2008 IEEE International Conference on

Robotics and Automation

Pasadena, CA, USA, May 19-23, 2008

An active connection mechanism for modular self-reconfigurable robotic

systems based on physical latching
A. Sproewitz, M. Asadpour, Y. Bourquin and A. J. Ijspeert

Abstract— This article presents a robust and heavy duty

physical latching connection mechanism, which can be actuated
with DC motors to actively connect and disconnect modular
robot units. The special requirements include a lightweight and
simple construction providing an active, strong, hermaphrodite,
completely retractable connection mechanism with a 90 degree
symmetry1 and a no-energy consumption in the locked state.
The mechanism volume is kept small to fit multiple copies
into a single modular robot unit and to be used on as
many faces of the robot unit as possible. This way several
different lattice like modular robot structures are possible.
The large selection for dock-able connection positions will
likely simplify self-reconfiguration strategies. Tests with the
implemented mechanism demonstrate its applicative potential
for self-reconfiguring modular robots. Fig. 1. Top view on a prototype of the active connection mechanism,
shown is one connection plate, the DC motor (1) with gearbox (2), a spur
I. I NTRODUCTION gear system (3), the distributing gearbox (4) and the latch mechanism (5).

Modular self-reconfigurable (MSR) robotic systems are

characterized as being composed of multiple robotic units,
each unit having few degrees of freedom (dofs) and a high
degree of autonomy (energy- and computational autonomy,
an own sensor system and communication capabilities).
A single MSR unit cannot usually do much on its own,
but by combining several units into a larger structure various
tasks can be performed.
The number of different types of MSR units in the MSR
system is very low, typically one ([1], [2], [3], [4], [5]) or
two active units [6] are used, or a combination of one active
and passive unit type ([7], [8]). These main characteristics Fig. 2. Bottom view. Scale of the ruler in [cm], the shown board size
is 8 cm by 8 cm. The shown board, but without the actuation (DC motor)
distinguish MSR robotic systems from monolithic robot and with two (diagonally) oriented latches being replaced by spring loaded
structures (e.g. humanoid robotic systems) and create specific blockers is used for the load experiments described in section III-A.
advantages and disadvantages when comparing both of them.
A. Advantages of MSR
able to experiment with different robot configurations using
Common reasons for developing MSR robotic systems are only a limited set of units makes MSR and modular robotic
the a) generation of redundant systems. Because the overall (MR) systems also a c) very attractive tool for scientific
robot structure is made from a lot of identical units, one research [11]. Another application is within d) swarm robot
failing module can be easily replaced. In the best case this applications [12]. By dividing into several smaller units, area
is done autonomously, resulting in a self-repairing system. exploration can be performed in parallel. Once an optimal
b) If needed, a MSR system assembled into e.g. a quadruped- route is found, the MSR structure is re-assembled and can
shaped robot can alter itself into a snake-shaped system, follow the optimal path. Often mentioned is e) an advantage
crawling through a hole non-accessible by the quadruped in price for MSR robotic systems. If many of them are being
configuration [9]. Similar strategies are used in substrate- produced costs can be reduced. However, this argument is
MSR robotic systems [10]. For transport MSR systems can mainly valid for a mass-market product. Projects like the
self-reconfigure into a small-sized and dense cube. Being open Molecubes Project [13] could support the development
Alexander Sproewitz, Masoud Asadpour, Yvan Bourquin and Auke Jan of low-cost MR systems.
Ijspeert are with the School of Computer and Communication Sciences at To sum it up, MSR robotic systems are specially suited
EPFL, Swiss Federal Institute of Technology, CH 1015 Lausanne, Switzer- in areas where a high degree of autonomy is needed, such
land. Email: (alexander.sproewitz, yvan.bourquin, auke.ijspeert)@epfl.ch,
[email protected] as hazardous areas, disaster sites or in places for which it
1 commonly also used: 4-way symmetry is too expensive to send human workers (space or deep sea

