Google “grey goo”. You’ll find the definition, one of which is “Gray goo is a mass of small, destructive, self-replicating nanorobots.” But amongst other things, you’ll find, in the results, “fear of runaway robots”; “the grey goo threat”; “worse than grey goo”; and yes, even “the end of the world may be nearer than you think.” Clearly, grey goo is not a thing you’d like to be associated with.
This variety of goo, according to some, will be the way the world ends. We’ll have more to say about grey goo later; for now, we’re hoping that swarms of nanorobots will remain under control and do our bidding-exploring planets, repairing arteries, and other noble deeds.
So how did we get from ants to nanobots-robots so tiny they’re measured on the nanoscale? It’s scientific, as you’ll find out. Let’s get back to the ant.
Having no commander, the ant yet gathers its food. This is a marvel of nature: does an ant even have a brain? Even if it does, it doesn’t go a long way in helping the ant gather its food. What helps the ant more is the fact that it is part of a larger entity called the hive. The hive exhibits intelligence that an individual ant is incapable of by itself-the whole is greater than the sum of its parts.
Philosophy alert! Did we just touch upon a contentious issue? We did, apparently; we can already hear the clamours from those who disagree. Yes, wholes and holism and reductionism are touchy subjects, and we will not veer where many philosophers have gone before.
Back to the ant again: our glorification of that creature is not to imply that the bee is far behind. Each individual bee is reasonably dumb, but a hive of them can be dangerously co-operative. This we naïvely attribute to “collective intelligence”. We are re-stressing our point, but the concept is essential. And “It turns out that what makes sense in the biological world often makes sense in the computational world as well,” confirms Christian Jacob, leader of the Evolutionary and Swarm Design Research Group at the University of Calgary, Canada.
So what does all this have to do with computers? Not much as graphics cards go, but it has a lot to do with the idea of swarm intelligence (SI). If you like formal definitions, SI is the property of a system whereby the collective behaviours of unsophisticated agents interacting locally with their environment cause coherent functional global patterns to emerge. In English, SI is the idea that a large number of relatively stupid things can together exhibit useful, intelligent behaviour.
This is a story about how the group behaviour of insects and other fauna led scientists to dream of expansive swarms of tiny robots doing useful work.
The Flight Of The Nano-Bee
Some time from now-it could be anywhere between five and twenty years-we might or might not see “roboswarms” all around us. Given that, we ask: what will the roboswarms be doing? What will each nanobot be like? And why might we not be seeing them? These are pertinent questions, and it is our purpose here to answer them.
Intelligence in any form is valued by scientists: they notice it, then they study it, then they model it. We’re not talking about artificial intelligence-in fact, we happen to be talking about natural intelligence. Mankind has observed a lot of things in nature, including the stars with their patterns, the migration of birds, the evolution of chimpanzees into George W Bush, etc. One such thing was the social behaviour of insects and birds. Ants foraging for food, birds flying in formation, schools of fish swimming and turning together, are all examples of “swarm behaviour”-behaviour that reeks of intelligence, and which has therefore been studied and modelled by computer scientists. The verdict is out: the resulting “swarm intelligence” can be applied to a number of problems and situations, such as optimisation, certain military scenarios, robotics, and more.
The TETwalker robot at NASA’s Goddard Space Flight Center. Such bots will be miniaturised and grouped togeather,to be deployed as swarms!
That last got us interested, so we decided to investigate. Robotics and insect swarms taken together would mean, well, useful swarms of insect-like creatures. What might such things do? They could explore the surface of Mars, for a start.
In 1999, the Mars Polar Lander failed. It was, like so many technological edifices today, monolithic: when one thing fails, everything else sinks with it. Instead, imagine a roboswarm trying to explore the planet. It could spread out into different areas: some in the sky, some burrowing into the ground, some cruising along on the surface. If a few of the members of the swarm failed, there would be no problem-does an ant colony shut down shop if a few ants die? The swarmbots would have their sensory apparati in place, and they would gather data: about soil composition, about temperature, humidity, gases, and so on. Each swarmbot, we’re assuming, will have transmitters in addition to sensors, so all this data would be transmitted to someplace useful-like a central database, which could be on earth. No Mars Lander, no mess, no fuss!
