Journal Club

Highlighting recent, timely papers selected by Academy member labs

Robot swarms communicate best when messages spread neighbor to neighbor

A swarm of 50 robots adapted better to change when each robot had a small number of neighbors. Image credit: Andreagiovanni Reina

A swarm of 50 robots adapted better to change when each robot had a small number of neighbors.                                                    Image credit: Andreagiovanni Reina

In the not-so-distant future, swarms of robots could help contain environmental disasters such as wildfires, by locating and dousing the most dangerous patches, even as flames move and spread. But how would these individual robots coordinate effectively to pull off such a complex task? A recent paper suggests a surprising answer: a smaller, less-connected social network. According to the work, published in Science Robotics, a flock of simple robots is better at homing in on a dynamic target when individual bots communicate only with their close neighbors, rather than globally across the network.

“Less is more,” says senior author Andreagiovanni Reina, a postdoctoral computer engineer at IRIDIA, Université Libre de Bruxelles in Brussels, Belgium. “The robots, by having a smaller social network, were able to spread the newly discovered information more efficiently.” In general, researchers would expect that when more individuals are connected, information should spread more efficiently; simply put, more nodes are talking to each other. In fact, Reina and collaborators found the opposite. Highly connected robot flocks became inflexible because vocal majorities drowned out the contributions of minorities, even when the minority had better information.

The first hints of this unexpected discovery came from a network of 50 coin-sized Kilobots, each moving randomly around Reina’s lab benchtop by vibrating and sending infrared messages to each other. Reina had organized the benchtop into three zones: red, blue, and green. The zones appeared, disappeared, and changed quality over time and were preprogrammed so that, at any moment, some sites were superior to others; for instance, initially red was better than blue, which was better than green. When each bot encountered their first zone, the bot began to broadcast an infrared message advocating that other bots should become that color too. Bots in the highest quality zone, for instance red, messaged the most frequently, followed by blue, then green. As the bots moved randomly around the benchtop, they would change the color they advocated for in their message if they encountered a better color—say moving from blue to red, or if they picked up very frequent messages advocating for another color from other bots.

The distance each bot could communicate turned out to matter quite a bit. When bots could broadcast over the whole lab bench, then the majority opinion swept across the network; if most individuals happened to drift into the green zone, for example, then bots picked up green flashes most often, and so green swept the group. But when the researchers shortened the bots’ communication distance, so that only close neighbors could talk, then even a small minority in red would spread word of that best color, until it overtook the network.

The authors followed up their experiment by hammering out the mathematical models underpinning the robot swarm’s unexpected behavior. Essentially, “they showed they could tune the system,” says biologist Iain Couzin, director of the Max Planck Institute of Animal Behavior in Konstanz, Germany. Couzin, who was not involved in this recent work, has found similar trends in networks of schooling fish. In studies in 2015 and 2019, Couzin and collaborators found that golden shiner fish move closer to their neighbors in response to a perceived threat, reducing the number of individuals each fish communicates with. These small neighbor networks become the hubs from which sudden waves of change sweep across the school, allowing the fish to collectively evade predators. Couzin hopes that this latest robotics paper illuminates new mathematical rules governing biological networks, from neurons in the brain to animal groups.

The unexpected results and updated models offer “a tool that we engineers can use in the future,” says Vito Trianni, a computer science engineer at the Italian National Research Council in Rome. He does note, however, that further investigation will be necessary to discern just how generalizable these network rules might be. The takeaway: “Instead of sharing information with everybody, it’s better to convince others one by one,” he says.

Reina next plans to conduct an experiment in which the distance the bots can communicate will vary, spurred by environmental cues. Researchers could, for example, speed up the bots. This is not unlike what happens in ant colonies: an individual with important information will start running through the nest. Moving at high speed changes the ant’s communication network, spreading a message farther by encountering more nest mates.

Broadly speaking, the work may hold lessons for networks of humans as well. “In a globally connected society,” says Reina, “a minority that has better ideas cannot really share well or help the group to improve.”

Other recent papers recommended by Journal Club panelists:

Entropy-driven translocation of disordered proteins through the Gram-positive bacterial cell wall

Chemically recyclable thermoplastics from reversible-deactivation polymerization of cyclic acetals

A fossil record of land plant origins from charophyte algae

Macroevolutionary foundations of a recently evolved innate immune defense

Categories: Animal Behavior | Computer sciences | Journal Club and tagged | | | | | | | | | |
Print Email Comment

Leave a Comment

Your email address will not be published. Required fields are marked *