NASA Technical Reports Server (NTRS) 20060051846: A Modular Robotic System with Applications to Space Exploration

9 MB·English

by  NASA Technical Reports Server (NTRS)

il
obotic System wit ~ I i ~ a tt~o Sop~acse
Matthew D. Hancher Gregory S. Hornby
NASA Ames Research Center University of CaGornia Santa Cruz
Moffett Field, eA, 94035 Moffett Field, CA, 94035
mdh @email.arc.nasa.gov hornby @ema$arc.nasa.gov
Abstract a single module type. While interesting to study, this form
of modularity is not appropriate to the real-world needs
Modular robotic systems offer potential advantages as of space robotics. A more practical modular robot would
versatile, fault-tolerant, cost-effective plat$orms for space include a number of general-purpose and special-purpose
exploration, but a suficiently mature system is not yet avail- module types, all designed with the anticipated range of
able. We describe the possible applications of such a sys- mission tasks and configurations in mind.
tem, and present prototype hardware intended as a step in In this paper we outline a modular robotic system in-
the right direction. We also present elements of an auto- tended as one step in the right direction. We first summarize
mated design and optimizationf ramework aimed at making related work in modular robotics, and we discuss a num-
modular robots easier to design and use, and discuss the re- ber of potential applications for modular robotics to future
sults of applying the system to a gait optimization problem. space exploration missions. We then describe the design
Finally, we discuss the potential near-term applications of and manufacture of our initial prototype hardware. Next,
nwdular robotics to terrestrial robotics research. we outline a simulation and design environment capable of
automated morphology and controller design and optimiza-
tion, and we present initial results of a quadrupedal gait op-
1. Introduction timization. Finally, we discuss tlie immediate application
of this form of modular robotics to the laboratory and class-
room setting as a low-cost generic platform for robotics re-
Future space exploration missions will require au-
search.
tonomous robotic platforms that can be adapted to a vari-
ety of tasks, both autonomously and in cooperation wilh
humans. A modular robotic architecture has the potential 2. Related work
to offer a high degree of flexibility and fault-tolerance at
a low cost and mass. However, such architectures involve Reconfigurable modular robots first emerged in the
trade-offs that are currently poorly understood and have his- eighties in the context of manipulator arms [l]. The idea
torically rendered modular robots inappropriate for use in a was popularized by the work of Mark Yim at Stanford in
space mission setting. the early nineties, whose work extended the idea to fully au-
A modular robot is a robot built from components with tonomous robots including locomotion [2]. Since that time
standard electromechanicali nterfaces, making it possible to the modular robotics community has focused primarily on
assemble the components in a variety of ways to suit a 'va- homogeneous robots, those constructed from many copies
riety of purposes. The chief advantage of modular robotics of a single module type.
to space missions arises from the ability to reuse hardware Existing modular robots can be divided into two cate-
to perform multiple functions. In a sufficiently complex gories according to whether the primary actuator is rota-
or open-ended mission this could result in a considerable tional or prismatic. RotationaI modules are typically de-
savings in mass and hence also in cost. In addition fault- signed to be connected in chains, sometimes with branch-
tolerance can be achieved with a small number of spare ing, and are well-suited to arm- and snzke-like behav-
-
modules, requiring far less mass than would otherwise be iors and legged locomotion. The PARCRJPenn PolyEot
needed €or redundancy. A self-reconfiguring modular robot robot [3], the Cornell Molecubes [4], and to som, ex-
would even capable of self-repair. tent the DartmouWMIT Molecule robot [SI fall into this
The modular robotics research community has focused category. Some rotational modular robots support self-
primarily on the extreme case of robots made entirely from reconfiguration, while others are manually reconfigured. By
contrast, in prismatic modular robots the function and re- ticipated or changing mission needs. Its capab
configuration of the robot are tightly intertwined. Mod- be extended by launching additional modules at a later date
ules translate relative to each other, typically in a lattice-like as long-term mission needs evolve, which could then inte-
structure, in order lo achieve higher-level goals. For exam- grate seamlessly into the existing robotic community. The
ple, locomotion may be achieved by cyclically moving the ability to easily remove and replace individual modules, ei-
rear modules to he front through a process of continuous ther manually or in an automated fashion, could extend mis-
reconfiguration. The DartmoutWMIT Crystalline robot 161 sion lifetimes.
