In this article, we review the concept of a cellular robot that is capable

Transcription

1 Self-Reconfigurable Robots Shape-Changing Cellular Robots Can Exceed Conventional Robot Flexibility BY SATOSHI MURATA AND HARUHISA KUROKAWA EYEWIRE AND IMAGESTATE In this article, we review the concept of a cellular robot that is capable of reconfiguring itself. This Self-Reconfigurable (SR) Robot exemplifies a new trend in robotics. Indeed, we can now build various kinds of SR robots with off-the-shelf technologies of processors, actuators, and sensors. These SR robots, based on modern mechatronics, are still not as adaptable as the liquid metal robot in The Terminator 2 but are just as flexible as any conventional robots. A self-reconfigurable (SR) robot is a cellular robot that is capable of adapting its shape and functions to changing environments and demands. The basic component of the SR robot is a mechanical cell ; namely, a robotic module capable of computation. These modules can rearrange their mutual mechanical connection to change the robot s outward features. A multicellular living organism and an SR robot do share some similarities. Both consist of small components, living cells for the former and robotic modules for the latter. Communication among the components is achieved by the diffusion of chemical substances in the former and by the exchange of digital information through module-to-module communication in the latter. In both cases, the components cooperate with one another to adjust their configuration to the environment. The mechanism of cooperation is embedded as the genomic information in the living cell and as a distributed program for each processor in the module. Designing an SR robot requires drastic modification of the conventional approach of robotics. Many of the most successful robots have mimicked the underlying dynamical function of living creatures but not the underlying form. In SR robotics, however, we start with form. The point is that an SR robot is not a robot designed to perform a specific task but a system that develops into various types of robots and executes a variety of tasks. (Note: robots designed for a specific task always work better than SR robots.) Conventionally, the necessary functionalities are determined top-down, from the highest level of abstraction to the lowest (i.e., from the task the completed robot should perform down to the smallest components such as bolts and transistors needed MARCH /07/$ IEEE IEEE Robotics & Automation Magazine 71

2 to build such a robot). In SR robotics, the first thing to be determined is the module s specifications, which in turn determine the potential ability of the completed robot. Therefore, the design of the whole and that of the module are intrinsically indivisible. In such a system, the key issue is how to embed a mechanism into the module so that the developed robot has the desired functions. The Potential of SR Robots By having the ability to transform themselves into different shapes, SR robots have the potential to exceed their conventional counterparts in multifunctionality, flexibility, and robustness. Table 1 summarizes the research issues concerning SR robots, and we describe some of these issues in more detail here. Morphogenesis (Self-Assembly). Traditional robots have a fixed configuration. For example, if a robot is designed and built to be four-legged, it stays four-legged forever. However, SR robots actively change their configuration to achieve their goals. This process is called morphogenesis, or shapeshifting. It is also noted as self-assembly (Figure 1) [1] [3]. Among the many different aspects of morphogenesis, the most important logic is how to differentiate modules from identical to different. Typical problems in morphogenesis [4], [5] include finding an algorithm to shift from any shape to a specific one, finding a reconfiguration path from one certain configuration to another; and finding a suitable configuration under the given environment. Self-Repair. An SR robot is able to adapt not only to the changes in its external environment but also to the changes in its internal environment, like functional disturbances or failures in modules. In such cases, an SR robot can solve the problems by self-reconfiguration [6], [7]. The self-repair process involves The part and the whole Diversity Morphogenesis Robustness Self-reproduction Evolution System architecture Table 1. Research issues concerning SR robots. Self-similar structure, boundary of system Flexibility, multifunctionality; adaptation to changing/uncertain/unstructured environments Logic of growth