February 5, 2005
Social Robots Without Leaders
How a flock of birds or school of fish may go in a single direction without having a permanent leader is a mystery of social organization. But adding a mechanized component to the question raises the possibility of robotic swarms which may lack much in the way of social skills or leadership qualities.
Astrobiology Magazine -- A study led by Princeton biologists has revealed a remarkably simple mechanism that allows flocking birds, schooling fish or running herds to travel in unison without any recognized leaders or signaling system.
The finding, published in the Feb. 3 issue of Nature, helps settle age-old questions about how animals coordinate their actions. Previously, scientists had looked for subtle signals or other explicit systems that animals may use in disseminating information through groups. The new study showed that such complexity is not necessary: Large groups easily make accurate decisions about where to go even when no individuals are regarded as leaders and very few individuals have any pertinent information.
In addition to shedding light on the graceful coordination of animal groups, the results may be useful in understanding how humans behave in crowds and in designing robots that explore remote locations such as the ocean or other planets.
Using computer simulations, the researchers found that group coordination arises naturally from two basic instincts: the need to stay in a group; and the desire by some individuals to act on their own information about where to go. First, the researchers programmed their simulated animals with a basic urge to stay near others, but not collide with them. This instinct alone caused individuals to form close-knit, evenly spaced groups like those of real animals, which pay a high price -- such as being eaten -- if they stray from their group.
Second, a few animals were programmed to have a preferred direction, as if aware of a food source or other valued destination. These animals were instructed to balance their desire to move toward their goal with their desire to stay in the group. In repeated simulations, it required only a few individuals with specific goals to set hundreds of others on the same course.
"It demonstrates the power of the little guy," said Daniel Rubenstein, chair of ecology and evolutionary biology. "You don't need avowed leaders, you don't need complex signaling."
In the simulations, a small number of "informed" individuals was just about as effective in leading a large group as a small one. When the size of the group increased from 10 to 200, the percentage of informed individuals needed to accurately guide the group reduced from approximately 50 percent to less than 5 percent. As a result, the number of informed individuals required to lead a small group was about the same as the number needed to lead a large one, Couzin said
The researchers extended their experiments by introducing two subgroups with differing preferences about which direction to move. The overall group consistently chose the direction preferred by the bigger subgroup even if the two subgroups differed in size by only one member, demonstrating that groups can come to an accurate consensus decision even though individuals are not aware of the preferences of others or whether they are in a majority or minority.
Couzin, who divides his time between Princeton and Oxford University, collaborated with Simon Levin of Princeton as well as Jens Krause of the University of Leeds and Nigel Franks of the University of Bristol.
In addition to its biological insights, the research may provide ideas for designing robots that work together in groups, said Couzin, who has collaborated with roboticists at Princeton on designs for fleets of underwater vehicles.
"Say you had a group of robots exploring a planet or the ocean, and each individual robot was moving around collecting information locally," he said. "By returning to the group and following the type of algorithm we propose, they could select collectively the direction associated with the best quality information or select collectively the majority direction."
Couzin and colleagues are now testing their findings in experiments with animals, such as schools of fish in which a few have been taught to seek food in a certain location. He also believes the results may shed light on how people move in crowds and is currently organizing experiments that test the principles in large groups of student volunteers.
On the Net:
William "Red" Whittaker, Robotics Institute