and the PARClUPenn Telecube robot [7] typify this cate- One important reason existing flight robots exhibit a
gory. highly centralized, monolithic design is the need to protect
A homogeneous design can introduce unnecessary com- the control electronics within a thermally controlled envi-
plexjcy into robot and module designs. For example, rolling ronment. Robots exhibiting a high degree of modularity
locomotion in a traditional rotational modular robot in- will depend on emerging technologies for electronics that
volves a long chain of modules connected in a tread-like can operate directly in extreme temperature and radiation
topology [SI, when in many situations it may be simpler environments.
to add wheel modules directly. Modular manipulator arms
offering a range of rotational joint modules with differ- 3.1. In-space operations
ent properties have been developed, such as the Rapidly
Deployable Manipulator System at Carnegie MelIon [9].
In-space operations include autonomous assembly, in-
These robots begin to flesh out the spectrum of modularity
spection, and maintenance and providing assistance to as-
that lies between single-purpose custom-engineered hard-
tronauts engaged in orbital extra-vehicular activity (EVA).
ware and pure homogeneous modularity. Unfortunately,
These tasks involve moving about the structure and manipu-
no general-purpose heterogeneous modular architecture has
lating both small-scale connectors and instrumentation and
yet been presented.
large-scale components, which is time-consuming and dan-
gerous C12-143. The diversity of tasks at different scales
3. Modular robotics for space exploration
typically requires human involvement or several different
robots.
Space exploration missions continue to place ever- Existing robotic arms, such as the
increasing demands on robotic systems [lo]. For the lim- tor System (RMS) [l4], are good for moving large objects
ited missions of the next decade the traditional approach, around the ISS but are too large for manipulating smaller
using a small number of specialized robots whose capabili- ORUs for repair tasks, are hard to transport and position,
ties span the range of mission requirements, will suffice. As and lack the ability to perform many
we develop permanent robotic and human outposts in space without aid. Humanoid robots, such as
and on the surface of the moon and Mars, however, the var- [15] and the Canadian Space Agency’s Dextre, are more ca-
ious advantages of modular robotic systems become more pable for jobs involving smaller objects but are limited to
appealing [ 111. tasks that are well-suited for a human, such as those that re-
In a human-robot collaborative mission the reconfigura- quire no more tlian two arms. None of these robotic systems
tion process could be facilitated by the astronauts, greatly can be adjnsled to handle tasks outside the range for which
simplifying robot design. For example, when assembling a they were designed.
large structure, astronauts could configure a robotic manip- A reconfigurable moduh robotic system, on the other
ulator system as one large manipulator or as several smaller hand, would be able to form robotic structures at different
ones as desired for each phase of assembly. The same elec- scales and with different manipulation capabilities. Here
tromechanical interface could be used to install the end- the modules types might include a number of actuators with
effectors best suited to each subtask. different specifications, branching modules for construct-
In a purely robotic mission, a modular robotic system ing multi-limbed robots, a variety of end-effectors such as
must be capable of reconfiguring itself. The approach taken grippers and dextrous manipulators, and specialized sensors
in labs today, adding the capability for automatic connec- such as stereo caweras OE LIDAR. These could be assem-
tion and disconnection to each module, is not the only bled to provide the necessary manipulation and sensing ca-
option. For example, a specialized reconfiguration end- pability for the currentjob, such as by forming three or more
effector module could be included, thereby reducing the arms to hold two truss elements in place while manipulat-
complexity of the remaining modules. ing the necessary tools to weld them together and perform
Beyond simply reusing mass for multiple purposes, a inspection. A self-reconfiguring system could form a robot
modular robotic system has the additional advantage that it large enough to manipulate a docking spacecraft, but could
can be configured into entirely new forms to address unan- also form a chain small enough to squeeze through a re-
Figure 1. Artists’ conceptions.of possible scenarios for a modular robotic architecture in space (a)
and in a surface mission in a wheeled (b) and legged (c) operating mode.