978-1-4244-1647-9/08/$25.00 ©2008 IEEE. 3508

applications). two MSR units has to lock all the open degrees of freedom
(dof). Therefore, blocking of x, y and z-translation (see also
B. Disadvantages of MSR Fig. 4) movement or blocking of rotation around the z-axis
Arguments against the usage of MSR robotic systems are: and translation in the z-direction are needed. The z-axis is
because a MSR system is made from identical units, it is orthogonally oriented towards the connection face, the y and
imperatively featuring A) less optimal dynamic and geomet- x-axes follow the right hand rule.
ric properties when comparing it to a monolithic structure Guidance, x and y position: This is needed to bring together
optimized for a specific task. MSR systems usually feature two opposing faces in such a way that the actual connection
B) many active and complex units, hence the overall system mechanism is able to lock the connection. Several principles
can become costly. A further challenge is the C) coordination are used: a) Passive guidance by structured surfaces (e.g.
of a high amount of units for e.g. locomotion tasks. Avoiding used in [16], [21]). By using the pins as guidance, together
internal collision and creating the right connection for self- with a wiggling movement of the MSR units, positioning can
reconfiguration is only possible if D) efficient reconfiguration be achieved as well2 . However, the MSR unit must be able
strategies are available. to perform the required wiggling movement. This requires
an appropriate choice of dof. The bigger the pins, latches,
C. Hardware of MSR grooves and surface structures are, the easier and more robust
A big part of the feasibility of a MSR system derives is the docking procedure3 . b) Active guidance is achieved by
from a highly robust, cheap, well balanced mechanical and using e.g. IR-sensors [2] or camera systems [9] to guide
electronic design of the single units. That includes the pos- the searching movement of the ACM. c) When using ACMs
sibility for high speed communication among the modules, based on magnetic or electrostatic forces, the passive self-
sufficient energy and processing power, a sensor architecture alignment features by the magnetic flux or the electric field,
delivering autonomy to the MSR units, and a properly chosen respectively, can be used.
actuation power. Finally the MSR robotic units need to Grasping movement, z-translation: If two connection faces
feature a reliable and robust active connection mechanism are aligned in the x and y directions, it is advantageous
for interconnecting the MSR units. if the ACM is able to overcome gaps in the z-direction
This work concentrates on active connection mechanism by a grasping/scooping movement. These gaps can derive
principles based on physical latching, rather than on connec- from dust and obstacles between the connection plates, from
tion mechanisms based on electro-static, electro-magnetic, or uneven surfaces (e.g. after usage or from bending under
permanent-magnetic forces. We give a brief discussion for stress), or from a slight angular misalignment of the plates.
this choice (Section II). We explain the resulting robustness Magnetic and electrostatic principles automatically include
against misaligned connection plates and the background for forces in the z-direction, certain physical latches based on
heavy-duty connectors, together with details for the imple- hook designs do this as well (see Fig. 3 for hook trajectories).
mentation of our prototype (Section III). Section IV refers However physical latching systems based on pin & hole
briefly to Self-Reconfiguration Planning to show its impact systems do not support this ability. A single MSR unit (unit-
on the latch design. In Section V we give details for the new a) must be fixed in position and the approaching unit-b needs
feature of an active connection mechanism for MSR robotic to insert the pin into the hole until the latching procedure
systems implemented in Webots, a simulation environment can lock the mechanism. This can lead to opposing forces
we are using. We discuss our approach in Section VI. between the two MSR units if they are not perfectly aligned.
In case unit-a is not sufficiently fixed and the mounting
II. C HOOSING A PRINCIPLE process is not completely aligned, unit-a might simply be
Active connection mechanisms (ACMs) can roughly be pushed away by unit-b. Connection systems based on screws
characterized by the forces they are using to hold MSR (using rough threads ressembling connectors for cameras
robot units together; a) permanent-magnet based magnetic with tripods) can be seen as an extreme case of a hooking
force, e.g. [1], [14] b) electromagnetic forces, e.g. [15], [16] system-the challenge will most likely be to feed-in the thread.
c) electrostatic forces [17] and d) ACMs based on physical Communication, power supply: Any connection mecha-
latches or pins, e.g. [4], [8], [18], [19], [2], [20], [21], [7], nism presents a barrier between the modules, but stable
[12], [5]. There is no MSR robotic system known to the and high-speed communication between the MSR units is
authors using e) atmospheric pressure (e.g. suction-power crucial. It is also sometimes desirable to have power lines
connectors), however hydrostatic pressure can be used to interconnecting the modules (for relayed recharging, in case
connect modules [21]. Some of the above principles are of power-units. . . ). For communication, some systems use
actually used in combination; Molecubes are using electro- IR-communication or even re-use the IR-sensors. Others rely
magnets that “selectively weaken and strengthen connec- on wireless communication between the modules [23]. Like
tions” [16], using additional permanent magnets. this direct connections anywhere inside the robot assembly