from homogeneity to heterogeneity; computational complexity Self-repair, graceful degradation, scalability Driving force of evolution, novel method of production Co-evolution between morphology and motion; acceleration of evolution Centralized/decentralized, homogeneous/ heterogeneous, local/global communication four steps: 1) detect a failure; 2) remove the damaged modules; 3) transport undamaged modules to where the damaged modules were; and 4) reassemble the part (Figure 1). Self-Reproduction. A system s ability to produce a copy of itself is called self-reproduction [8]. Because of its modularity, self-reproduction is much easier for an SR robot than for an ordinary robot [34]. Scalability. When a system s functionality depends linearly on the number of modules, the system is called scalable. Graceful degradation is another term for scalability; however, it is only concerned with partial malfunction. Scalability, one of the basic properties of an information network, is difficult to realize in a mechanical system. However, in an SR robot it is relatively easy. Motion Generation. As long as it is called a robot, an SR robot has to be able to move and accomplish meaningful tasks, just like an ordinary robot. Indeed, motion generation is a central issue in robotics. Motion control based on the model equation of a fixed configuration in conventional robotics is not satisfactory for SR robots because they frequently change configurations. Autonomous motion generation for an arbitrary configuration is most desirable; however, it is quite difficult. One feasible solution is semi-autonomous motion generation to reduce human intervention. For instance, if the repertoire of possible motion patterns for a typical configuration is generated a priori, then it is easier to synthesize a suitable motion pattern by combining these motion patterns. Co-Evolution of Shape and Motion. The shape of a robot constrains its motion and the motion affects its shape. Shape and motion are intrinsically indivisible, and they actually are two different aspects of the evolutionary cycle. Karl Sims study on artificial creatures [9] aimed to visualize this evolutionary process by computer simulation. With SR robots, we are able to examine the same process in the real world by hardware, not in the virtual world of computer graphics. Brief History of SR Robotics The idea of building systems by homogeneous components dates back to von Neumann s Theory of self-reproducing cellular automata [8]. His model assumes complete homogeneity of initial modules. This is an analogy to biological systems in which all the differentiated cells are Random Shape Assemble Target Shape Detect Failure Cutting Off Reassemble Figure 1. Self-assembly and self-repair [12]. 72 IEEE Robotics & Automation Magazine MARCH 2007

3 produced by cell division from a single cell after fertilization. However, biological systems need a vast number of cells to form their bodies. In artificial systems, we have to admit heterogeneity among components to some extent, or use complex modules with more functions within themselves. CEBOT (1988) was the first SR robot designed based on heterogeneous components. It is composed of several different types of modules, such as transportation, rotational joint, telescopic arm, and grasping modules. The combination of these modules made it possible for a CEBOT to perform a variety of tasks [10]. The study of SR robots with homogenous modules followed that of those with heterogeneous modules. In the homogenous system, where all modules are identical, a set of common rules must sufficiently describe the differentiation and the behavior of each module. When this is realized, we can replace any module with any other module. This property facilitates self-repair and self-reproduction schemes. Lattice-Type SR Robots Most SR robots can be classified as lattice type. Latticetype modules are just like biological cells or crystal atoms [Figure 2]. These modules are aligned to make a periodic structure with certain geometrical symmetry. Actuators and connectors are designed so that each module can move to adjacent lattice points and connect to others by itself or with the help of a few other modules without collision. In order to assure this property, the latticetype module requires many connectors and actuators. Self-reconfiguration is generically easy and possible motion patterns are well specified by the crystal; therefore, relative alignment of the modules is not necessary. Based on this concept, several two-dimensional (2-D) SR robots were Module developed. In 1994, two SR systems were proposed. The