harvard swarm.jpg

Harvard’s Kilobots assemble to form complex shapes using simple programmed rules (via Harvard)

Harvard isn’t usually an institution we associate with robotics like we do with MIT, Carnegie Mellon or Georgia Tech, however the Ivy-League school can hold their own when it comes to ‘The Swarm’. Why have only one complex robot to perform tasks when you could use thousands? That’s the notion behind Harvard’s tiny, swarming Kilobots, which can ‘self-assemble’ to form complex shapes (something other than a circle or square). Swarms are highly beneficial in terms of productivity and complex problem solving, just look at all the biological examples that can be found all around us. Ants can grab onto one another to overcome obstacles, bees work collectively to produce honey and even trillions of cells come together to create individual organisms. So why not use thousands of robots to help solve problems? Back in 2011, Harvard researchers began developing the Kilobot project in an effort to make it easy to test ‘collective algorithms’ on hundreds and even thousands of tiny simplistic robots. When the tests began 3-years ago, the researchers demonstrated the swarming behavior of just 25 of the robots through forging, formation control and synchronization.

Fast-forward a mere 3-years and the Kilobots have evolved into a swarm of 1,024 strong, able to assemble themselves collectively into complex shapes such as letters of the alphabet. The robots themselves are about the size of a quarter with three rigid legs and moves using two vibrational motors (powered by a 3.4-volt Li-ion battery), allowing to travel forward and side to side. An infrared transmitter and receiver allow the robots to communicate with others that are in close proximity.

To form a 2D shape, a command is given and then processed through the robots onboard microcontroller. In the 1,024 robot ‘flash-mob’ demonstration, a researcher places four ‘seed’ robots to mark the location where the shape is to be formed. The bots on the outer edge of the mass then begin to move and follow around the edge until they meet up and coalesce on the four seeder bots (with the others following suit) and form the desired shape. The ‘assembly’ algorithm the researchers designed uses three ‘primitive behavior’ programs- Gradient, Distributed Coordinate System and Edge Following.

To account for movement error, the software is also comprised of a continuous space algorithm along with platforms for the detection of rare errors and the compensation for variation between robots. The software platform accounts for unreliable robots that misinterpret (or flat-out ignore) the desired shape command and head out on their own or others that are blocking their neighbors from traveling to the appointed destination.

The Kilobots will not be implemented to perform real-world tasks but are rather a platform to test group-centered AI applications for robotic collaboration. That doesn’t mean that future swarming robots won’t be implemented to perform tasks such as disaster relief, environmental monitoring or scan the retinas of people living in large apartment complexes while looking for criminals (Minority Report).


See more news at: