Here's how my recent article for IEEE Spectrum started off, with some more references, info and links.
If science is to reach beyond a myopic fixation on incremental advances, it may need bold and visionary dreams that border on myth-making. There are plenty of those in the field called programmable matter, which aims to blend micro- and nanotechnology, robotics and computing to produce substances that change shape, appearance and function at our whim.
The dream is arrestingly illustrated in a video produced by a team at Carnegie Mellon University in Pittsburgh. Executives sit around a table watching a sharp-suited sales rep make his pitch. From a vat of grey gloop he pulls a perfectly rendered model of a sports car, and proceeds to reshape it with his fingers. With gestures derived from touchscreen technology, he raises or flattens the car’s profile and adjusts the width of the headlamps. Then he changes the car from silver-grey to red, the “atoms” twinkling in close-up with Disney-movie magic as their color shifts.
This kind of total mastery over matter is not so different from the alchemist’s dream of transmuting metals, or in contemporary terms, the biologist’s dream of making life infinitely malleable through synthetic biology. But does the fantasy – it’s little more at present – bear any relation to what can be done?
Because of its affiliation with robotic engineering and computer science, the idea of programmable matter is often attributed to a paper published in 1991 by computer scientists Tommaso Toffoli and Norman Margolus of the Massachusetts Institute of Technology, who speculated about a collection of tiny computing objects that could sense their neighbors and rearrange themselves rather like cellular automata . But related ideas were developed independently in the early 1990s by the chemistry Nobel laureate Jean-Marie Lehn, who argued that chemistry would become an information science by using the principles of spontaneous self-assembly and self-organization to design molecules that would assemble themselves from the bottom up into complex structures . Lehn’s notion of “informed matter” was really nothing less than programmable matter at the atomic and molecular scale.
Lehn’s own work since the 1960s helped to show that the shapes and chemical structures of molecules could predispose them to unite into large-scale organized arrays that could adapt to their circumstances, for example responding to external signals or having self-healing abilities. Such supramolecular (“beyond the molecule”) self-assembly enables biomolecules to become living cells, which need no external instructions to reconfigure themselves because their components already encode algorithms for doing so. In some ways, then, living organisms already embody the aspirations of programmable matter.
Yet in the information age, it is we who do the programming. While living cells are often said (a little simplistically) to be dancing to the evolutionarily shaped program coded in their genomes, technologies demand that we bring matter under our own direct control. It’s one thing to design molecules that assemble themselves, quite another to design systems made from components that will reconfigure or disassemble at the push of a button. The increasingly haptic character of information technology’s interfaces encourages a vision of programmable matter that is responsive, tactile, even sensual.
Many areas of science and technology have come together to enable this vision. Lehn’s supramolecular chemistry is one, and nanotechnology – its extreme miniaturization and interest in “bottom-up” self-organizing processes – is another. Macroscopic robotic engineering has largely abandoned the old idea of humanoid robots, and is exploring machines that can change shape and composition according to the task in hand . To coordinate large numbers of robotic or information-processing devices, centralized control can be cumbersome and fragile; instead, distributed computing and swarm robotics rely on the ability of many interacting systems to find their own modes of coordination and organization . Interacting organisms such as bacteria and ants provide an “existence proof” that such coordination is sometimes best achieved through this kind of collective self-organization. Understanding such emergent behaviour is one of the central themes in the science of complex systems, which hopes to harness it to achieve robustness, adaptability and a capacity for learning.
Meanwhile, thanks to the shrinking of power sources and the development of cheap, wireless radio-frequency communications for labelling everything from consumer goods to animals for ecological studies, robotic devices can talk to one another even at very small scales. And making devices that can be moved and controlled without delicate and error-prone moving parts has benefitted immensely from the development of smart materials that can respond to their environment and to external stimuli by, for example, changing their shape, color or electrical conductivity.
In short, the ideas and technologies needed for programmable matter are already here. So what can we do with them?