Metamorphic Robot developed at Johns Hopkins University was made of hexagonal link modules driven by servomotors [2]. Fracta, developed by the National Institute of Advanced Industrial Science and Technology (AIST), Japan, (our group) was also a hexagonal module (Figure 3) [3], [33]. It is driven by an electromagnet and has no moving parts (solid-state). Several 2-D SR robots have been proposed thus far [11], [12]. In three-dimensional (3-D) space, SR modules require more connectors and actuators than in 2-D space. Geometrical symmetry of 3-D modules can be defined by space-filling polyhedra such as a regular cube, a truncated octahedron, and a rhombic dodecahedron. Several systems are based on a regular cube [13] [17], [22] and a rhombic dodecahedron [18]. None of them, however, are fully implemented because the large number of degrees of freedom (DOF) and connectors complicate the mechanical design, and they cannot achieve a high enough power/weight ratio. Recently, a group at the University of Southern Denmark developed a lattice-type module called ATRON, depicted in Figure 4 [19], [20]. This spherical module is made of two hemispheres actuated by a servomotor (1-DOF). Because of its simplicity, ATRON has a high power/weight ratio and succeeded in self-reconfiguration of tens of modules. Another way to reduce the 3-D module s mechanical complexity is to combine two modules into one. This kind of Joint Module Branch Figure 2. Types of SR robots. Lattice type. Chain type. (c) Figure 3. Self-assembly by the Fracta robot. Homogeneous distributed control by onboard processors is installed on the Fracta. MARCH 2007 IEEE Robotics & Automation Magazine 73

4 bipartite module does not need a connection mechanism between the parts. However, it loses symmetry in exchange, making SR planning difficult. (It is possible to make a metamodule, which is an isotropic group of modules, by combining several anisotropic modules.) A typical SR robot in this category is the Molecule [21]. Chain-Type SR Robots The chain type SR robot is aimed for dexterous motion [Figure 2]. These basically serial-link robots look like snakes; a combination of an actuated joint and connectors Figure 4. ATRON robot. (Courtesy of USD.) Figure 5. Chain-type SR robots (Polypod). (Courtesy of PARC.) between the links comprises a module. If some branch modules with more than two connectors are added to the system, an SR robot with multiple limbs can be built. Because of its ability to change the length and number of limbs by SR according to given tasks, application for maintenance task of space structure have been studied [28]. The first robot in this category was the Polypod (1993) developed at Stanford [23]. This robot was not exactly an SR robot because the connector was not automated. (Its successor, the Polybot (Figure 5) [24], has automated docking function.) Compared to the lattice-type robot, the chain-type is less symmetrical; thus, it can be built with fewer actuators and connectors and has high torque/weight ratio. Naturally, motion generation on the chain-type robot is easier than that of the lattice-type. However, self-reconfiguration is difficult because multi-dof cooperative manipulation and precise measurement of relative position/orientation between the connectors is necessary. Therefore, SR experiments in 3-D space that require control of at least six axes have not yet been conducted for chaintype robots. Some reported results are basically 2-D using the floor to restrict motion; even so, they require remote control by a human operator or they are very slow, taking minutes for a single reconnection step [24]. Polybot, CONRO [25] [27], and RBR [28] are typical SR robots in this category. A Hybrid-Type SR Robot: M-TRAN As mentioned in the previous section, both types of SR robots have some peculiar drawbacks resulting from their lattice or chain nature. We have proposed a hybrid-type SR robot, called M-TRAN (modular transformer), shown in Figure 6 to solve the problems [29] [31]. The hybrid module design of the M-TRAN realizes dexterous motion generation while maintaining the properties of lattice-type modules. As a lattice-type module, the M-TRAN module is classified as a combination of two cubic modules. It has two parallel rotational actuators embedded in a link between two semi-cylindrical cubes. Each cube has three connection Link Connection Active (Male) Block 9 Passive (Female) Block Module Rotation 13 Link Active (Male) Block 180 Rotation Passive (Female) Block 18 Module Neighbor Module Figure 6. M-TRAN robot. 74 IEEE Robotics & Automation Magazine MARCH 2007

