INVISIBLE COMPUTING

tain a 3D shape, • communicate with other catoms Programmable in an ensemble, and • compute state information with possible assistance from other Matter catoms in the ensemble. In the preliminary design, each Seth Copen Goldstein, Carnegie Mellon University catom is a unit with a CPU, a network device, a single-pixel display, one or Jason D. Campbell, Intel Research Pittsburgh more sensors, a means of locomotion, Todd C. Mowry, Intel Research Pittsburgh and Carnegie Mellon University and a mechanism for adhering to other catoms. Although this sounds like a

n the past 50 years, computers have shrunk from room-size main- Researchers are using frames to lightweight handhelds. nanoscale technology I This fantastic miniaturization is primarily the result of high-vol- to create microrobot ume nanoscale manufacturing. ensembles for modeling While this technology has predomi- nantly been applied to logic and mem- 3D scenes. ory, it’s now being used to create ad- vanced microelectromechanical sys- tems using both top-down and bot- movement, visual appearance, sound, microrobot, we believe that imple- tom-up processes. and tactile qualities. menting a completely autonomous One possible outcome of continued microrobot is unnecessarily complex. progress in high-volume nanoscale Synthetic reality Instead, we take a cue from cellular assembly is the ability to inexpensively One application of an ensemble, reconfigurable robotics research to produce millimeter-scale units that comprised of millions of cooperating simplify the individual robot modules integrate computing, sensing, actua- robot modules, is programming it so that they are easier to manufacture tion, and locomotion mechanisms. A to self-assemble into arbitrary 3D using high-volume methods. collection of such units can be viewed shapes. Our long-term goal is to use as a form of programmable matter. such ensembles to achieve synthetic Ensemble principle reality, an environment that, unlike Realizing this vision requires new CLAYTRONICS virtual reality and augmented reality, ways of thinking about massive num- The Claytronics project (www.cs. allows for the physical realization bers of cooperating millimeter-scale cmu.edu/~claytronics) is a joint effort of all computer-generated objects. units. Most importantly, it demands of researchers at Carnegie Mellon Hence, users will be able to experience simplifying and redesigning the soft- University and Intel Research Pitts- synthetic reality without any sensory ware and hardware used in each catom burgh to explore how programmable augmentation, such as head-mounted to reduce complexity and manufactur- matter might change the computing displays. They can also physically ing cost and increase robustness and experience. Similar to how audio and interact with any object in the system reliability. For example, each catom video technologies capture and repro- in a natural way. must work cooperatively with others duce sound and moving images, respec- in the ensemble to move, communi- tively, we are investigating ways to Catoms cate, and obtain power. reproduce moving physical 3D objects. Programmable matter consists of a Consequently, our designs strictly The idea behind claytronics is nei- collection of individual components, adhere to the ensemble principle: ther to transport an object’s original which we call claytronic atoms or A robot module should include only instance nor to recreate its chemical catoms. Catoms can enough functionality to contribute composition, but rather to create a to the ensemble’s desired functionality. physical artifact using programmable • move in three dimensions in rela- Three early results of our research matter that will eventually be able to tion to other catoms, each highlight a key aspect of the mimic the original object’s shape, • adhere to other catoms to main- ensemble principle: easy manufactura-

June 2005 99 Invisible Computing

and volume exceed those of the robots themselves. To provide sufficient energy to each catom without incurring such a weight and volume penalty, we’re developing methods for routing energy from an external source to all catoms in an ensemble. For example, an ensemble could tap an environmental power source, such as a special table with pos- itive and negative electrodes, and route that power internally using catom-to- catom connections. To simplify manufacturing and Figure 1. Claytronic atom prototypes. Each 44-mm-diameter catom is equipped with 24 accelerate movement, we believe it’s electromagnets arranged in a pair of stacked rings. necessary to avoid using intercatom connectors that can carry both supply bility, powering million-robot ensem- coils varies roughly with the inverse and ground via separate conductors bles, and surface contour control with- cube of distance, whereas the flux due within the connector assembly. Such out global motion planning. to a given coil varies with the square of complex connectors can significantly the scale factor. Hence, the potential increase reconfiguration time. HIGH-VOLUME force generated between two catoms For example, in previously con- MANUFACTURABILITY varies linearly with scale. Meanwhile, structed modular robotic systems such Some catom designs will be easier to mass varies with the cube of scale. as the Palo Alto Research Center’s produce in mass quantity than others. These relationships suggest that a PolyBot (www2.parc.com/spl/projects/ Our present exploration into the design 10-fold reduction in size should trans- modrobots/chain/polybot/index.html) space investigates modules without late to a 100-fold increase in force rel- and the Dartmouth Robotics Lab’s moving parts, which we see as an inter- ative to mass. Energy consumption and Molecule (www.cs.dartmouth.edu/ mediate stage to designing catoms suit- supply will still be an issue, but given ~robotlab/robotlab/robots/molecule/ able for high-volume manufacturing. sufficient energy, smaller catoms will index.html) it can take tens of seconds In our present macroscale (44-mm have an easier time lifting their own or even minutes for a robot module to diameter), cylindrical prototypes, shown weight and that of their peers, as well uncouple from its neighbor, move to in Figure 1, each catom is equipped with as resisting other forces involved in another module, and couple with that 24 electromagnets arranged in a pair of holding the ensemble together. newly proximal module. stacked rings. To move, a pair of catoms Our finite-element electromagnetic- In contrast, our present unary-con- must first be in contact with another physical simulations on catoms of dif- nector-based prototypes can “dock” in pair. Then, they must appropriately ferent sizes appear to confirm this less than 100 ms because no special energize the next set of magnets along approximation and closely match our connector alignment procedure is each of their circumferences. empirical measurements of magnetic required. This speed advantage isn’t These prototypes demonstrate high- force in the 44-mm prototypes. We’re free, however: A genderless unary con- speed reconfiguration—approximately also studying programmable nanofiber nector imposes additional operational 100 ms for a step-reconfiguration adhesive techniques that are necessary complexity in that each catom must involving uncoupling of two units, to eliminate the static power drain obtain a connection to supply from movement from one pair of contact when robots are motionless, while still one neighbor and to ground from a dif- points to another, and recoupling at the maintaining a strong bond. ferent neighbor. next pair of contact points. At present, Several members of the Claytronic they don’t yet incorporate any signifi- POWERING MICROROBOT team have recently developed power cant sensing or autonomy. ENSEMBLES distribution algorithms that satisfy The current prototypes can only Some energy requirements, such these criteria. These algorithms require overcome the frictional forces opposing as effort to move versus gravity, scale no knowledge of the ensemble config- their own horizontal movement, but with size. Others, such as communi- uration—lattice spacing, ensemble downscaling will improve the force cation and computation, don’t. As size, or shape—or power-supply loca- budget substantially. The resulting force microrobots (catoms) are scaled tion. Further, they require no on-catom from two similarly energized magnet down, the onboard battery’s weight power storage.