Indeed, NASA has plans. Their Web site talks of a shape-shifting robot called the TETwalker (because it’s a tetrahedron) as follows: “Engineers at NASA’s Goddard Space Flight Center… (successfully tested) a shape-shifting robotic pyramid. Robots of this type will eventually be miniaturised and joined together to form ‘Autonomous NanoTechnology Swarms’ (ANTS) that alter their shape to flow over rocky terrain or to create useful structures like solar sails.” We gather that since NASA is talking about shape-altering ANTS, the idea that a roboswarm can do something useful can’t be all stupid.
Nanoswarms are some way into the future, but we’ll see a swarm of baseballs tested right this year! July 20, it was reported that a team of MIT researchers led by Prof Steven Dubowsky has conceptualised a swarm of baseball-sized bots to explore the Martian surface. Several thousands of them would explore difficult terrain, going where no probe has gone before-merrily rolling and bouncing away! Each “ballbot” would carry environmental sensors and cameras. They would communicate by forming a LAN, and a base station would relay their data to Earth. A prototype of the swarm will be soon tested on this planet.
How are each of these nanobots-the “agents”-aware of each other, and how do they co-operate for the job they’re doing? That’s precisely where SI comes in. As Yang Liu and Kevin Passino-whom we quoted above-say in Swarm Intelligence: Literature Overview:
“The agents use simple local rules to govern their actions and via the interactions of the entire group, the swarm achieves its objectives. ‘Self-organisation’ emerges from the collective actions of the group. SI is the collective intelligence of groups of simple agents… The autonomous agent does not follow commands from a leader, or some global plan. For example, for a bird to participate in a flock, it only adjusts its movements to co-ordinate with the movements of its flock-mates.”
The autonomous agent is the Ant in the Bible, but how exactly can intelligent behaviour emerge from simple interactions? It’s complex, but an example should suffice. Think of streams of ants converging towards a speck of food. You’ve seen it: the ants are all moving in various directions. They sometimes go straight, and sometimes they turn. It’s been proved that they’re as likely to turn left as to turn right! Then one ant finds the food. Its duty is to rush back to headquarters and report the find-and herein lies its simple intelligence: it can follow its own trail to go back to where it came from. While doing so, it deposits a “food” pheromone along the way. The next ant that passes the pheromone will follow it-ant instincts dictate that they move in the direction of “food” pheromone. This second ant returns, and the pheromone along the path is now twice as strong-smelling. It’s now twice as likely for the next passing ant to take that path, and so on until the entire colony gravitates towards the food the first ant found.
Underneath this, we find something more interesting: ants can find the food closest to them-meaning more chances of successfully transporting it home! How? It’s simple: a given ant has a certain amount of pheromone inside it, and the ant that finds the closest food source can leave behind more of the pheromone. (The ant that finds food far away will have its pheromone exhausted by the time it gets home.) As a result, the colony gravitates towards not just any food, but towards what food is closest to them!
On the surface, the ants seem to be searching for food and finding it effectively. But backstage, it’s all probability and pheromones. This isn’t the whole story, but here’s one way intelligence can come from simplicity.
Essentially, the ants “self-organise.” Self-organisation refers to the fact that “good” patterns and configurations-as in the ants finding the nearest food-emerge from interactions between ants, and between the ants and the environment. These interactions must, naturally, be conducive to the emergence of intelligence: for example, Darwin would quickly kill off a species of ant where individuals simply lunched on the food they found.
We, Robots
Intelligence emerging from group activities is fine, but what about the individuals themselves? Leaving behind ants and/or bots for the moment, let’s just call them “agents.” So each agent’s properties sheet would read something like the following.
Alignment: Each agent has the ability to align itself with the others: just find all the agents in the vicinity and average where they’re headed. Then align yourself in that direction. Except for the first agent, who is headed god knows where, every other agent gets aligned because its neighbour is aligned because its neighbour…
Cohesion: An agent can approach, and form a group with, nearby agents. Find some of your neighbours, figure their “average” location, then move to that location.
Obstacle avoidance: An obvious requisite. Have you ever seen an ant bump into a wall, then go back, and bump into the same wall again? Well, it does seem like ants have an inbuilt obstacle-avoidance algorithm, and any agent participating in a swarm needs one.
Separation: The ability to maintain a distance from nearby agents. This prevents crowding, which allows the agents to scout a wider area.
That’s about behaviour intra-swarm. As for interactions with the environment, if we’re talking about a little robot, it would need equipment that enables it to take in things from the environment-such as solar cells for power, sensors of various kinds, and more-and it needs to be built such that it can contribute to the environment: it needs grippers, wheels, and so forth.