A. Demands on an active connection mechanism

2 also ”blindly” or by judging from sensor values measuring internal
Locking the dof: If no movement between the connection bending [22]
plates is desired, a connection mechanism interconnecting 3 extreme values are achieved with e.g. the Dragon connector [19]

3509
are possible, plus they are not dependent on the quality of the
contact connectors. Contact connectors are widely used and
can relay information and power. If the interconnecting hooks
or pins are reused as contact connectors one can provide
perfectly flat surfaces for the MSR units [8]. Standard contact
connectors are based on spring contacts will slightly protrude
the surface.
Symmetry and gender: When assembling a MSR robot sys-
tem from several units, a high number of docking orientations
is desirable. A 90 degree symmetry of the connectors on the
connection faces refers to 4 possible docking positions. If the
ACM is a hermaphrodite, the self-reconfiguration planning
can be simplified (see Section IV for details.).
Connector strength, power consumption, geometry: The
overall connector must provide shear strength and withstand
tensile stress. Structured surfaces can be designed to cope
partly or completely with shear forces (e.g. for permanent
magnetic connectors being vulnerable to shear forces). Be-
Fig. 3. In the left: applying forces in several hook design principles, on the
cause MSR robotic systems are designed as autonomous right trajectories for the corresponding latch design. Walls are like in Fig. 5,
systems, they usually have a restricted battery capacity. Fext (e.g., Ftensile ) is an external force pulling away the neighboring module.
Therefore, the ACM should consume power only during [a] simple hook principle, [b] proposed design, [c] clamping principle as
used with AMAS [8] and MTRAN III [9], the actual shape and sizes differ.
the docking and decoupling process. This is mostly not In design [b] and [c] applying force-lines will be directed only through the
valid for electro-magnetic ACMs4 . If a simple hook is used latch and the direct hull area, in design [a] force-lines go through the hook,
and the actuation system is reversible , Fx (see Fig. 3) will axis and hull structure. The “scooping” motion in design [b] is visible, a
property taken from the simple hook design.
insert a torque to the hook trying to open the connection.
A trivial demand is a flat and small-sized ACM design-the
more attaching surfaces are available the more global robot
configurations are possible. The desired properties of our
new connector include properties of a MSR system similar
to [9], [2], with MSR units based on a cubic grid (110 mm
grid size). Humans will interact with the MSR system (safety
issue), and the MSR systems should be able to withstand
relatively high forces when combined into a global robot.
It should be small sized and should not consume energy
in the docked state. Self-alignment in x- and y- directions
and robustness against z-axis misalignment is preferred. In
summary, the aimed features are: Fig. 4. Connector axes. For the load experiments described in section III-
A two real connectors are assembled, positioned as in the above picture
1) Physical latching with no Fx -fraction (Fig. 3a). The tensile and being connected by two latches, any actuation (DC motor) is removed
force should possibly remaine inside the hook (as in [8]). An during load test. Fshear in the load experiments is directed −y, Ftensile along
electro-static ACM is not human-friendly (current systems a z-axis. Mshear uses the z-axis for rotation and a 40 mm lever (half the cube
size).
use high voltages directly applied on the surfaces), whereas
pure electr-magnetic ACMs consume energy in the holding
state. Permanent magnetic force-based systems would feature
using a similar latching system remain, satisfying almost all
self-alignment, however the applied forces on the global
requirements: AMAS [8] and MTRAN III [18]. Because we
robot would likely require very strong permanent magnets.
are planning for a MSR robotic system with relatively high
To create strong magnetic fields, enough magnets and suffi-
applied forces5 and less accuracy for z-alignment6 , we were
cient flux-conducting material is required and the connector
searching for a design featuring the grasping movement of a
is likely to become heavy.
hook (Fig. 3), together with the capability to trap Fext inside
2) A hermaphrodite connector with 90 degree symmetry. If
the hook only (Fig. 3 representing–roughly–the AMAS and
possible, the latches should serve a dual-purpose for power
the MTRAN III latching system).
and communication relay.
Therefore, we use a system producing a “scooping” move-
III. D ESIGN & IMPLEMENTATION ment (Fig. 6 and Fig. 5a to Fig. 5f). To redirect Fext -lines
Looking at the above demands and the existing solutions into the hook, a latch must be shifted between the hook
for active connection mechanisms, two MSR robot systems
5 Therefore connection surfaces might get bent by strain, resulting in z-
4 Only if the electro-magnetic ACM is used in parallel to permanent misplacement.
magnetic forces weakening them for disconnection. 6 longer chains for self-reconfiguration movements than MTRAN III