100 Computer SHAPE CONTROL WITHOUT GLOBAL MOTION PLANNING A hole approaches a but rather than the edge consumes contracting edge, reflecting, the hole. Classical approaches to creating an arbitrary shape from a group of mod- ular robots involve motion planning through high-dimensional search or gradient descent methods. However, in the case of a million-robot ensemble, (a) global search is unlikely to be tractable. An edge selected creates one or more launching them into Even if a method could globally plan to expand holes, the ensemble. for the entire ensemble, the communi- cations overhead required to transmit individualized directions to each mod- ule would be very high. In addition, a global plan would break down in the face of individual unit failure. To address these concerns, we’re (b) developing algorithms that can control shape without requiring extensive Figure 2. Hole motion. Edges can (a) contract by consuming a hole or (b) expand by planning or communication. While creating a hole, purely under local control. this work is just beginning, Claytronics researchers have had early success outside of the ensemble, pulling in the volume manufacturing techniques using an approach inspired by semi- surface at that location. Similarly, bridged a similarly large gulf in cost and conductor device physics. expanding edges create holes and inject physical scaling in computer hardware. This approach focuses on the them into the ensemble, pushing its Achieving the Claytronics vision motion of holes rather than that of contour out in the corresponding local won’t be straightforward or quick, but robots per se. Given a uniform hexag- region. by taking on some of the problems onal-packed plane of catoms, a hole is Importantly, all edge contouring and associated with operating and building a circular void due to the absence of hole motion can be accomplished using these ensembles, we hope to advance seven catoms. Such a seven-catom hole local rules, and the overall shape of an the state of the art in modular recon- can migrate through the ensemble by ensemble can be programmed purely figurable microrobotics and encourage appropriate local motion of the adja- by communicating with the catoms at others to undertake related research. I cent catoms. the edges. Hence, we use probabilistic Holes migrate through the ensemble methods to achieve a deterministic Seth Copen Goldstein is an associate as if moving on a frictionless plane, and result. Our initial analyses of the cor- professor in the Computer Science bounce back at the ensemble’s edges. responding 3D case suggest surface Department at Carnegie Mellon Uni- Just as bouncing gas molecules exert contour control will be possible via a versity. Contact him at seth.goldstein@ pressure at the edges of a balloon, similar algorithm. cs.cmu.edu. bouncing holes interact frequently with each edge of the ensemble without the Jason Campbell is a senior researcher need for global control. As Figure 2 ur initial research results suggest at Intel Research Pittsburgh. Contact illustrates, edges can contract by con- it may be possible to construct, him at [email protected]. suming a hole or expand by creating a O power, and control large microro- hole, purely under local control. bot ensembles to model 3D scenes. Todd C. Mowry is director of Intel We initiate shape formation by “fill- While many difficult problems remain, Research Pittsburgh and an associate ing” the ensemble with holes. Each hole successful implementation of a dynamic professor in the Computer Science receives an independent, random veloc- physical rendering system could open Department at Carnegie Mellon Uni- ity and begins to move around. A shape the door to a new era of human- versity. Contact him at [email protected]. goal specifies the amount each edge computer interface design and new edu. region must either contract or expand applications. to match a desired target shape. Economic feasibility also poses a high A hole that hits a contracting edge is bar to the manufacture and deployment Editor: Bill Schilit, Intel Research consumed. In effect, the empty space of multimillion-robot ensembles. How- Seattle; [email protected] that constitutes the hole moves to the ever, the innovative application of high-

June 2005 101