stricted space, reconfiguring into a different shape on the peared in the in-space case arise here as well. For exam-
* other side to perform the desired task. One artist’s concep- ple, the same actuators that are used for end-effector place-
tion of a modular robotic system performing various tasks ment in one phase of the mission could be used for legged
about truss structure in space is shown in Figure l(a). locomotion in another. A limited number of spare mod-
Since any orbital s’ystem will likely require at least one ules could be used to provide total coverage for the entire
large arm, but such an arm may only be used a fraction system, and could also be used to enhance functionality in
of the time, the ability to decompose it into a number of the absence of faults. For instance, the artist’s conception
smaller robots for day-to-day operations could represent a in Figure l(b) shows a redundantly actuated rover; in Fig-
significant reduction in total robotic mass. The associated ure l(c) the robot has taken advantage of these actuators to
reduction in the number of spare parts needed to be able enter a legged locomotion mode to traverse rough terrain
to recover from any fault could also be significant. These without the need for reconfiguration. If a module failed, the
cost savings must be weighed against the increased com- robot’s functionality could gracefully degrade to one or the
plexity of each particular robot configuration relative to a other operating mode.
single-purpose robot designed to perform the same task. These artists’ renderings reff ect the conventional, but
Such an analysis is inherently mission-specific, but clearly probably unrealistic, approach to modular robotics in which
as the number of fundamentally different tasks increases the virtually all functionaIity derives from a large number of
potential advantages of a modular architecture increase as small modules. A hybrid approach is also possible and
well. might prove more appropriate for some mission scenar-
ios. For example, a modular manipulation system could be
3.2. Surface operations used in conjunction with a more traditional rover locomo-
tion platform. A rover supporting a lunar base, for instance,
could quickly be outfitted with whatever arms and effectors
Even more expensive than putting a spacecraft in orbit is
are needed for a given task. A self-reconfiguring &pula-
landing one on another planetary surface, since additional
tor system could potentialty even address one of the long-
mass is required for landing mechanisms and additional fuel
standing challenges in rover design, the problem of righting
for deceleration. Thus, the potential mass and cost savings
a rover that has fallen over, by configuring one or more arms
associated with modular robotics are even greater in this do-
with the appropriate geometry to push itself back up.
main. Once on the surface, useful capabilities include loco-
motion, instrument placement and sample operations, and
support of lunar and planetary bases [16]. In addition to 4. Prototype Hardware
the module types that would be useful in space, there are a
number of surface-specific types to consider. Special wheel We have developed prototype modular robot hardware
modules could allow efficient long-range locomotion, while with an eye towards gaining a greater understanding of
force-sensitive foot modules could enable legged locomo- the practical issues that will impact the design of modular
tion over rougher terrain. A range of scientific sensors, robotic equipment for use in space. As discussed later, this
drills, shovels, and other tools could be included as end- hardware also serves a dual purpose in supporting near-term
effector modules. terrestrial robotics research.
The versatility and fault-tolerance advantages that ap- Our modular robots differ from most existing robots in
Figure 2. The three primary module varieties in the first-generation prototype system: a hinge-type
actuator (a), a five-connector hub (b), and a battery and communications module (c).
several respects. Most importantly, they are Izetemgerzeous, Figure 2(a), similar to those found in existing homoge-
consisting of modules of several different types each de- neous robots such as PolyBot [3]a nd the NASA Snakebot.
signed to perform a particular simple function. A number For this generation we used inexpensive s&vomotors of the
of other labs have focused their attention on the problem of sort designed for the hobby industry and manufactured in
self-reconfiguration [3-71. However, many of the advan- volume. We selected a medium-sized servo with a high
tages of modular robots, such as reusability of hardware torquehass ratio, and the module scale was chosen to be
for a range of tasks and reduction in the number of spare as small as possible while accommodating that motor.