3510
 Fig. 7. Exploded view for the latching system with four latches. Screws are
not shown, the board size is 80 mm by 80 mm. We also designed a version
with only two latches, what simplifies the torque distribution, lowers the
friction of the overall setup and, as the main reason, decreases the board
size to a circle of 65 mm diameter at a height of 16 mm.

Fig. 5. Latching mechanism in the sectional view. The latch shifting in A. Load and alignment tests for the latching system
(Fig. 5e) is closing the force-lines.
As we are aiming for a heavy-duty latching mechanism,
we tested our design under the following conditions: an
“active” connection board (B1) as in Fig. 1 is fixed by screws
against a wall. We removed the DC motor (the screw drive
is non-reversible) and used only two latches (see Fig. 5 for
a latch design). We replaced the other two latches each by
a spring-loaded blocker without latching capability to cope
for applying torque between the two connectors. Each of the
latches is made from two identical pieces and has a thickness
of 3.2 mm (two 1.6 mm-thick glass-fiber reinforced plastic
pieces being copper-plated and soldered together) with a
cross-section area of approximately 7.7 mm2 at its weakest
point.
The opposing connection board (B2) is a blank board
Fig. 6. Steps of 1 mm of the slider motion are overlaid, the tip of the hook
with notches only, such that the two latches from board B1
shows a scoop-like motion path (bold continuous line, due to pin-in-slot can grab onto it. The two blockers from B1 lock against
guidance) other than a circular motion (bold, dashed line) when using a rotation only. Three different load tests were carried out
pin-in-hole guidance. hs − hc corresponds approximately to the maximum
misplacement dz discussed in section III-A.
successfully: Ftensile , Fshear and Mshear (described in Fig. 4),
with each using a load of 18 kg (static load). For Mshear
the torque was applied using the housing as a lever (lever
length of 40 mm), resulting in a torque of Mshear = 7 Nm.
Using only two latches instead of four simplifies the design,
and the connection faces. In case Fext is not too high, the however forces and torque are not applied as evenly as with
system can be used without the translational latch and the four latches, and the supporting walls of B1 and B2 will
specially shaped driving pin, basically representing a dense eventually bend.
standard hook design. The latch is directly attached to a Our latch design is aiming to compensate for an angular
slider moving the leading pin, however the rotation of the misalignment of the y-axis (dyy = 1.7 degrees) or a distance
hook must be stopped (Fig. 5e). Therefore the driving pin is misalignment (z-axis) up to dz = 1 mm purely by the “scoop-
specially shaped and enables a detachment of the driving pin ing” movement into the notches of B2. These values are
from its rotating position, after reaching the position shown measured using the 3D CAD model in Solidworks and a latch
in (Fig. 5e). The slider continues shifting a latch underneath designed to grab a plate of 1.6 mm thickness. The bigger the
the hook until positioned (Fig. 5f). For the inverse movement latch can be designed, the higher the misalignment is allowed
of the slider, the latch is positioned such that the rotation of to be. The slope at the tip of the latch should compensate
the hook starts immideately after the latch has left the gap for 1 mm more dz -misalignment.
between the wall and the hook. This timing is important and
can be supported by a spring holding the hook. The position B. Actuation
of the hook might otherwise become undefined before the Because we are aiming for a minimal motor torque and an
driving pin reaches its rotation point again. acceptable connection time, the lever for the driving pin and