Seth Goldstein and his team at Carnegie Mellon, in collaboration with Intel Research Pittsburgh, were among the first to explore the idea seriously. “I’ve always had an interest in parallel and distributed systems”, says Goldstein. “I had been working in the area of molecular electronics, and one of the things that drew me into the field was a molecule called a rotaxane that, when subjected to an electric field, would change shape and as a result change its conductivity. In other words, changing the shape of matter was a way of programming a system. I got to thinking about what we could do if we reversed the process: to use programming to change the shape of matter.”
The Carnegie Mellon group envisions a kind of three-dimensional, material equivalent of sound and visual reproduction technologies, in which millions of co-operating robot modules, each perhaps the size of a dust grain, will mimic any other object in terms of shape, movement, visual appearance, and tactile qualities. Ultimately these smart particles – a technology they call Claytronics  – will produce a “synthetic reality” that you can touch and experience without any fancy goggles or gloves. From a Claytronics gloop you might summon a coffee cup, a spanner, a scalpel.
“Any form of programmable matter which can pass the ‘Turing test’ for appearance [looking indistinguishable from the real thing] will enable an entire new way of thinking about the world”, say Goldstein. “Applications like injectable surgical instruments, morphable cellphones, 3D interactive life-size TV and so on are just the tip of the iceberg.”
Goldstein and colleagues call the components of this stuff “catoms” – Claytronic atoms, which are in effect tiny spherical robots that move, stick together, communicate and compute their own location in relation to others. Each catom would be equipped with sensors, color-change capability, computation and locomotive agency. That sounds like a tall order, especially if you’re making millions of them, but Goldstein and colleagues think it should be achievable by stripping the requirements down to the bare basics.
The prototype catoms made by the Pittsburgh researchers since the early 2000s were a modest approximation to this ambitious goal: squat cylinders about 44 mm across, their edges lined with rows of electromagnets that allow them to adhere in two-dimensional patterns. By turning the magnets on and off, one catom could ‘crawl’ across another. Using high-resolution photolithography, the Carnegie Mellon team has now managed to shrink the cylindrical catoms to the sub-millimetre scale, while retaining the functions of power transfer, communication and adhesion. These tiny catoms can’t yet move, but they will soon, Goldstein promises.
Electromagnetic coupling might ultimately not be the best way to stick them, however, because it drains power even when the devices are static. Goldstein and colleagues have explored the idea of making sticky patches from carpets of nanofibers, like those on a gecko’s foot, that adhere due to intermolecular forces. But at present Goldstein favors electrostatics as the best force for adhesion and movement. Ideally the catoms will be powered by harvesting energy from the environment – drawing it from an ambient electric field, say – rather than carrying on-board power supplies.
One of the big challenges is figuring out where each catom has to go in order to make the target object. “The key challenge is not in manufacturing the circuits, but in being able to program the massively distributed system that will result from putting all the units together into an ensemble”, says Goldstein. Rather than drawing up a global blueprint, the researchers hope that purely by using local rules, where each catom simply senses the positions of its near neighbors, the ensemble can find the right shape. Living organisms seem to work this way: the single-celled slime mold Dictyostelium discoideum, for example, aggregates under duress into a mushroom-shaped multicellular body without any ‘brain’ to plan it. This strategy means the catoms must communicate with one another. The Carnegie Mellon researchers plan to explore both wireless technologies for remote control, and electrostatic interactions for nearest-neighbour sensing.
To be practical, this repositioning needs to be fast. Goldstein and colleagues think that an efficient way to produce shape changes might be to fill the initial catom “blob” with little voids, and then shift them around to achieve the right contours. Small local movements of adjacent catoms are all that’s needed to move holes through the medium, and if they reach the surface and are expelled like bubbles, the overall volume shrinks. Similarly, the material can be expanded by opening up new bubbles at the surface and engulfing them.
