Science Works to Make Robots Smarter With Swarm AI!

Science Works to Make Robots Smarter With Swarm AI!

January 16, 2016

When we think of robots, what comes to mind these days is usually something like a Roomba, arobot vacuum, or a huge arm that can be used to assemble cars, or a walking, talking, bipedal science-fiction style android. What all of these seemingly disparate robotic forms have in common is that they’re individual ‘creatures’ with a central control system and a single ‘body’, much like humans and many other animals. However, more recent developments in robotic technology have allowed researchers to explore alternatives to the single self-contained robot, and they’ve been taking leaves out of the book of our social insect cousins to do so. Welcome to the age of swarm AI.

The Power of the Swarm

Ants, bees and termites are well known to form colonies of individuals that work together to advance the hive, or the community at large, making seemingly collective decisions that are far beyond what the individual insects could manage. This collective behaviour of decentralized, self-organized systems is known as swarm intelligence. In some ways this is not dissimilar to the human nervous system, with the brain itself being a vast collection of simple neurons that work together to produce complex behaviours. The difference is that in swarm intelligence, rather than being held in place in a centralized brain, the elements of swarm systems are relatively free to interact with their environment and each other.

Scientists have been quick to take advantage of the benefits of swarm intelligence, and to create their own version. Some example algorithms model natural systems like ant and bee colonies, but that’s just the tip of the iceberg. Artificial immune systems have been developed that mimic the abstract structure and function of the immune system to solve computational problems in various fields, and even the echolocation behaviour of bats has been exploited to allow swarms to discover the properties of a new space. Other applications involve crowd simulation, for example to evaluate boarding times for aircraft, and there has even been some cultural outreach in the form of swarmic art.

It’s an exciting time for swarm AI research, and as these robots move out of the lab and into the real world, they’re going to need a wide variety of human helpers to keep them going, from electronics technicians ready to repair them when they fail to Automation technicians that can implement the algorithms that control the swarm. Let’s take a look at a couple of different directions that this technology might go in.

Self-organizing organisms

One example of a research project comes from Harvard University, where swarms of specially designed robots called Kilobots can assemble themselves into particular shapes defined by their operators. This echoes the behaviour of army ants, who can link together to build rafts and bridges across difficult terrain, and social amoebas, who join together to form a fruiting body to escape the local environment when food is scarce. The Kilobots represent an advance in swarm technology because they require no management or intervention once a command has been delivered, cutting down on expensive processing time.

Once sent the 2D image that they are required to mimic, four robots mark the origin of a coordinate system and then all of the machines take turns in moving toward an acceptable position. To do this they use very simple behaviours, like following the edge of a group, tracking the distance from the origin, and maintaining a sense of relative location. Because these behaviours are so simple, they require no help from outside once their instructions have been received. Crucially, the Kilobots also correct their own mistakes, and because of their low-cost design their abilities are rather variable from robot to robot, so such error-correction is essential. At scale, the smart algorithm that controls the robots guarantees both physically and mathematically that the robots can complete the specified task.

Swarming and swimming

Another example is the EU-funded Collective Cognitive Robotics (CoCoRo) project, which has built the world’s largest collection of autonomous underwater vehicles that show collective cognition. Unlike the Kilobots, the CoCoRo robots are of three distinct types: the Jeff robot, which is agile and fast, the Lily robot, which functions as a relay and communicator, and a base station that floats on the water and is able to pass information to scientists on the ground. With these differently specialized types of robots, the swarm functions even more like an ant colony (with its differentially specialized workers, soldiers and Queen).

Examples of its use involve the exploration of seas, rivers and ponds, with the base station able to pull a bunch of robots behind it like a tail to keep them at a minimum useful distance. The robots communicate using blue LEDs rather than sonic or echo technology to ensure that they are well contained near the base station and don’t get lost. The robots can also sense electric pulses, and their self-charging mechanisms that allow them to dock with the base station autonomously will stop them running out of power.

The future looks bright for swarm AI (and programming in general) and who knows where this technology will flow next?

If you are looking to enhance your career in technology, find out more about the George Brown College online technical training programs

Add new comment