components required, do not depend on this ability at all, The hub, shown in Figure 2(b), has a novel structure with
and hardware to support self-reconfiguration can add signif- five connection points arranged to provide a variety of con-
icantly to module weight, size, cost, and complexity. Thus nection angles including 90" and 120". With this design it
the modules discussed in this paper were alI designed for is possible to construct both rectangular and hexagonal lat-
quick and easy manual reconfiguration, in this case using tices for use in assembling larger structural configurations.
thumb-screws. The hub modules also provide power distribution and com-
We present two generations of prototype hardware. The munications switching between neighboring modules.
first generation was a quick and inexpensive design fea- The battery and communications module, shown in Fig-
turing a small number of module types intended lo pro- ure 2(c), allows the robot to operate fully autonomously or
vide some insight into the issues that arise in the design, in a tethered mode, and can be configured with either five
construction, and use of manually-reconfigurable modular or ten AA-size NiMH batteries. Finally, we have also con-
robots. It has also subsequentiy been used as a platform structed passive "foot modules" which can be installed us-
for research in gait optimization and robot arm calibration. ing the module connector to protect the other modules and
The second generation, currently undergoing construction to provide a more uniform surface €or locomotion.
and testing, builds an the lessons of the first and expands The module connectors, shown in Figure 3(a), were de-
the number of module types available. signed for quick and easy manual reconfiguration. They are
four-way symmetrical and slide together using alignment
4.1. First-generation hardware pins. Spring-loaded gold-plated contacts establish the elec-
trical connections, and up to four thumbscrews may be used
Our first-generation prototype modular robotic system to lock each pair of modules together. There are gendered
included three module types: a rotational actuator module, male and female connectors, but this is not restrictive since
a five-connector hub module, and a power and communica- each face on each hub may be configured with either a male
tions module. The design goals were low cost, low mass, or female connector. (On the hub shown in Figure 2(b), the
and small size. Low mass and size are important so that top, fTont, and left faces are configured with male connec-
the behavior of the robot is not limited by motor torque: a tors while the bottom and right faces have female connec-
snake-like arm, for example, is less useful if it cannot sup- tors.)
port its own weight. Achieving low small-quantity module cost and low mod-
The joints are hinge-type actuated modules, shown in ule mass also guided the selection of the primary moduIe
Figure 3. The first-generation gendered electromechanical connectors (a) and second-generation
hermaphroditic connector (b) used in the prototype systems.
material and manufacturing process. The parts were first
printed in an ABS-like plastic using stereo-lithography and
were then plated with a layer of copper followed by a layer
of nickel. The resulting parts have essentially the same den-
sity as common plastics but considerably greater stiffness,
strength, and durability. This is a rapid-prototyping pro-
cess with ;irtually zero set-up and tooling costs, making it
much more attractive than traditional machining processes
for production in research quantities. Moreover, experimen-
tation with minor variations or new module types incurs no
additional cost.
The resulting 6Omm-scale modules are smaller and
lighter than those of other reconfigurable modular robots
in the research community. The hinge modules weigh ap-
proximately 125gm, and each hub module, with no motor
but considerably more structural material, weighs roughly
115gm. The heftier battery module weighs 390gm with
Figure 4. A quadrupedal walking robot.
the usual ten batteries and 240gm with five. For tethered
operation the batteries may be removed entirely, in which
case this module weighs only 90gm. The feet are virtually robots with multiple arms, and legged robots with three and
weightless at just over lOgm each. four legs. A small quadmpedd walking robot is shown in
Each powered module is controlled by an Atmel FPSLIC Figure 4.