3511
the axis of evolution is kept relatively big. We can currently V. ACTIVE CONNECTION MECHANISM IN SIMULATION
achieve an overall design of 16 mm height including the ENVIRONMENT
actuators. A four-latch and a two latch design fit into a circle In order to validate our design and to be able to make pre-
of an 80 mm and 65 mm diameter, respectively. We have the dictions for bigger MSR structures using our latch design, we
choice for two low-cost DC motor/gear box combinations: are collaborating with Cyberbotics Ltd and using their We-
a 12 mm flat design motor with a 75:1 gear ratio (15 sec bots software described as “a rapid prototyping environment
to retract four latches) and a 15 mm flat design motor with for modeling, programming and simulating mobile robots”
a 30:1 gear ratio (5 sec to retract four latches). We are [24]. In the case of modular robots, the Webots controller-
still working on decreasing the friction inside the actuation program can access simulated sensors and actuators, detect
system (resulting from distributing the torque from one DC the presence of compatible connectors during run-time and
motor onto 4 latches, including 4 worm gears). Eventually also dynamically connect and disconnect robot units from
this will enable us to use smaller gearing ratios and it should each other. Because possible connection mechanism princi-
lead to faster docking procedures. ples differ greatly, Webots provides a general framework, that
allows for both symmetric and asymmetric mechanisms, any
number of rotational docking positions, user-defined docking
IV. S ELF - RECONFIGURATION PLANNING tolerance, etc. We recently added several new features for
the active connection mechanism to get as close as possible
Several features of the latching design, such as its to real-world MSR robotic systems. The newly developed
hermaphrodite properties and the small sized design were “snap” mechanism allows to automatically align connectors
chosen to support the self-reconfiguration planning (SRP). when they are close enough and when the latch mechanism
The goal of self-reconfiguration planning is to design an is triggered. Two snapping parameters distanceTolerance and
optimal algorithm that minimizes the number of steps re- rotationTolerance correspond to the above (section III-A)
quired to reach a final configuration, starting from an initial measured parameters dz and dyy , respectively. Transferring
configuration. The SRP process must address three essential both values from the latch design to the simulation will hope-
questions: fully provide a robust simulation basis for active latching.
1) Possibility: Does any path exists between the initial An additional parameter axisTolerance depends on the angle
configuration of MSR units and the final configuration of between the two z-axes. When the controller program “locks”
the MSR robot? the connection mechanism, the simulation automatically
2) Optimality: In cases where multiple paths exist, which aligns the connectors; the two plates are pulled together, and
of them is optimal in terms of some optimality criteria e.g. they are rotated to the closest matching 90 degree docking
number of reconfiguration steps? position. Another feature we recently implemented is the
3) Computability: How long does it take to find the optimal ”rupture simulation”, that allows to simulate the rupture of
solution out of the possible solutions? And, how does the the connection mechanism under external stress. An example
computation time scale up with an increase in the number is shown in Fig. 8 where two connected robot units (A, B)
of modules? are dropped from 2 different heights onto a solid block. The
In many cases, hermaphrodite connectors are helpful in connection breaks when the units are dropped from a higher
simplifying the reconfiguration process, but they can raise location (Fig. 8-1 and 8-2) while it does not break when
some problems as well. In male/female connections, attention dropped from a lower location (Fig. 8-3 and 8-4). In this
must be paid to bring male connectors close only to female setup, the simulator measures the instantaneous force exerted
ones and vice versa. However, in hermaphrodite connections at each time step on the connection and simply destroys the
all connection points are compatible. Therefore it is feasable connection when a peak of force exceeds the user specified
to reach a larger number of configurations from a specific shear strength or tensile strength.
configuration. This means a bigger solution space as well as a
bigger search space. A bigger solution space gives a higher VI. D ISCUSSION
chance for finding a feasible reconfiguration path between The proposed design shows a possibility to merge the
two configurations (possibility). It thus increases the chance elegant latch design from AMAS [8] and MTRAN III [18]
of finding shorter reconfiguration sequences (optimality). with a higher grasping range, aiming at a higher robustness
A bigger search space, however, results sometimes in a and reliability for the self-reconfiguration sequence in case of
simpler reconfiguration process and sometimes makes it small dz and dyy misalignment between the connecting MSR
more complicated. If by having hermaphrodite connectors a units. The additionally introduced property does not come
kind of high-level planning is made possible, the SRP process without a cost: our current prototype is more complex and
would not only take a shorter time but would also scale very not as compact as the above latching designs. The possibility
well to large numbers of modules (computability). On the to compensate for misalignment depends on the size of the
other hand, if high-level planning is not possible, the SRP latches and the size of the notches. Naturally, a compact MSR
process usually applied would be a random search in a big design does not leave a lot of space for both. For bigger mis-
search space. That means a longer computation time would alignment as compensated by the latching mechanics itself a
be required. “wiggling” movement and compliance in the joints [22] can