5 surfaces, with male (active) surfaces on one cube and female (passive) surfaces on the other. There are two different types of connection mechanisms for the M-TRAN series. M- TRAN I and II adopt magnetic connection, and M-TRAN III has a hooking mechanism. Important features of these connection mechanisms are that two surfaces are connected in any four relative orientations and that they are completely retracted into the surface when disconnecting. These properties along with the parallel axis of one module enable a single module or two to move and reconnect without collision, as is the property of lattice-type SR robots. The semi-cylindrical cube shell innately gives the modules structural strength; therefore, they are stackable and can support a large lattice structure without power consumption. Serially linked M-TRAN modules can be regarded as chain-type, thus having a high ability to generate motion. In Table 2, three types of SR robots are compared. Self-Reconfiguration and Motion Control of M-TRAN In planning and controlling self-reconfiguration, we usually have to consider several different types of constraints, such as avoiding collisions, dealing with the torque limit of servomotors, and maintaining the connectivity of the robot. The M-TRAN s anisotropic symmetry makes the problem more challenging, and it is generally difficult to find a self-reconfiguration path between current and desired configurations. An SR robot is not a robot designed to perform a specific task but a system that develops into various types of robots and executes a variety of tasks. One approach to reduce the difficulty is to limit the structure. We have achieved fully automatic self-reconfiguration for a class of periodic lattice structures (e.g., linear structures with bends and planar and 3-D structures) (Figure 7). For instance, a linear structure can move by sending tail modules to the head along the structure, using predefined sequences of local self-reconfiguration. In order to obtain a self-reconfiguration path between specific configurations such as a legged configuration and a snake configuration, we usually need some human intervention. We have developed a GUI-based simulator for this purpose. This simulator displays the possible self-reconfiguration segments (a few steps of reconfiguration) under the constraints and executes kinematics/dynamics simulation to evaluate designed segments. The results are recorded in a motion script file. By using this simulator, a human planner can efficiently design appropriate reconfiguration paths and motion Table 2. Comparison of lattice, chain, and hybrid robots. SR Motion Generation No. of Actuator No. of Connectors Homogeneity Symmetry Lattice easy hard many many homo isotropic Chain hard easy few few hetero anisotropic Hybrid easy easy few few homo anisotropic (4) (3) (2) (1) Figure 7. Amoeba-like transportation. MARCH 2007 IEEE Robotics & Automation Magazine 75

6 sequences. Figure 8 presents an example of an SR sequence for the M-TRAN designed with the use of a simulator. Figure 8. Metamorphosis from a four-legged structure to a linear structure. Figure 9. Robustness of CPG control: CPG control is able to automatically adjust motion patterns to local and global fluctuations. Local periodic control by CPG is regulated by a stable rolling motion. External disturbances such as additional load or change of friction coefficient or inclination of the floor are successfully absorbed by CPG control. Figure 10. Demonstration of the M-TRAN III at World Expo We also study motion control for M-TRAN as a chaintype robot. We need a generic and more flexible method allowing frequent configuration changes. It is also necessary to consider the modular nature of the SR system. We have developed a design method that combines a central pattern generator (CPG) and a genetic algorithm (GA). In this method, a CPG, or a neural oscillator, drives every actuated joint at the same frequency. These CPGs are randomly networked initially, thus generating only a random motion in the beginning. The GA optimizes the connection matrix among the CPG network, where performance (speed and efficiency of transportation) is evaluated by the M-TRAN simulator [32]. This method gives efficient and stable gait patterns suitable for various configurations without manual tuning (Figure 9). Challenges Facing SR Robotics Hardware Issues The most important part of the hardware design of SR robots is the fundamental properties of a module, especially the shape of the module and the geometrical arrangement of actuator axes and connection. In the M-TRAN design, we successfully created