Hive hopes

Swarm robotics is a mainstay of science fiction. Armies of little robots that can work together to accomplish a variety of tasks have appeared in everything from Star Trek to Doctor Who and all points in between.

The reality is just beginning to take shape in laboratories around the world and is proving just as fascinating as the fiction, with potential applications including military reconnaissance, mining, search and rescue, and space exploration.

In reality swarm robotics is a highly challenging form of engineering in which the usual rules of robotic systems are useless. Rather than being targeted directly at solving specific problems in industry, it is a more theoretical proposition, with surprisingly close links with, and cross-fertilisation from, the life sciences. Computer simulation techniques can be a misdirection rather than a useful tool. And instead of building complex tools, the focus is on simplicity.

Alan Winfield of the Bristol Robotics Laboratory, one of the UK’s leading experts in swarm robotics, said the defining feature of swarm robotics is the way the robots are controlled. ‘Swarm robotics is trying to develop a system based on the principles of insect swarm intelligence, in which there is no centralised command or control at all. We don’t fully understand how swarms of insects work, but we do understand that there is no ‘brain-ant’ directing the actions of all the millions of individual ants. And not only that, there is no hierarchy at all in an organisational sense.’

This is an important difference from multi-robot systems, where a small number of relatively complicated robots are used to complete a task, such as welding together the various parts of a car body.

‘Those systems use a conventional, hierarchical, top-down command and control structure, but swarm robotics eschews that altogether and goes for a 100 per cent distributed control approach,’ said Winfield.

There is no central computer directing each robot, and no task leader: instead, the goal is to equip each robot in a group numbering tens or millions, with a system that governs how it reacts to its neighbours and its surroundings and with a specific set of capabilities. The desired problem-solving behaviour should then arise almost spontaneously when the robots are placed into the environment — a phenomenon known to swarm roboticists as ‘emergence’.

It is fiendishly complicated, because engineers have no model for how to design such a system. It is completely unlike any conventional form of robot control, where specific actions can be programmed into the robots and executed in order.

So what is the attraction? As Winfield explained, it is all to do with the concept of safety in numbers. ‘If you try to approach a problem using one, typically very complicated and expensive, robot, and it goes wrong, then you’re stuffed,’ he said. ‘But if you can approach the problem using a very large number of simple robots, then the solution is more robust and resilient.’

There is no central computer that could be disabled, and each individual robot is disposable or, at least, not crucial to the task. In fact, scalability is an essential characteristic of a swarm system, Winfield said. ‘If your system would stop working if you had a very large number of robots, it suggests that you’re not doing it right,’ he said.

While most swarming projects use a relatively small number of robots — between 10 and 100 is typical — the envisaged operation system could involve hundreds or thousands of very small robots, or even millions of microscopic units.

This concept has served insects extremely well. ‘If we can achieve it, then it promises to build a system which has an extraordinary level of robustness. It would tolerate very high levels of failure of individual robots, with the overall system just carrying on regardless and doing the job, albeit possibly slower and less efficiently,’ said Winfield

Swarms are so efficient that they might even be able to tolerate individual robots doing completely the wrong thing. ‘Leaf-cutter ants will always have a number of individuals carrying bits of leaf the wrong way, but it doesn’t matter,’ he said.

An individual swarm robot will be equipped with some way of moving around, an array of sensors so that it can detect other robots and aspects of its environment, and some method of affecting the environment, such as a system for attaching to other robots; as well as some processing power to operate the control algorithms.

Although Winfield calls the individual robots ‘simple’, this actually leads to complex little machines; and advances in miniaturising these types of technology have led to the development of swarm robotics.

The field has emerged only in the last 10 years, Winfield said. ‘It has allowed us to build robots small enough, and reliable enough, to have significant numbers of them and to run meaningful experiments,’ he said. ‘You need sufficient computing, communication and battery power for these fairly complex systems, and power management and sensing are critical. Experiments tend to take several hours, so they need to be able to run for a fairly long time and be reliable.’

Being able to build robots is crucial, because computer simulation is not a reliable method for studying swarms. This is because there is inherent randomness in a swarm of robots; wheeled robots will have wheels which are not all precisely round; there will be minor differences in gearboxes; the environment will interact with the robots in unpredictable ways.

‘These small differences and the noise in the environment somehow assist or enable the self-organisation and emergence of behaviours in ways that we don’t understand,’ said Winfield. ‘Computers aren’t good at randomness; they’re always pseudo-random. Roboticists have found that they can demonstrate things in simulation, but when they transfer them to real robots, they don’t see the same behaviour.’

In terms of applications, swarms are well suited for any activity that falls into the ‘foraging’ category. ‘Any application where you have stuff — any sort of stuff — spread out in the environment, and you have to find it and do something with it, whether it’s on land, sea, in the air or in space, is a possibility for swarm robotics,’ said Winfield.

‘All sorts of exploration and surveying, through to mining and harvesting, and areas like construction and search and rescue are also good targets, as are medical applications.’

Medicine is one area where even one robot doing the wrong thing could be disastrous. However, borrowing another property from nature, the swarm could be equipped with an immune system that detects aberrant units and isolates them so that they can do no damage.

Winfield is involved with a European Union project called Symbrion, which aims to build a swarm of land-going robots that can join together to form an ‘artificial organism’, capable of pooling its energy and computational resources. This will create an ‘evolve-able’ robot system that will be able to reprogramme itself to take account of its environment and the changing nature of its task, Winfield said. ‘One part of the project is to develop an auto-immune response, so that if one robot goes bad, the others can nullify it.’