3512
 [2] M. Yim, Y. Zhang, K. Roufas, D. Duffa, and C. Eldershaw, “Connect-
ing and disconnecting for chain self-reconfiguration with polybot,”
hdrop hdrop IEEE/ASME Transactions on mechatronics, special issue on Informa-
tion Technology in Mechatronics, 2003.
[3] D. Rus and M. Vona, “A physical implementation of the self-
reconfigurable crystalline robot,” in Proceedings of IEEE International
Conference on Robotics and Automation 2000, vol. 2, 2000, pp. 1726–
[1] [3]
1733.
[4] S. Murata, E. Yoshida, H. Kurokawa, K. Tomita, and S. Kokaji, “Self-
repairing mechanical systems,” Autonomous Robots, vol. 10, no. 1, pp.
7–21, January 2001.
A B A
[5] M. Jørgensen, E. Østergaard, and H. Lund, “Modular atron: Modules
for a self-reconfigurable robot,” in IEEE/RSJ Int. Conf. on Robots and
Systems, Sendai, Japan, 2004, pp. 2068–2073.
B [6] D. Rus, Z. Butler, K. Kotay, and M. Vona, “Self-reconfiguring robots,”
[2] [4] Communications of the ACM archive, vol. 45, pp. 39–45, March 2002.
[7] C. Ünsal, H. Kiliççöte, and P. K. Khosla, “A modular self-
Fig. 8. Rupture experiment in simulation: dropping MSR robot units reconfigurable bipartite robotic system: Implementation and motion
connected by an active connection mechanism. The weight of module A planning,” Autonomous Robots, vol. 10, no. 1, pp. 23–40, January
is largely higher than that of module B in both dropping experiments, to 2001.
keep it staying on the block. [8] Y. Terada and S. Murata, “Modular stucture assembly using blackboard
path planning systems,” in International Symposium on Automation
and Robotics in Construction 2006, May 2006, pp. 852–857.
[9] S. Murata, K. Kakomura, and H. Kurokawa, “Docking experiments
help. It is not yet known how the mechanism will perform of a modular robot by visual feedback,” in IEEE/RSJ International
Conference on Intelligent Robots and Systems, October 2006, pp. 625–
at detaching a global MSR robot having the single units 630.
under tensile and shear stress. Hopefully tests with single [10] K. Kotay and D. Rus, “Generic distributed assembly and repair algo-
connection mechanisms will enable us to gain sufficient data rithms for self-reconfiguring robots,” in IEEE Intl. Conf. on Intelligent
Robots and Systems, Sendai, Japan, 2004.
to feed it to the simulation environment and estimate the [11] D. Daidié, O. Barbey, A. Guignard, D. Roussy, F. Guenter, A. Ijspeert,
behavior for bigger robot structures. and A. Billard, “The dof-box project: An educational kit for con-
figurable robots,” in Proceedings of the 2007 IEEE/ASME Interna-
VII. C ONCLUSION tional Conference on Advanced Intelligent Mechatronics (AIM2007),
September 2007.
We present a robust and heavy-duty active connection [12] F. Mondada, M. Bonani, S. Magnenat, A. Guignard, and D. Floreano,
“Physical connections and cooperation in swarm robotics,” in 8th
mechanism based on physical latching, combining the pos- Conference on Intelligent Autonomous Systems (IAS8), 2004, pp. 53–
sibility to keep applying external forces inside the latches 60.
(similar designs are available with e.g. [8], [18]) with a [13] “Molecubes for everyone.” [Online]. Available:
http://www.molecubes.org
