UMass Boston

As a part of his PhD research at UMass Boston, Shoubhik Banerjee used geospatial data from homing pigeons to see what strategies they used to navigate their way home and found that it may not be as complex as researchers had assumed.

Homing pigeons have the remarkable ability to find their way back to their coop, even when released many miles away in a new location. If they are released from the same location multiple times, they can become more efficient on each flight. However, scientists are investigating how the birds make these improvements. When pigeons fly together, can inexperienced birds actually learn from birds who have flown the route before, and find shorter paths home?

In a new paper published in eLife earlier this year, a team of researchers from UMass Boston used computational models to simulate different learning mechanisms that homing pigeons could use for navigation. PhD student Shoubhik Banerjee, his advisor Albert Kao, and post-doctoral researcher Fritz Francisco found something surprising: Despite their impressive abilities, homing pigeons are relying on cognitively simple techniques to improve their route when they fly together.

Banerjee’s research began as a response to a 2017 paper in which scientists recorded homing pigeons’ flights using GPS trackers strapped to their backs. The pigeons were released in pairs, with new, “naïve” pigeons flying alongside more experienced pigeons, which had flown the 8.4-kilometer route before. After 12 flights together, the naïve bird took the experienced bird’s place and was given a new naïve partner, which could then learn from it in turn. Because the birds improved their routes over time, the authors believed that pigeons might be using a complex process called cumulative cultural evolution (CCE), in which a group of animals learns from each other to improve their task performance over generations by accumulating beneficial behaviors.

“We have a lot of evidence of this in human culture, but not much in non-human animals, so this was a very significant paper,” says Banerjee. After a long brainstorming session at a research retreat, Banerjee and his colleagues decided to look deeper at exactly how the pigeons were learning.

A homing pigeon, like those studied in the 2017 paper, in flight.

The team recreated the experiment using virtual birds. They created seven strategies, each increasing the cognitive ability required by the birds, to test what strategies the real birds were using.

In the most basic model, when a pair of virtual pigeons plotted a route to their goal, they simply took the average of their two preferred routes. More complex simulated birds could recognize which of the pair had more experience, and gave more weight to that bird’s decisions. The birds with the most complex learning model, which fulfilled the requirements for CCE, also had the ability to evaluate their own performance, both individually and as a pair, and make adjustments in mid-flight.

As Kao explained, “This study bridges two different worlds. There’s the world of collective intelligence, in which the mechanism of simple averaging is the default assumption. Then there’s this other world of CCE, which assumes something more complicated, where the birds can judge the quality of different strategies and identify the one that’s better, and ditch the one that’s worse.”

When Banerjee ran the models, he found that all the different learning strategies led to more efficient flights over time. To find out which model was the closest to that used by real homing pigeons, they compared the virtual flights to data from the 2017 study.

To the researchers’ surprise, the homing pigeons’ flights were closest to those made by the simplest learning model. When released to fly home, the two real pigeons just took the average of their preferred routes.

“We found that all these complexities that we were creating around pigeon navigation didn’t seem to hold,” Banerjee said. “It was a pretty shocking result, but I really appreciate when, in nature, simpler learning mechanisms that have been studied for so many decades can actually lead to improvements or efficiency increase in these systems.”

That pigeons’ simple strategy isn’t a failing, he suggests, but an adaptation. Rather than trying to gauge the experience of every other bird, they can rely on the knowledge of the whole flock, while saving resources. “I’m trying to understand different tradeoffs and learning mechanisms, and how animals choose one over another,” Banerjee says.

He’s curious whether the pigeons’ strategy makes even more sense in large flocks, where more birds’ experience can be pooled. He also hopes to collect more data on wild birds, unburdened by GPS backpacks.