Born To Bond
Summing up what we’ve said thus far: each robot needs to have some means and characteristics by which it can act as part of a swarm. Intelligence and useful behaviour can emerge from the swarm, and what kind of behaviour emerges depends on the properties of each member. The idea is to design little bots such that a swarm of them can do something useful. And various uses for robotic swarms have been envisaged, including terrestrial, outer-space and submarine exploration.
An individual s-bot at the SWARM-BOTS project
A collection of s-bot actually crossed a ridge in this fashion!
SWARM-BOTS was a project funded by the Future and Emerging Technologies programme of the European Community. Its focus was “self-organising, self-assembling, biologically-inspired robots.” The project calls the swarm itself a “swarm-bot,” which is an aggregate of s-bots. See illustration above: the little critters can self-assemble by connecting and disconnecting from each other using their claw-like grippers! The SWARM-BOTS project hoped to create groups that could “explore, navigate and transport heavy objects on rough terrains in situations in which a single s-bot would have problems to achieve the task alone.”
Each s-bot itself has limited computational power, but apart from that, the spec-sheet of the s-bots is impressive. Each has “proximity sensors, light sensors, accelerometers, humidity sensors, sound sensors, an omni-directional colour camera, force sensors…” The swarm-bot as a whole was designed to reconfigure itself as and when needed: for example, falling into single file to go through a narrow passage. The project mandated that the swarm-bot aggregate should form as a result of the collective intelligence of the s-bots, rather than hard-coded rules.
Potential applications for the s-bots include space exploration, rescue searches, and underwater exploration. The project ended in March of last year; what came of it? Did the s-bots form swarms and explore Mars? Of course they didn’t, because they were never sent there. One of the things that did happen was that, as in the artist’s impression above, they were able to cross a gap in a ridge by holding on to each other. They aggregated; they moved in a co-ordinated fashion; they lifted heavy objects, as promised; they navigated rough terrain; and did many other things worthy of a second round of applause. If you’re not on dial-up, see www.swarm-bots.org/index.php?main=3&sub=35 &conpage=s41r for videos of s-bot(s) on various kinds of terrain.
Intelligent Small World Autonomous Robots for Micro-manipulation, or I-Swarm, is a project at the Institute for Process Control and Robotics of the Universität Karlsruhe in Germany. It aims at facilitating the mass-production of microbots. These are expected to be deployed as swarms consisting of up to a thousand units, and the researchers expect the swarm to be able to do lots of things, including micro-assembly and biological, medical, and cleaning tasks. (No, these bots won’t be sent to Mars either.) Amongst the broad objectives of the project are to demonstrate collective task execution, to showcase the collective intelligence of robots, and to “use advanced sensors and tools for manipulation in the small world.”
The bots being designed at the 1-Swarm project look very much like adorable insects
The artist’s impression below is of the bots being designed at I-Swarm. They’re a few millimetres large, and are pretty basic in structure as well as capabilities. This project extends to 2008: be sure to check back in this space to find out what happened! If your band is broad enough, visit http://dsc.discovery.com/news/media/swarmrobotvideo.html for a video of cute little swarmbots doing equally cute things.
Then, there’s the Smart Dust project on at the University of California at Berkeley. This one aims at building a “millimetre-scale sensing and communication platform for a massively distributed sensor network. This device will be around the size of a grain of sand and will contain sensors and bi-directional wireless communications, while being inexpensive enough to deploy by the hundreds.” So what would the grain-sized sensing device do when deployed by the hundreds? Lots of things, possibly.
“Yes, personal privacy is getting harder and harder to come by. Yes, you can hype Smart Dust as being great for big brother. Yawn. Every technology has a dark-side deal with it”
Kris Pister,
Professor of Electrical Engineering and Computer Science,
University of California at Berkeley
Of Devils And Dust
Kris Pister, Professor of Electrical Engineering and Computer Science at the University of California at Berkeley, has imagined several applications for Smart Dust, his baby. Of course, each of us is free to imagine, and if you’ve got what seems a good idea, write in to Pister-he’s listening!
Think of this: you could glue a Smart Dust mote to each of your fingernails. Then, accelerometers will sense the orientation and motion of your fingertips, and talk to your computer. If it knows where your fingers are, you could sculpt 3D shapes in virtual clay, play the piano, and do much more!
Then, there can be defence-related sensor networks that can perform battlefield surveillance, and transportation monitoring. Smart Dust could find its way into quality monitoring: things such as temperature, humidity of meat and dairy products, and so on, could be monitored.
