At Ecole Polytechnique Fédérale’s Laboratory of Intelligence Systems in Lausanne, Sabine Hauert, Severin Leven and Dario Floreano have found a way to make small, fixed-wing machines fly together, migrate and avoid crashing. The swarms can be used for imaging and mapping the ground. In the future they may fly on search and surveillance missions.
The swarming behavior is based on a three-dimensional algorithm that represents the movements of schools of fish and flocks of birds. The algorithm, developed in 1986 by Craig Reynolds, was first used as a computer graphics tool. In the algorithm, as in real flocks, the individual agents behave simply. They respond to their close neighbors without considering the movements of the group. Yet out of the noise, larger patterns emerge, coherent and beautiful.
“Flocking requires three things. You need to move with the same speed and direction as your neighbors, you need to avoid hitting them and you need to stay close,” said Hauert, who is now a post-doctoral student at the Massachusetts Institute of Technology. When programming the robots, Hauert and Floreano added in a fourth ability: migration. With this ability, the robot swarm can travel to a set location, making them more useful as search and surveillance tools.
The robots were built by senseFly, a Swiss startup founded in late 2009, as a spin-off of the work done at the Floreano’s lab. The bots communicate with each other via a simple WiFi dongle connected to the on-board Linux computer that runs the autopilot program. In this project, Hauert and Floreano searched for the best balance between the robots’ weight, cost, turning ability and the range of their communications, while keeping their ability to flock.
“You can imagine two robots working together like holding hands with someone,” Hauert said. “If one of you changes direction, but the other person can’t respond quickly enough, then you break the connection.”
To perfect the technology, they tested the robots over 200 times in the field, without any crashes. The trials began simply, eventually reaching a flock of 10 autonomous fliers. More could be added, they believe. They’ve simulated up to 100 flocking robots. Next Floreano is interested in adding different kinds of sensors, such as cameras modeled on insect eyes, so the robots would be able to avoid obstacles while staying as a flock.
Hauert and Floreano will present their work at the International Conference on Intelligent Robots and Systems, held from Sept. 25 to Sept. 30 in San Francisco. The research is part of the SMAVNET project, supported by the Swiss Federal Department of Defense.