Hasta La Gadget!

A masters student at Georgia Tech University has created a system that allows a group of robots to move into formations without communicating with the other robots it is forming shapes with. The robots have no predefined memory or no prior knowledge of their location.

In the video above, the 15 Khepera robots make independent decisions based on the same information and by trial and error move their way into a formation that has been assigned to them. They move to spell out the word “GRITS”, standing for Georgia Robotics and Intelligent Systems.

Ted MacDonald, who devised the multi-robot system, explained to Wired.co.uk that most other systems tend to split robot formation into two parts. The first is figuring out where each robot should go and the second is actually making the formation happen. He wanted to find a way to do both of these things at the same time, i.e. be moving and working out where to go simultaneously.

He told Wired.co.uk: “Imagine if a group of 10 people were asked to form the shape of a box, but weren’t allowed to speak. People would look around to see if there were any lines forming and try and find holes for them to move into.”

His robotic formation used a 3D motion-tracking camera above the robots, which could assess the position of each of the robots and then broadcast that information to all of the robots over Wi-Fi — so they each had access to the same information. They then use an Iterative Closest Point (ICP) algorithm which assesses the difference between where they are positioned currently and the formation they need to adopt.

The robots can be placed completely randomly to start with. Each robot then makes an initial guess as to where it thinks the formation might be (perhaps noticing an early pattern emerging that resembles part of the letter). All of the robots follow the same rules and eventually converge to a solution.

The robots can be made to spell out any words in real time by following the same rules. Macdonald is now working on a variation that sees a leader robot (which does not follow the algorithm) be automatically assigned a role in the formation. This means that a single robot could be controlled remotely (by a human) to lead a phalanx of other robots in a formation, which could change en route.

Clearly in real-world situations you do not always have a motion tacking camera, but the same effect could be achieved by the robots having GPS devices relaying their positions to the others.

Macdonald is also keen to find out what happens if the robots cannot see every other robot in the group. So far it seems that they can still get into formation if they start off close to the right formation, but so far he hasn’t been able to mathematically prove when it works and when it doesn’t.

Potential applications for the technology include moving a group of robotic vehicles from point A to point B without experiencing congestion problems (which could have uses for the military). Because of the sophisticated 3D motion tracking, the same system could be used to move aerial robots into 3D formations.

This story originally appeared on Wired UK.