBA runs on all major operating system, including
If full speed in required for more than 12 robots, additional Android tablets. We plan to industrialize the wireless module
IEEE 802.15.4 channels can be used, at the price of increased and release it under an open-hardware license. Therefore,
collisions with Wi-Fi. its diffusion will not be encumbered by expensive fees or
restrictive licenses. This is important for the adoption in [17] M. Bonani, S. Magnenat, P. Rétornaz, and F. Mondada, “The hand-
public schools, which run on tight budgets and cannot afford bot, a robot design for simultaneous climbing and manipulation,” in
Intelligent Robotics and Applications, pp. 11–22, Springer, 2009.
non-sustainable, restrictive solutions. Therefore, we believe [18] J. Fink, F. C. Vaussard, P. Rétornaz, A. Berthoud, F. Wille, F. Mondada,
that the combination of Thymio II, ASEBA and the wireless and P. Dillenbourg, “Motivating Children to Tidy up their Toys with a
interface is a significant progress to the diffusion of robotics Robotic Box,” in HRI pioneers workshop, 2013.
[19] F. Rochat, P. Schoeneich, O. Marti, H. Bleuler, and F. Mondada,
and programming activities in schools. “Cy-mag3De: magnetic climbing inspection robot,” in Field Robotics:
Proceedings of the 14th International Conference on Climbing and
R EFERENCES Walking Robots and the Support Technologies for Mobile Machines,
pp. 407–414, World Scientific, 2011.
[1] S. Magnenat, F. Riedo, M. Bonani, and F. Mondada, “A programming [20] D. Yang, Y. Xu, and M. Gidlund, “Wireless coexistence between ieee
workshop using the robot “thymio II”: The effect on the understanding 802.11-and ieee 802.15. 4-based networks: A survey,” International
by children,” in Advanced Robotics and its Social Impacts (ARSO), Journal of Distributed Sensor Networks, vol. 2011, 2011.
2012 IEEE Workshop on, pp. 24–29, IEEE, 2012. [21] A. Dunkels, B. Gronvall, and T. Voigt, “Contiki-a lightweight and
[2] F. Riedo, P. Rétornaz, L. Bergeron, N. Nyffeler, and F. Mondada, “A flexible operating system for tiny networked sensors,” in Local Computer
two years informal learning experience using the thymio robot,” in Networks, 2004. 29th Annual IEEE International Conference on,
Advances in Autonomous Mini Robots, pp. 37–48, Springer, 2012. pp. 455–462, IEEE, 2004.
[22] G. Montenegro, N. Kushalnagar, J. Hui, and D. Culler, “Transmission
[3] R. Balogh, “Educational robotic platform based on arduino,” in Pro-
of ipv6 packets over ieee 802.15.4 networks,” IETF RFC, vol. 4944,
ceedings of the 1st international conference on Robotics in Education,
September 2007.
RiE2010. FEI STU, Slovakia, pp. 119–122, 2010.
[23] B. Lauwens, B. Scheers, and A. Capelle, “Performance analysis of
[4] M. Quigley, K. Conley, B. Gerkey, J. Faust, T. Foote, J. Leibs,
unslotted CSMA/CA in wireless networks,” Telecommunication Systems,
R. Wheeler, and A. Y. Ng, “Ros: an open-source robot operating
vol. 44, no. 1-2, p. 109–123, 2010.
system,” in ICRA workshop on open source software, 2009.
[24] C. W. Reynolds, “Flocks, herds and schools: A distributed behavioral
[5] S. Magnenat, P. Rétornaz, M. Bonani, V. Longchamp, and F. Mondada,
model,” in ACM SIGGRAPH Computer Graphics, vol. 21, pp. 25–34,
“Aseba: a modular architecture for event-based control of complex
ACM, 1987.
robots,” Mechatronics, IEEE/ASME Transactions on, no. 99, pp. 1–9,
2010.
[6] P. Levis and D. Culler, “Maté: a tiny virtual machine for sensor
networks,” in ACM Sigplan Notices, vol. 37, pp. 85–95, ACM, 2002.
[7] R. Muller, G. Alonso, and D. Kossmann, “A virtual machine for sensor
networks,” ACM SIGOPS Operating Systems Review, vol. 41, no. 3,
pp. 145–158, 2007.
[8] P. Levis, D. Gay, and D. Culler, “Active sensor networks,” in Proceed-
ings of the 2nd conference on Symposium on Networked Systems Design
& Implementation-Volume 2, pp. 343–356, USENIX Association, 2005.
[9] C. Lombriser, D. Roggen, M. Stager, and G. Troster, “Titan: A tiny task
network for dynamically reconfigurable heterogeneous sensor networks,”
in Kommunikation in Verteilten Systemen (KiVS), pp. 127–138, Springer,
2007.
[10] S. Madden, M. Franklin, J. Hellerstein, and W. Hong, “Tinydb: an
acquisitional query processing system for sensor networks,” ACM
Transactions on Database Systems (TODS), vol. 30, no. 1, pp. 122–173,
2005.
[11] P. Pichler, B. Weber, S. Zugal, J. Pinggera, J. Mendling, and H. Reijers,
“Imperative versus Declarative Process Modeling Languages: An
Empirical Investigation,” in Business Process Management Workshops,
vol. 99 of Lecture Notes in Business Information Processing, pp. 383–
394, Springer Berlin Heidelberg, 2012.
[12] C. Buckl, S. Sommer, A. Scholz, A. Knoll, and A. Kemper, “Generating
a tailored middleware for wireless sensor network applications,” in Sen-
sor Networks, Ubiquitous and Trustworthy Computing, 2008. SUTC’08.
IEEE International Conference on, pp. 162–169, IEEE, 2008.
[13] A. Dunkels, N. Finne, J. Eriksson, and T. Voigt, “Run-time dynamic
linking for reprogramming wireless sensor networks,” in Proceedings
of the 4th international conference on Embedded networked sensor
systems, pp. 15–28, ACM, 2006.
[14] S. Magnenat, P. Rétornaz, B. Noris, and F. Mondada, “Scripting
the swarm: event-based control of microcontroller-based robots.,” in
SIMPAR 2008 Workshop Proceedings, 2008.
[15] F. Mondada, M. Bonani, X. Raemy, J. Pugh, C. Cianci, A. Klaptocz,
S. Magnenat, J.-C. Zufferey, D. Floreano, and A. Martinoli, “The e-
puck, a robot designed for education in engineering,” in Proceedings
of the 9th conference on autonomous robot systems and competitions,
vol. 1, pp. 59–65, 2009.
[16] M. Bonani, V. Longchamp, S. Magnenat, P. Rétornaz, D. Burnier,
G. Roulet, F. Vaussard, H. Bleuler, and F. Mondada, “The marxbot,
a miniature mobile robot opening new perspectives for the collective-
robotic research,” in Intelligent Robots and Systems (IROS), 2010
IEEE/RSJ International Conference on, pp. 4187–4193, IEEE, 2010.