Inventory control applications are interesting: “The carton talks to the box, the box talks to the palette, the palette talks to the truck, and the truck talks to the warehouse, and the truck and the warehouse talk to the Internet. Know where your products are and what shape they’re in any time, anywhere.”
You could put Smart Dust motes on a quadriplegic’s face, “to monitor blinking and facial twitches-and send them as commands to a wheelchair / computer / other device.” (This was an idea someone sent in.)
Pister also speaks of smart office spaces, where ambient conditions are tailored to the needs of each individual. “Maybe soon we’ll all be wearing temperature, humidity, and environmental comfort sensors sewn into our clothes, continuously talking to our workspaces which will deliver conditions tailored to our needs.”
Naturally, many others, including us, have tried to imagine what use smart dust, or a swarm of motes, or a roboswarm, or whatever, could be put to. If a swarm were to explore the ocean floor, it could be programmed to look for precious metals or minerals. When one unit finds something that seems precious, it could send out signals that would cause others to gravitate to that spot-but not all of them, because the area under surveillance has to be populated. Then, if a few bots get eaten by a shark, their absence would be felt, and others would come in to fill the gap.
Think of traffic surveillance: if you needed to know the general direction of traffic, or the traffic pattern in a particular area, you could just spray the entire area with a near-invisible swarm of cam-equipped bots! You’d get the entire picture-including, if you wanted such data, where there are more yellow cars and where the humidity correlates with accidents-without the need for complex computer vision algorithms, and with the entire swarm reprogrammable at the press of a button.
Think of medical applications. Nanobots could do things like repairing clogged arteries, and a variety of procedures that involve going where medical equipment can’t go directly. Think of dangerous missions like clearing out minefields. Think of operations that involve going where no man has gone before. What seems a better idea-sending in a programmed humanoid and risking it getting destroyed, or using a spray can to fill the area with smart motes? If some bots died, the others would rush back, warning of something evil going on down there. Need to destroy toxic waste? Let each swarmbot carry a little bit of it, and what you have is a toxic swarm…
The applications of a swarm of tiny, suitably-fitted robots are limited only by the imagination. But what’s scary is: suppose that toxic swarm were to veer in the direction of Parliament House?
Goo… D?
We’re now back to grey goo, which, as you’ll recollect, we started off with. The fear is about swarms getting out of control, clogging the atmosphere, and doing a variety of other bad things. The problem comes when we make the robots self-replicating. (There is good reason to do so, from the cost as well as redundancy points of view.) And once they can self-replicate, what would happen if something went wrong at the control centre? Would they multiply, swim through toxic waste, come into populated areas, and spread death and disease? Would they consume people?
The scare began when molecular nanotech pioneer Eric Drexler first used the term “grey goo” in his 1986 book Engines of Creation. In that book:
“Early assembler-based replicators could beat the most advanced modern organisms. ‘Plants’ with ‘leaves’ no more efficient than today’s solar cells could out-compete real plants, crowding the biosphere with an inedible foliage. Tough, omnivorous ‘bacteria’ could spread like blowing pollen, replicate swiftly, and reduce the biosphere to dust in a matter of days.”
Drexler doesn’t now talk in such terms; he has recently made attempts to focus worry regarding nanotechnology in the right directions. But then some people read Michael Crichton’s 2002 novel Prey, which talks about precisely the same thing-the threat of nanobots going beyond control, replicating like nobody’s business, and generally creating a scene. Some of these readers got scared.
Bill Joy, co-founder of Sun Microsystems, once wrote a column in Wired Magazine titled Why The Future Doesn’t Need Us. There, he famously stated that we’re fools rushing in-into territory best left unexplored by the GNR technologies (genetics, nanotechnology, robotics). Joy, certainly no Luddite, gauging from his profile, goes so far as to say that there are some research areas we ought not to pursue because the consequences might be dire. If a man of Joy’s stature is worried, what about the rest of us?
Grey goo seems more of a public scare than a real problem. But there do exist scientists trying to prove that grey goo is impossible, and pseudo-scientists who say the scenario is not entirely implausible.
Space does not permit a plausible-or-not discussion, but it’s essentially another doomsday and big-brother scenario. As Pister puts it, “Yes, personal privacy is getting harder and harder to come by. Yes, you can hype Smart Dust as being great for big brother. Yawn. Every technology has a dark-side deal with it.”
Yawn, yes. But Google “grey goo” anyway.
It’s interesting.