Monday 12 January 2015

Intelligent swarms

This essay was published in Luxury Briefing in January 2015

Intelligent swarms
by Alexander Gallé
12 January 2015 

The rise of Tesla over the last couple of years has brought with it an increased awareness of what some people call 'smart factories': factories where most human labour is replaced by multi-purpose robots that are aware of certain variables that inform them of one another's activity and where they amend their own activity accordingly.

I've discussed the rise of smart companies before, citing Fabergé as an example of a luxury-sector company that could vacate its premises and set up shop within 24 hours in another location, because its entire production chain has been digitised – with every stone coded from the moment it leaves the mine to the moment it is sold. The knowledge, which is the core of the company, is held in the cloud. Tesla takes this a step further by having much of the work itself performed by multi-purpose robots that operate directly from this knowledge bank and feed their own knowledge back into it. The factory, as a whole, acts like a big, self-optimising machine.

So, what is the next step for this method of production? In two words: intelligent swarms.

Limitations have existed for a long time in robotics. One of the big hurdles was communication between the robots that would enable fluid interaction and self-organising systems. Another was robots' freedom of movement. Simply put, getting robots from A to B without bumping into things, crashing into each other or losing their balance was always quite tricky. Furthermore, until now, the idea of them flying always seemed near impossible.

However, with the amount of progress made in quadcopter flying technology, flying them has now turned out to be the most logical option. Just as flocks of starlings can fly in huge groups (called "murmurations") without knocking each other out, so too can quadcopter robots now fly in very close-knit formations without hitting each other, moving in space as a group without the limitations of physical terrain.

What's more, each quadcopter robot can be assigned a small part of a complex task that is to be completed by the robot swarm as a whole. Since one quadcopter is very much like another, the swarm can then be re-organised in very flexible ways to optimise its performance. For example, let's say that you’d like a group of robots to build a large structure, like a house. The freedom of movement of the robots would be a big issue to solve: how do you get the robots to go back and forth from the bricks deposit to the wall they are building on the third floor? With quadcopter technology, these issues are no longer relevant: the quadcopters just fly to and from the deposits, each one performing its small task over and over until the bigger task is complete. The swarm redirects its efforts as and when the work is done in some places and still to be completed in others. The swarm may be made up of thousands of small, flying robots performing individual actions, but it "thinks" as one.

We tend to think of industrial robots as relatively large things, which is how they have been until now. However, the larger they are, the narrower their purpose, the less flexible they are the less intelligent. The swarm approach allows for a "many small hands make big things" solution to a much wider range of complex tasks, just like bees making honey.