a simple form but sacrificed its isotropy. We need to make the module more functional without further complication of the hardware. In order to find such a solution, we have to design more systematically. Specifically, we need a method of quantitatively estimating self-reconfiguration functionalities, based on the geometric properties of a module. Another important issue is the reliability of the hardware. Major factors restricting the size of the SR system (the number of working modules) are mechanical reliability, electrical reliability, and high cost (typically, several thousand dollars per module in current models). Electrical reliability and high cost can be resolved by mass production. However, the mechanical reliability of the mechanism (the ability to work regardless of the number of modules) cannot be predicted until the system is actually constructed. One crucial point in mechanical design is the connection mechanism. M-TRAN III has a mechanical hooking connector that is faster, stronger, and more reliable than the magnetic connectors in previous versions Figure 10 shows the M-TRAN III being demonstrated at World Expo 2005 in Aichi, Japan. Fifty M-TRAN III modules were produced for demonstration. During the 11- day exhibition, self-reconfiguration and locomotion of M- TRAN were successfully displayed more than 200 times in front of a large crowd of visitors from all over the world. See is/dsysd/mtran3/ for more information. 76 IEEE Robotics & Automation Magazine MARCH 2007

7 Control System Issues The control system of an SR robot can be regarded as an information network of many module processors. In an SR system where the network structure is dynamic, various operations must be realized: differentiation of modules or role sharing, synchronization among modules for coherent motion generation, and high-level decision making in a distributed manner. One of the possible solutions to realize these operations is to introduce hierarchy into the system (e.g., a system maintenance layer, a morphology control layer, a motion control layer, and a decision-making layer). Each layer or each process in the layer requires a different communication capability among modules, such as speed and distance. Animals have centralized nerve systems as well as chemical inter-cellular communication in order to realize fast motion generation. Likewise, our control system needs different types of communication devices and protocols with different properties, such as peer-to-peer communication using ID/broadcasting and a tethered/radio communication channel. Algorithm Issues SR robots, of course, are not almighty. The range of tasks executable by combined modules is restricted by module granularity and functionality. We need to find feasible morphology/ motion within a given resolution of the modules. Simulation is necessary in the search for a solution. In the M-TRAN system, motion generation is automated by the use of forward kinematics/dynamics simulation and genetic algorithm; however, the concurrent search of morphology and motion is quite difficult because of explosive computational complexity. To achieve a breakthrough, both more powerful computers and more profound understanding of the SR system (such as a new description method of morphology and motion) are essential. Another topic in the area of algorithms is the connectivity (wholeness) issue. In most SR robots developed thus far, the robot is assumed to be a single cluster with a fixed number of modules. The robot s boundary is naturally defined, and its behavior is given for the cluster. However, if we consider a more general case (i.e., two separated robots merge into one, or one robot is divided into two individual robots, or a robot self-reproduces itself by collecting scattered modules), the system s boundary can no longer be defined. The fundamental definitions of system architecture should be extended to deal with the full flexibility of the SR system. The Future of SR Robots SR robotics research can take the application-oriented direction or the fundamental-science-oriented direction. The application-oriented direction will emphasize research for practical application of the SR system, especially that utilizing real-time changeability of morphology and functionality. Selfreconfiguration is most suitable for applications in which robustness and multifunctionality rather than optimality are required; examples include tasks in extreme environments where human operators have no access. Self-reconfiguration ability will pay for itself in such applications. An SR robot is able to adapt not only to the changes