The ultimate goal of this project is to develop a system that could be used in disaster relief and search and rescue. ‘You might have thousands of robots out looking for survivors, then when they find them, they’d self-organise: some would deliver first aid, others could physically remove obstacles and rubble; others might form a communication chain with the outside world,’ said Winfield. At the moment, however, the project in its early stages. The project teams, in 10 universities across the UK, Germany, Austria, France and Belgium, are now developing a basic robot model that can link together to forage for power supplies.

The Symbrion project, like many swarming projects, uses identical robot units, unlike another European Union project, called Swarmanoids. Involving researchers in Switzerland, Italy and Belgium, this project is developing three different types of robot that will act together in a single swarm.

Project co-ordinator Marco Dorigo of the Universite Libre de Bruxelles explained the three robot types. ‘We have eye-bots, which can sense and analyse the environment from a high position; they are essentially small helicopters. We have hand-bots, which can climb vertical surfaces; and foot-bots, which are wheeled robots that can move over rough terrain and transport objects or other robots.’

Although other projects have looked at so-called heterogeneous swarms, the Swarmanoid project is the first to use a system that can work in three dimensions.

Each robot is about 10cm in diameter; the foot-bot has grasping hands that will allow it to grab other foot-bots or hand-bots, while the hand-bots are able to fire a rope up to the laboratory ceiling and climb up it with its pair of gripper-equipped arms. The eye-bot carries an array of visual sensors on a small four-rotor flying platform.

The robots were developed from an earlier ground-only project called Swarm-bots, with which Dorigo was also involved. ‘We’ve had to develop completely new ways of coordinating the robots,’ he said. ‘For example, the communication systems we had before were no good for a flying robot that is directing a robot from the ground. We’ve also had to equip the ground-based robots with cameras so that they can see robots on the ceiling.’

The Swarmanoid robots are equipped with control programs that cover four basic tasks: searching for items, collection and grasping of those items; transport; and deposition. These were chosen, Dorigo explained, because a huge number of real-world tasks require a combination of these operations.

Each robot has its part to play in the overall task, and the robots will communicate in a variety of insect-like ways — by a wireless signal, analogous to insects releasing a pheromone that the whole swarm can smell, or by a visual cue, such as the ‘dance’ used by bees to direct workers to nectar.

Another heterogeneous swarm project, based at the University of Southern California, is SuperBot, a space-oriented project to construct a modular, multi-functional robot. ‘The SuperBot could configure into a “flying arm” for extra-vehicular crew assistance and maintenance; then reconfigure into multiple “eye/hands” for in-vehicle crew assistance and autonomous health monitoring and management,’ says coordinator Wei-Min Shen.

It could also pack itself into a landing capsule, then reassemble on the surface into a rover, a climber, a platform for drilling, building or sample collection, for example.

It is a typically ambitious goal, but Shen’s initial aim is to develop a swarm of 100 reconfigurable modules that will be tested under desert conditions, and be capable of assembling into the four configurations. Its final test — for which Shen does not state a target date — will be to start from a ‘lander’ configuration; reconfigure into the rolling version; travel to a sand dune; reconfigure to climb the dune; then at the top form into a platform that will plant and tend a set of seeds until they sprout.

The project, based around ‘Lego-like’ robot modules with docking mechanisms and a novel distributed software control system, is being funded by $8m (£4m) from government bodies including DARPA and NASA.

Also in the US is one of the most science fiction-like swarm projects. At Carnegie Mellon University in Pittsburgh, Seth Goldstein is directing the Claytronics project, which he describes as not so much swarm robotics as ‘programmable matter’. He aims to develop robots in the 100micron scale that will be able to self-assemble into tools, 3D models, or even ‘moving statues’, which will assist in communication.

‘We started about four years ago,’ he said. ‘[Co-researcher] Todd Maury and I were interested in improving teleconferencing, and I was interested in programmable materials, and we thought it would be great if instead of having a phone call, we could have a face-to-face meeting with me in his office and him in mine.’

To meet this mind-boggling goal, Goldstein and his team are working on both macro- and micro-scale robots. At the large scale, they have developed ‘cube’ robots, whose faces are mounted on rods that can extend from 8cm to about 24cm in length, and which are equipped with electrostatic latches to attach to other ‘cubes’.

On the micro scale, Goldstein is using photolithography, the technique used to etch electronic components onto microchips, to make robots that can self-assemble from a flat sheet into a three-dimensional form. ‘We’re in the process of making a robot that’s essentially a cylinder that can rotate anticlockwise or clockwise,’ he said. ‘A sub-millimetre robot has enough surface area to put a processor onto it, with a few hundred kilobytes of memory, and communication and support circuitry.’

Goldstein said the development area is more in software than hardware. ‘We’ve developed two new programming languages which can be used to write very efficient and concise programs that in the past would have required pages and pages of code,’ he said. ‘Our philosophy is that it’s important for the programmer to think about the ensemble as a whole, rather than individual robots, and these concise programmes allow them to think about optimisation techniques which could have been missed otherwise.’

Goldstein’s programmable matter goals are decades off, but he’s certain that his research will pay off in the short term. ‘We’re looking at 3D fax machines and programmable antennas, and we’re also interested in medical applications,’ he said. ‘You could imagine swallowing a robot, then it could change shape to perform diagnostic and surgical procedures.’ Such a system could be ready for trials within five years, he claimed.

Alan Winfield sounded a note of caution for commercial deployment of swarming systems, however. ‘You still have the down-to-earth engineering problem of validating the swarm,’ he said. ‘You’ll need to give guarantees that it’ll not only do the right thing, but it won’t do the wrong thing. We don’t know how to do that, because swarms break the rules of normal engineering design. But there’s no reason to assume that when they do have real applications, they’ll be any less reliable than any other engineered system. There’s nothing intrinsically scary about them.’