microcontroIIerFPCA. The FPGA provides as many com-
munications ports as the module has connectors and inter- 4.2. Second-generation hardware
faces to other on-board hardware such as motors. The mi-
crocontroller,a 25MHz AVR core, manages the higher-level Our second-generation system has been designed to ad-
communications and control functions. The modules com- dress several limitations of the first, and is now partially
municate with each other, and optionally with one or more complete. The gendered connectors of the first system,
control computers, using a simple ad hoc peer-to-peer net- while not StriCKly limiting, s inconvenient,
work scheme. Any module can send data packets to any and the new system featur ditic connector
other module, and the intermediate modules route the pack- shown in Figure 3@). Two entirely new module types have
ets accordingly. This permits true distributed control of the been introduced. The first, shown in Figure 5(a), is an actu-
modules in addition to the usual masterhlave control strat- ated wheel intended for rover-like locomotion at velocities
egy. up to approximately a meter per second. The control elec-
We have tested several different robot configurations us- tronics and the motor are contained entirely within the hub
ing these modules, including the classic snake-like arm, of the wheel. The second new module type is a digital cam-
Figure 6. A simulated modular quadrupedal
walking robot used for evolutionary gate op-
timization.
able future, other aspects axe amenable to automated design
and optimization.
It has been shown that, using an evolutionary search al-
gorithm combined with a representation that- directly en-
(b) ‘codes the underlying hierarchical structure of designs, it is
possible to search a space of modular robot morphologies
Figure 5. Two new module types introduced and controllers to find candidate designs with desired prop-
in the second-generation p e s erties [17]. The problem of premature convergence, which
an actuated wheel (a) and a ngl has historically plagued population-baseds earch algorithms
era (b). in complex spaces, has recently been addressed by using a
specially-designed population structure f IS].
Ultimately, a unified design tool &needed that will give
engineers the freedom to quickly develop and test designs
era, shown in Figure 5(b). The camera transmits images
in simulation while providing them access to automated de-
wirelessly to a controlling computer, thus avoiding the need
sign and optimization tools where applicable. We have be-
for a high-bandwidthi nter-module communications system.
gun developing a few key elements of such a tool, discussed
One of the chief limitations of the first-generation system
below.
was the low accuracy with which the hobby-grade motors
could be controlled. The second-generation system is there-
5.1. ~o~~~~ robot sirnulation
fore being redesigned with high-precision brushless DC ser-
vomotors and backlash-free harmonic gearboxes. The feet
We have developed a physics-based so€tulare simulation
modules are also being upgraded with tactile sensing capa-
bility based on QTC force sensors. Finally, the remaining environment for modular robots in C++, which allows users
electronics are being updated with a larger FPSLIC proces- to construct robots using a variety of module types and
to extend the simulation by adding additional types with
sor and support for a new higher-voltage power bus.
compatible connectors. The physical dynamics simulation,
based on a modified version of the Open Dynamics Engine
5. Automated design and optimization
[1 91. was designed for high-speed medium-fidelity simula-
tion. Thus it is suitable for use both by engineers wishing to
To make the most use of a modular robotic system it rapidly explore design spaces by hand and also within the
must be combined with software tools to assist in select- evaluation loop of a search or optimization tool. The simu-
ing or designing the best morphology and control structure lator supports several levels of photo-realism, ranging from
for each task. While many aspects of this design process simple block representations useful for quick visualization
will inevitably rely on human intelligence for the foresee- (as shown in Figure 6) to fully rendered images that can be
3.5
where it performed as advertised.
3
6. Modular robots in the research lab
2.5
(0
ul
.c2_ 2 Robotics is playing m ever-increasing role in educa-
L.=L tional, industrial, and governmental research labs as well
1.5
a as in the classroom. Many such labs woFk with a variety
8
1 of robot morphologies, including rovers, legged robots, ma-
nipulator arms, and collaborative robot teams. However,
05 research budgets are often small, and hardware budgets are
often smaller. This can constrain the range of morpholo-
0
0 500 1000 1500 2000 2500 3000 gies with which a given la5 can experiment, and may limit
Number of Evaluations
the ability of the robotics community as a whole to explore
new morphologies that are not already commercially avail-
Figure 7. Ten gait optimization runs. Grey
able. Many educational institutions have attempted to work
crosses represent evafuated gaits, and black
around this by developing curricula based on LEG0 com-
lines indicate the best found so far in each
ponents or other toy-grade modular design kits, but these
run.
have well-known limitations.