At MIT, computer scientist Daniela Rus and her collaborators have a different view of how smart, sticky ‘grains’ could be formed into an object. Their “smart sand” would be a heap of such grains that, by means of remote messages and magnetic linkages, will stick selectively together so that the target object emerges like a sculpture from a block of stone. The unused grains just fall away. Like Goldstein, Gilpin and colleagues have so far explored prototypes on a larger scale and in two dimensions, making little units the size of sugar cubes with built-in microprocessors and electromagnets on four faces. These units can communicate with each other to duplicate a shape inserted into the 2D array. The smart grains that border the master shape recognize that they are at the edge, and send signals to others to replicate this pixellated mould and the object that lies within it .
Rus and her collaborators have hit on an ingenious way to make these ‘grains’ move. They have made larger cubes called M-blocks, several centimeters on each side, which use the momentum of flywheels spinning at up to 20,000 r.p.m. to roll, climb over each other and even leap through the air . When they come into contact, the blocks can be magnetically attached to assemble into arbitrary shapes – at present determined by the experimenters, although their plan is to develop algorithms that let the cubes themselves decide where they need to go.
M-blocks in action
Programmable matter doesn’t have to be made from an army of hard little units. Hod Lipson at Cornell University and his colleagues think that it should be possible to create “soft robots” that can be moulded into arbitrary shapes from flexible smart materials that change their form in response to external signals.
“Soft robotics” is already well established. Shape-memory alloys, which bend or flex when heated or cooled, can provide the ‘muscle’ within the soft padding of a silicone body , for example, and polymeric objects can be made to change shape by inflating pneumatic compartments . What made the soft robot designed by Lipson and his colleague Jonathan Hiller particularly neat is that the actuation didn’t require a specific signal, but was built into the structure itself. They used evolutionary computer algorithms to figure out how to arrange tiny blocks of silicone foam rubber so that raising and reducing the air pressure caused the rubber to contract and expand in a way that made the weirdly-shaped assembly crawl across a surface .
Lipson and his coworkers have also devised algorithms that can mutate and optimize standardized components such as rods and actuators to perform particular tasks, and have coupled this design process to a 3D printer that fabricates the actual physical components, resulting in “machines that make machines.” They have been able to print not just parts but also power sources such as batteries, and Lipson says that his ultimate goal is to make robots that can “walk out of the printer”.
These are top-down approaches to programmable matter, emerging from existing developments in robotic technology. But there are alternatives that start from the bottom up: from nanoscale particles, or even molecules. For example, currently there is intense research on the behavior of so-called self-propelled or “living” colloids: particles perhaps a hundred or so nanometers across that have their own means of propulsion, such as gas released by chemical reactions at their surface. These particles can show complex self-organized behavior, such as crystalline patterns that form, break and explode . Controlling the resulting arrangements is another matter, but researchers have shown they can at least move and control individual nanoparticles using radiofrequency waves and magnetic fields. This has permitted wireless “remote control” of processes in living cells, such as the pairing of DNA strands , the triggering of nerve signals , and the control of insulin release in mice .
Nature programs its cellular matter partly by the instructions inherited in the DNA of the genome. But by exploiting the same chemical language of the genes – two DNA strands will pair up efficiently into a double helix only if their base-pair sequences are complementary – researchers have been able to make DNA itself a kind of programmable material, designed to assemble into specific shapes and patterns. In this way they have woven complex nanoscale DNA shapes such as boxes with switchable lids , nanoscale alphabetic letters  and even world maps . By supplying and removing ‘fuel strands’ that drive strand pairing and unpairing, it is possible to make molecular-scale machines that move, such as DNA ‘walkers’ that stride along guide strands . Eventually such DNA systems might be given the ability to replicate and evolve.
In ways like this, programmable matter seems likely to grow from the very small, as well as shrinking from robots the size of dimes. Goldstein says the basic idea can be applied to the building blocks of matter over all scales, from atoms and cells to house bricks. It’s almost a philosophy: a determination to make matter more intelligent, more obedient, more sensitive – in some respects, more alive.
Box: What might go wrong?
Isn’t there something a little sinister to this idea of matter that morphs and even mutates? What will the sculptors make? Can they be sure they can control this stuff? Here our fears of “animated matter” are surely shaped by old myths like that of the Jewish golem, a being fashioned from clay that threatened to overwhelm its creator.
