Robots marry natural neuroscience, experimental control to probe animal interactions

Studying animal behavior in the wild often gets hairy, with little experimental control and an abundance of extraneous data. And when multiple animals get together, the way they look, act and smell all influence one another, making it difficult to parse complex social interactions, says Andres Bendesky, associate professor of ecology, evolution and environmental biology at Columbia University.

Robotic or animated partners, however, can simplify that equation. Studying animal-robot interaction gives researchers complete control over one partner during any tête-à-tête, Bendesky says. It makes it possible to present the same stimulus to an animal repeatedly or compare how different individuals react. And the method complements observation-based research: Scientists can use a robot- or animation-based paradigm to test ideas gleaned from studies that use artificial-intelligence tools to track behavior.

“It’s the opposite of a vicious cycle—a virtuous circle,” Bendesky says.

Bendesky is part of a growing cohort of neuroscientists turning to robots to help them decode social interactions. The quirks are still being ironed out, but the approach is already helping several groups tackle questions about schooling, fighting and chatting behaviors.

The rigor of the results depends on whether a critter believes what it sees, says Tim Landgraf, professor of artificial and collective intelligence at Freie Universität Berlin, who uses robots to study group behavior in guppies. That can be hard to gauge; there’s no handbook that describes what traits make a robot believable, he says.

But researchers can compare how animals act toward a real peer versus a counterfeit one, says Steve Chang, associate professor of psychology and neuroscience at Yale University, who doesn’t work with robots but studies the social behavior of macaques and marmosets.

Even then, it’s unclear if the animals are treating a robot as a social partner or as “some really special case of some nonsocial object,” Chang says. But the approach is “super exciting”—and essential to confirm that knowledge about the brain gained through controlled experiments holds up in real-life situations.

Faux fighting

Betta fish are born to fight. During a typical interaction, male fish take turns flaring their gill covers, turning their bodies to the side and spreading their fins. And then the biting begins.

Bendesky and his team animated a short sequence of a betta fish performing this visual display to measure how real fish would respond. By tweaking different aspects of the animations—such as swimming speed, elevation and the amount of gill flaring—they discovered that the swimming depth of the animated fish correlated with the aggression levels in the observer fish; swimming higher incites more aggression than swimming closer to the bottom of the tank. Bendesky says this observation surprised him at first, until he thought about the fact that betta fish build and guard nests for their eggs on the water’s surface.

To tease apart the biting behavior, the team developed a silicone 3D-printed fish attached to a clear rod that can be lowered into a tank and steered by cameras tracking a real betta fish nearby. The robot can follow the other fish and bully it into biting or take a programmed, nonthreatening route around the tank. In this way, it offers greater control over the interaction and prevents any injury to the real fish, Bendesky says.

The group is currently collecting data using the robot and plans to incorporate the method into an ongoing project focused on the genetic differences between wild betta fish and domesticated species bred for fighting. They hope to identify specific genes linked to aggression and then test how manipulating those genes affects fighting behaviors.

Chat bot

Budgerigars love to yap. These parrots hold “very rich conversations” with one another, mimic human speech and can learn the vocabulary of a new budgie in about a week, says Michael Long, professor of neuroscience and physiology, and of otolaryngology, at New York University. The bird’s neural circuitry is equally flexible: Activity in the central nucleus of the anterior arcopallium—the vocal output region—encodes chirps and songs by their acoustic properties, which is similar to how the human brain dictates speech, Long and his team reported earlier this year.

Now, Long and his team are working to decode the meaning of budgie vocalizations and probe the limits of their vocal abilities using a robotic bird—an idea that came from Andrew Bahle, a postdoctoral associate in the lab, who sourced a cheap pencil holder online that came with life-size and realistic animatronic budgies. The fake birds even have accurate sex-specific coloring on their beaks.

The real birds in Long’s lab warmed up to their robot companions right away, Bahle says: They sing to them, nuzzle them and do “all sorts of social behaviors.” Hopefully, they will also use the full breadth of their vocabulary with them, helping the team to explore the range of sounds the birds can mimic and whether they can learn, retain and use multiple vocabularies. Other team members are using machine learning to form hypotheses about what the different vocalizations mean.

“We’re hopeful that the robot will offer a way to actually test those [hypotheses], because we can’t really ask the birds to say what we want,” Bahle says.

Follow the leader

Using animal robots may be most successful with visual animals, Landgraf says: When he used a robotic bee to study the insect’s waggle dance, which involves both visual and vibration cues, it proved more difficult than using a robotic guppy to study schooling behavior, which depends mainly on visual signals.

A school of fish is essentially many one-on-one social interactions occurring at the same time, Landgraf says, and he wants to untangle how those interactions lead animals to act as a collective. Robot guppies—he uses a 3D-printed fish attached to a magnetic base that he can control from beneath the tank—make it possible to test the potential rules guiding those interactions.

As it turns out, guppies are more likely to follow the robot around the tank when the robot is programmed to factor in their avoidance behavior, Landgraf and his team reported in 2023. Fish that are standoffish at first eventually warm up to a more cautious robot but stay away from an aggressive one. The team is also using the robot to test the accuracy of models of schooling behavior.

“An adaptive leader is a better leader than a static leader,” Landgraf says. “It would not have been possible to show that without robots.”