in its external environment but also to the changes in its internal environment, like functional disturbances or failures in modules. Exploration tasks on an uncharted planet are a typical example. We do not know the surface condition of the planet; it may be hard and flat; it may be rocky, or sandy and easy to crumble. Our robot must move over the surface, despite its condition. It must not die even if its leg is caught in a crevasse or buried under debris: an SR robot can change its shape and escape from such situations. It is also capable of repairing itself. (SR robots are scalable; thus, total functionality will be gracefully degraded after they cut off damaged modules.) Various tasks for SR robots will be found in orbit, in the deep sea, and in areas disaster-stricken by radiation or toxic chemicals. Education and arts are other areas of application. SR robots are practical instantiations of modular systems. They are ideal material for robotics and systems science education. SR robots themselves are interesting objects. Because they evoke various emotions through their reconfiguration and motion, they may be used as tools of expression for visual artists. The science-oriented direction regards SR robotics as a part of fundamental science. It is the design science of artifacts, based on a deep understanding of the construction principle of living systems. SR robot systems have ultimate autonomy in the sense that they rise above the perpetual cycle of conventional engineering (design, production, and employment). Historically, artifacts have never reached this level of autonomy. Compared to living systems that have evolved through eons of repetition of reproduction and development processes, SR systems are artifacts in which every factor can be manipulated. We can rapidly improve such artifacts by acceleration of evolution, involving both large-scale simulations in the virtual world and hardware experiments in the real world. When the principle of SR robotics becomes clear, SR robots will cease to be merely biologically inspired artifacts and become super biological robots. Acknowledgments This work has been supported by the Ministry of Education, Culture, Sports, Science and Technology, Japan (Grants-in- Aid for Scientific Researches No ). Keywords Self-reconfigurable robots, modular robotics, shape shifting, self-assembly, self-repair, evolutionary robotics. MARCH 2007 IEEE Robotics & Automation Magazine 77

8 References [1] P. Liang and G. Beni, Robotic morphogenesis, in Proc. IEEE Int. Conf. Robotics and Automation, 1995, pp [2] G. Chirikjian, Kinematics of a metamorphic robotic system, in Proc. IEEE Int. Conf. Robotics and Automation, 1994, pp [3] S. Murata, H. Kurokawa, and S. Kokaji, Self-assembling machine, in Proc. IEEE Int. Conf. Robotics and Automation, 1994, pp [4] G. Chirikjian, A. Pamecha, and I. Ebert-Uphoff, Evaluating efficiency of self-reconfiguration in a class of modular robots, J. Robot Syst., vol. 13, no. 5, pp , [5] A. Pamecha and I. Ebert-Uphoff, Useful metrics for modular robot motion planning, IEEE Trans. Robot. Automat., vol. 13, pp , Aug [6] K. Tomita, S. Murata, H. Kurokawa, E. Yoshida, and S. Kokaji, A selfassembly and self-repair method for a distributed mechanical system, IEEE Trans. Robot. Automat., vol. 15, pp , Dec [7] S. Murata, E. Yoshida, H. Kurokawa, K. Tomita, and S. Kokaji, Selfrepairing mechanical system, Autonomous Robots, vol. 10, no. 1, pp. 7 21, [8] J. von Neumann, Theory of Self-Reproducing Automata. Urbana, IL: Univ. Illinois Press, [9] K. Sims, Evolving 3D morphology and behavior by competition, in Proc. Artificial Life IV, 1994, pp [10] T. Fukuda and S. Nakagawa, Dynamically reconfigurable robotic system, in Proc. IEEE Int. Conf. Robotics and Automation, 1988, vol. 3, pp [11] K. Hosokawa, T. Tsujimori, T. Fujii, H. Kaetsu, H. Asama, Y. Kuroda, and I. Endo, Self-organizing collective robots with morphogenesis in a vertical plane, in Proc. IEEE Int. Conf. Robotics and Automation, 1998, pp [12] N. Inou, K. Minami, and M. Koseki, Group robots forming a mechanical structure-development of slide motion mechanism and estimation of energy consumption of the structural formation, in Proc. IEEE Int. Symp. Computational Intelligence in Robotics and Automation, 2003, vol. 2, pp [13] H. Kurokawa, S. Murata, E. Yoshida, K. Tomita, and S. Kokaji, A three-dimensional self-reconfigurable system, Adv. Robot., vol. 13, no. 6, pp , [14] R. Fitch, D. Rus, and M. Vona, A basis for self-repair using crystaline modules, in Proc. Intelligent Autonomous Systems Conf. (IAS-6), 2000, pp [15] D. Rus and M. Vona, Self-reconfiguration