Heterogeneous modular robots, with their fundamentally
open-ended morphological range, have the potential to alle-
used for simulated closed-loop visual control. viate this probzem. A system of modules optimized for low
We have also developed an abstraction layer that allows cost could provide an extremely versatile research platform
robot control software to transparently operate both the sim- to researchers wishing to go beyond the standard rover and
ulated and real moduiar robots. This makes it trivial to arm morphologies without investing in custom hardware.
transfer controller designs from simulation to hardware for In this domain, manual reconfigurztion of the sort used in
testing or use, and also makes it possible to incorporate test- our modules as discussed above would be entirely accept-
ing on real hardware into an automated design cycle. able. Because all morphologies are derived from a relatively
small number of module types, the manufacturing efficien-
5.2. Quadrupedal gait optimization cies associated with mass production promise to drive the
price down even further.
As an initial test of the simulation and automated op-
timization system, we optimized the walking gait of the 7. Conclusion
quadrupedal robot shown in Figures 4 and 6. We held the
morphology fixed and assumed a periodic gait, with the tra- Autonomous robotic systems are critical to achieving
jectory of each joint parameterized by the first three Fourier sustainabiIity and reliability in NASA's exploration mis-
basis coefficients. We evaluated each candidate controller sion. The current monolithic design approach to robotics
for ten simuIated seconds. In order to search for efficient offers little room for reuse, adaptation, or maintenance on
gaits, the fitness was computed as the ratio of the distance long-duration or open-ended missions. Adopting a modu-
travelled to an energy measure derived from the control out- lar design could address these needs, by allowing a single
puts. We used a steady-state evolutionary search algorichm, system mass to be reconfigured to suit each task and by re-
simiiar to simulated an3ealing but with a population size of ducing the number of spare parts required to achieve redun-
four and a mutation rate that decreased over the course of dancy. However, there are many challenges to the scalabil-
each run. The results of ten runs are shown in Figure 7. ity, reliability, and usability of such a system that a must be
Seven of the ten runs converged to essentially the same addressed before it couid be put to use outside the Iabora-
high-performing result, while three experienced premature tory.
convergence to sub-optimal gaits. The seven all surpassed We have presented initial prototype hardware, intended
earlier efforts at manual optimization. They did so by dis- as a platform for beginning to address those challenges.
covering an unexpected gait which walks "sideways" rela- Though this hardware is still far from being immediately
tive to the originally-imagined direction of travel. Further- nseful in a space mission context, its versatility and US-
more, this somewhat unintuitive gait eliminates the need for ability is steadily increasing and we believe it may have
four of the eight actuators. In order to be sure that the gait immediate applications in the robotics research setting.
was not taking advantage of some unrealistic property of the Each module implements a single core function, reduc-
simulator, we transferred the gait to the real modular robot, ing individual module complexity and cost and aIlowing a
robotic system to be tailored as needed by including special- [9] C. J. Paredis, H. B. Brosn, and P. K. Khosla, “A rapidly
purpose modules. By designing for a rapid-prototyping deployable manipulator system,” in PI-OC1.9 96 IEEE
manufacturing technology, it is easy and inexpensive to add Iizt. Con$ Robot. Autoiiuzt., Minneapolis, MN, April
new module types when the existing types are insuf3icient or 1996, pp- 1434-1439.
to make incremental changes between manufacturing runs.