The malevolence of matter that is infinitely protean is evident in imagery from popular culture, such as the “liquid robot” T-1000 of Terminator II. The prospect of creating programmable matter this sophisticated remains so remote, though, that such dangers can’t be meaningfully assessed. But in any event, Goldstein insists that “there’s no grey goo scenario here”, referring to a term nanotechnology pioneer Eric Drexler coined in his 1986 book Engines of Creation.
Drexler speculated about the possibility of self-replicating nanobots that would increase exponentially in number as they consumed the raw materials around them. This sparked some early fears that out-of-control nanotechnology could to turn the world into a giant mass of self-replicating gray sludge—a theme that appeared repeatedly in later works of science fiction, including Will McCarthy’s 1998 novel Bloom, Michael Crichton’s 2002 thriller Prey, and even in tongue-in-cheek fashion in a 2011 episode of Futurama.
But the real dangers may be ones associated more generically with pervasive computing, especially when it works by Wifi. What if such a system were hacked? It is one thing to have data manipulated online this way, but when the computing substrate is tangible stuff that reconfigures itself, hackers will gain enormous leverage for creating havoc.
Goldstein thinks, however, that some of the more serious problems might ultimately be of more of a sociological nature. Programmable matter is sure to be rather expensive, at least initially, and so the capabilities it offers might only widen the gap between those with access to new technology and those without. What’s more, innovations like this, as with today’s pervasive factory automation, threaten to render jobs in manufacturing and transport obsolete. So they will make more people unemployable, not because they lack the skills but because there will be nothing for them to do.
Of course, powerful new capabilities always carry the potential for abuse. You can see hints of that already in, say, the use of swarm robotics for surveillance, or in the reconfigurable robots that are being designed for warfare. Expect the dangers of programmable matter to be much like those of the Internet: when just about everything is possible, not all of what goes on will be good.
1. T. Toffoli & N. Margolus, Physica D 47, 263–272 (1991).
2. J.-M. Lehn, Supramolecular Chemistry (Wiley-VCH, Weinheim, 1994).
3. K. Gilpin & D. Rus, Robotics & Automation Magazine, IEEE 17(3), 38-55 (2010).
4. J. C. Barca & Y. A. Sekercioglu, Robotica 31, 345-359 (2012).
5. S. C. Goldstein, J. D. Campbell & T. C. Mowry, Computer 38, 99-101 (May 2005).
6. K. Gilpin, A. Knaian & D. Rus, IEEE Int. Conf. on Robotics and Automation, 2485-2492 (2010).
7. J. Romanishin, K. Gilpin & D. Rus, abstract, IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (November 2013).
8 .B. A. Trimmer, A. E. Takesian, B. M. Sweet, C. B. Rogers, D. C. Hake & D. J. Rogers, Proc. 7th Int. Symp. Technol. Mine Problem, Monterey, CA (2006).
9. F. Ilievski, A. D. Mazzeo, R. F. Shepherd, X. Chen & G. M. Whitesides, Angew. Chem. Int. Ed. 50, 1890–1895 (2011).
10. J. Hiller & H. Lipson, IEEE Trans. On Robotics 28(2), 457-466 (2012).
11. J. Palacci, S. Sacanna, A. P. Steinberg, D. J. Pine & P. M. Chaikin, Science 339, 936-940 (2013).
12. K. Hamad-Schifferli, J. J. Schwartz, A. T. Santos, S. Zhang & J. M. Jacobson, Nature 415, 152 (2002).
13. H. Huang et al., Nat. Nanotechnol. 5, 602 (2010).
14. S. A. Stanley et al., Science 336, 604 (2012).
15. E. S. Andersen et al., Nature 459, 73-76 (2009).
16. B. Wei, M. Dai & P. Yin, Nature 485, 623-626 (2012).
17. P. Rothemund, Nature 440, 297-302 (2006).
18. T. Omabegho, R. Sha & N. C. Seeman, Science 324, 67-71 (2009).