planning with compressible unit modules, in Proc. IEEE Int. Conf. Robotics and Automation, 1999, pp [16] D. Rus and M. Vona, Crystalline robots: Self-reconfiguration with compressible unit modules, Autonomous Robots, vol. 10, no. 1, pp , [17] N. Inou, H. Kobayashi, and M. Koseki, Development of pneumatic cellular robots forming a mechanical structure, in Proc. Int. Conf. Control, Automation, Robotics and Vision, 2002, pp [18] M. Yim, J. Lamping, E. Mao, and J.G. Chase, Rhombic dodecahedron shape for self-assembling robots, Xerox PARC SPL, Tech. Rep. P , [19] M. Jorgensen, E. Ostergaard, and H. Lund, Modular ATRON: Modules for a self-reconfigurable robot, in Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems, 2004, pp [20] D.J. Christenssen, E. Ostergaard, and H. Lund, Metamodule control for the ATRON self-reconfigurable robotic system, in Proc. Intelligent Autonomous Systems (IAS-8), 2004, pp [21] K. Kotay, D. Rus, M. Vona, and C. McGray, The self-reconfiguring robotic molecule, in Proc. IEEE Int. Conf. Robotics and Automation, 1998, pp [22] C. Unsal, H. Kiliccote, and P. Kohsla, A modular self-reconfigurable bipartite robotic system: implementation and motion planning, Autonomous Robots, vol. 10, no. 1, pp , [23] M. Yim, A reconfigurable modular robot with many modes of locomotion, in Proc. JSME Int. Conf. Advanced Mechatronics, 1993, pp [24] M. Yim, Y. Zhang, K. Roufas, D. Duff, and C. Eldershaw, Connecting and disconnecting for chain self-reconfiguration with Polybot, in Proc. IEEE/ASME Trans. Mechatron, Dec. 2002, vol. 7, pp [25] A. Castano and P. Will, Mechanical design of a module for autonomous reconfigurable robots, in Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems, 2000, pp [26] W. Shen, B. Salemi, and P. Will, Hormone-inspired adaptive communication and distributed control for CONRO self-reconfigurable robots, IEEE Trans. Robot. and Automat., vol. 18, pp , Oct [27] K. Stoy, W. Shen, and P. Will, Using role based control to produce locomotion in chain-type self-reconfigurable robot, IEEE/ASME Trans. Mechatron., vol. 7, pp , Dec [28] H. Ueno, Y. Wakabayashi, Y. Ohkami, S. Matunaga, R. Hayashi, and T. Yoshida, Ground testbed of a reconfigurable brachiating space robot, Advanced Robot., vol. 14, no. 5, pp , [29] S. Murata, E. Yoshida, H. Kurokawa, K. Tomita, and S. Kokaji, Concept of self-reconfigurable modular robotic system, J. AI Eng., vol. 15, no. 4, pp , [30] S. Murata, E. Yoshida, A. Kamimura, H. Kurokawa, K. Tomita, and S. Kokaji, M-TRAN: Self-reconfigurable modular robotic system, IEEE/ASME Trans. Mechatron., vol. 7, pp , Dec [31] E. Yoshida, S. Murata, A. Kamimura, K. Tomita, H. Kurokawa, and S. Kokaji, A self-reconfigurable modular robot: reconfiguration planning and experiments, Int. J. Robot. Res., vol. 21, no.10, pp , [32] A. Kamimura, H. Kurokawa, E. Yoshida, S. Murata, K. Tomita, and S. Kokaji, Automatic locomotion design and experiments for a modular robotic system, IEEE/ASME Trans. Mechatron., vol. 10, pp , Sept [33] E. Yoshida, S. Murata, K. Tomita, H. Kurokawa, and S. Kokaji, Experiment of self-repairing modular machine, in Distributed Autonomous Robotic Systems 3, T. Luth, P. Dario, H. Worn, Eds. New York: Springer, 1998, pp [34] V. Zykov, E. Mytilinaios, B. Adams, and H. Lipson, Self-reproducing machines, Nature, vol. 435, no. 7038, pp , Satoshi Murata received his B.E., M.E, and Dr. Eng. in aeronautical engineering from Nagoya University, Japan. He joined the Mechanical Engineering Laboratory, National Institute of Advanced Industrial Science and Technology (AIST), Ministry of International Trade and Industry (Japan), in 1987 and became a senior researcher in Since 2001 he has been an associate professor of computational intelligence and systems science at the Tokyo Institute of Technology, Japan. His current interests include distributed autonomous systems, modular robotics, and molecular robotics. Haruhisa Kurokawa received the B.E and M.E in precision machinery engineering and the Dr. Eng. in aero- and astronautical engineering from the University of Tokyo in 1978, 1981, and 1997, respectively. He is currently the head of the Distributed System Design Research Group, Intelligent Systems Institute, National Institute of Advanced Industrial Science and Technology (AIST), Japan. His main research subjects are kinematics of mechanisms, distributed autonomous systems, and nonlinear control. Address for Correspondence: Satoshi Murata, Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology, G3-53, 4259 Nagatsuda, Midori, Yokohama, Japan. murata@dis.titech.ac.jp. 78 IEEE Robotics & Automation Magazine MARCH 2007