[lo] L. Pedersen, D. Kortenkamp, D. Wettergreen, and
Finally, we have described the first components of an au-
1. Nourbakhsh, ‘Space robotics technology assess-
tomated design and optimization system for modular robots,
ment report,” NASA Exploration Team (NEXT),
including a modular robot simulator and an evolutionary
Tech. Rep., December 2002.
controller optimization tooI. We have presented the sesults
of applying this system to the optimization of a walking [l 13 M. Yim, K. Roufas, D. Duff, Y Zhang, C. Eldershaw,
gait, and discussed how the system yas even able to out-
and S. Homans, “Modular reconfigurable robots in
perform the human engineer. As the capabilities of both the space applications,” Autorzomous Robots. vol. 14, no.
robots themselves and automated design tools grow, we ex-
2-3, pp. 225-237, March 2003.
pect such tools to be of increasing importance in the use of
modular robots. [ 121 W. Doggett, “Robotic assembly of truss structures for
space systems and future research plans,’’ in Proc.
2002 IEEE Aerospace Cor$, Big Sky, MT, 2002.
References
1131 S. M. Jones, H. A. Aldridge, and S. L. Vazquez,
[l] K. H. Wurst, “The conception and construction of a “Wrist camera orientation for effective orbital replace-
modular robot system,” in Proc. JSME 16th Int. Synzp. abIe unit (ORU) changeout,” NASA, Technical Report
Industrial Robotics ([SIR), 1986, pp. 37-44. TM-4776,1997.
[21 M. Yim, “Locomotion with a unit modular recon- [ 141 D. S. Portree and R. C. Trevino, Walking to Olyinpus:
An EVA Clzronology, ses. Monographs in Aerospace
figurable robot,” Ph.D. dissertation, Stanford Univ.,
History- Washington, D.C.: NASA, 1997, voI. 7.
1994.
[15] W. Bluethmann, R. Ambrose, M. Diftler, S. Askew,
(31 M. Yim, Y. Zhang, K. Roufas, D. Duff, andC. Elder-
E. Huber, M. Ooza, E Rehnmark, C. Lovchik, and
shaw, “Connecting and disconnecting for chain self-
D. Magruder, “Robonaut: A robot designed to work
reconfiguration with PolyBot,” IEEHASME Trans.
with humans in space,” Autonomous Robots, vol. 14,
Mechatiarz., vol. 7, no. 4, pp. 442-451, Dee. 2003.
pp. 179-197,2003.
141 v. Zykov, E. Mytilinaios, B. Adam and H. Lipson,
~161M . M. cohen, “Mobile lunar and planetary bases:’ in
“Self-reproducing machines,” Nature, vol. 435, no.
1 Proc. 2003 AIM $ace Conference, San Diego, CA,
7038, pp. 163-164,2005.
September 2003.
[5] K. Kotay,.D. Rus, M. Vona, and C. McGray, “Theself- [I71 G. S. Hornby, H. Lipson, and J. B. Pollack, “Genera-
reconfiguring robotic molecule,” in Proc. 1998 IEEE tive representations for the automat‘ed design of mod-
Int. Con$ Robot. Automat., Leuven, Belgium, May ular physicd robots,” IEEE Trans. Robot. Autonuzt ,
1998, pp. 424-431. vol. 19,110.4, pp. 703-719, August 2003.
[6] D. Rus and M. Vona, “Crystalline robots: Self- [18] G. S. Hvmby, “ALPS: The age-layered population
reconfiguration with compressible unit modules,” Au- structure for reducing the problem of premature con-
tonontousRobots, vol. IO, no. 1, pp. 107-124, January vergence,” in Proc. 2006 Genetic and Evolutionary
2001. Conzputatioiz Con$, July 2006, forthcoming.
J. W. Suh, S. B. Homans, and M. Yim, “Telecubes: [I91 R. Smith. Open Dynamics Engine (ODE). [Online].
Mechanical design of a module for self-reconfigurable Available: http:llwww.ode.orgl
robotics,” in Proc. IEEE Int. Con$ Robot. Automat.,
vol. 4,2002, pp. 4095-4101.
M. Yim, D. Duff, and K. D. Roufas, “PolyBot: A
modular reconfigurable robot,” in Proc. 2000 IEEE
Int. Con$ Robot. Automat., vol. 1, San Francisco, CA,
2000, pp. 5 14520.