“scooping” movement of the latch. We base our additional [14] J. W. Suh, S. B. Homans, and M. Yim, “Telecubes: Mechanical design
request for a higher z-direction grasping range on an analysis of a module for self-reconfigurable robotics,” in IEEE Intl. Conf. on
of available active connection mechanisms for modular self- Robotics and Automation (ICRA), 2002, pp. 4095–4101.
[15] S. Goldstein, J. Campbell, and T. Mowry, “Programmable matter,”
reconfigurable robots. We believe that extracted features such Computer, vol. 38, pp. 99–101, May 2005.
as genderless, hermaphrodite connectors based on our pro- [16] V. Zykov, E. Mytilinaios, B. Adams, and H. Lipson, “Self-reproducing
posed latch system will support robust self-reconfiguration machines,” Nature, vol. 435, no. 7038, pp. 163–164, 2005.
[17] M. E. Karagozler, J. D. Campbell, G. K. Fedder, S. C. Goldstein,
for our future MSR robot platform. To estimate the be- M. P. Weller, and B. W. Yoon, “Electrostatic latching for inter-module
havior of larger MSR robot assemblies we have developed adhesion, power transfer, and communication in modular robots,”
additional features for MSR in Webots, a physics-based in Proceedings of the IEEE International Conference on Intelligent
Robots and Systems (IROS 07), October 2007.
simulation environment. An example applying these features [18] S. Murata and H. Kurokawa, “Self-reconfigurable robot: Shape-
is shown for a simple rupture/dropping experiment in sim- changing cellular robots can exceed conventional robot flexibility,”
ulation. First tests with the prototype of the implemented IEEE Robotics & Automation Magazine, March 2007.
[19] M. Nilsson, “Connectors for self-reconfiguring robots,” IEEE/ASME
latching mechanism demonstrate its functionality. Transactions on mechatronics, vol. 7, no. 4, Dezember 2002.
[20] A. Castano, A. Behar, and P. Will, “The Conro modules for recon-
VIII. ACKNOWLEDGMENTS figurable robots,” IEEE/ASME Trans. Mechatronics, vol. 7, no. 4, pp.
403–409, December 2002.
This work was made possible thanks to financial support [21] V. Zykov and H. Lipson, “Fluidic stochastic modular robotics: Re-
from Microsoft Research Cambridge, EPFL, and the Swiss visiting the system design,” in Proceedings of Robotics Science and
Systems Workshop on Self-Reconfigurable Modular Robots, Philadel-
National Science Foundation. We gratefully acknowledge the phia PA, August 2006.
technical support of André Guignard, André Badertscher, [22] M. Vona, C. Detweiler, and D. Rus, “Shady: Robust truss climbing
Philippe Vosseler and Manuel Leitos. We would like to thank with mechanical compliances,” in International Symposium on Exper-
imental Robotics, 2006.
Sasha Voloshina for proof-reading. [23] R. Möckel, A. Spröwitz, J. Maye, and A. J. Ijspeert, “An easy to use
bluetooth scatternet protocol for fast data exchange in wireless sensor
R EFERENCES networks and autonomous robots,” in IROS2007, 2007.
[24] “Webots 5, fast prototyping and simulation of mobile robots.”
[1] S. Murata, E. Yoshida, A. Kamimura, H. Kurokawa, K. Tomita, [Online]. Available: http://www.cyberbotics.com/
and S. Kokaji, “M-tran: Self-reconfigurable modular robotic system,”
IEEE/ASME Transactions on Mechatronics, vol. 7, no. 4, pp. 431–441,
December 2002.

3513