Birds can teach the gliding drones of the future

Humans have always learned a lot from animals, and AI probably has a lot to learn from them too. For example, drones could save a lot of energy if they could harness the upward air currents that birds have in their “little fingers”. This is exactly what the MTA-ELTE Research Group on Airborne Group Behaviour is doing: using the latest technologies to study the mechanisms of collective behaviour in animals, and then trying to incorporate the knowledge gained into even more modern technologies.

Together, many animals can solve far more complex problems than would be expected from a simple combination of individual abilities. However, the behaviour of groups of many individuals is very difficult to study using classical ethological methods. However, physics, statistical physics and state-of-the-art computational techniques can provide a solution to this problem, as they can now be used to model the behaviour of the whole group and the variation in environmental factors in a very realistic way, assessing many factors simultaneously.

Máté Nagy, a research fellow at the Department of Biological Physics at ELTE and head of the MTA-ELTE Swing Group Behaviour Research Group, is working on this. The researcher returned from the Max Planck Institute of Ornithology to continue his research, in part, on behavioural mechanisms in birds and how this knowledge can be incorporated into algorithms for autonomous devices such as robotic aircraft. The research team works closely with the Max Planck Institute for Animal Behaviour, the University of Oxford and several other departments at ELTE.

“We are trying to automate classical ethological methods and try to analyse and interpret the very large amount of data we collect in novel ways. To achieve these goals, we often need to develop new methods, as we are dealing with big data type tasks,” said Máté Nagy. “We humans can learn from the problem solving of animal groups. There are many common problems where the question is whether the mechanisms that work well in groups of a few people are also effective in large groups,” he added.

For example, there is the question of whether new behaviours need to be reflected at the level of individuals for the group to function optimally, or whether it is even optimal to function together. The question is therefore how local behavioural rules between individuals lead to community-wide functioning. The amazing thing here is that very simple local rules can lead to complex behaviours at the group level that individuals are unaware of or would have no chance of understanding.

A good example is the building of a fortress by ants

Although there is no ‘designer ant’ that can see the whole process and be responsible for guiding the other ants, they can work together to build complex structures with ventilation systems and rooms optimised for different functions. It is often through similar joint solutions that flocks of birds find efficient pathways, or find thermals, upward air currents, in a coordinated way. Through a collaboration with the Max Planck Institute, data recorders have also been attached to hundreds of wild storks to continuously collect data.

“One of our main objectives within the Flight project is to study the group gliding of birds. While migratory birds fly up to thousands of kilometres in a single migration, they use the thermals they take off to gain altitude and then glide to stay airborne without wing flapping. What makes this interesting is that, from a physicist’s point of view, air can be seen as a multidimensional flow field, which also changes over time and contains fluctuations. It is in this medium that individual birds fly, and by observing each other they can gain information about this complex environment. One of our main questions is how they achieve this and how we can use the data from them to model the structure and composition of atmospheric updrafts at high resolution,” he explains.

In the simplistic model of thermals, these rising hot air currents act as a rising column of smoke. In reality, however, there are highly complex processes that birds routinely fly over and optimise their behaviour to these conditions. Winds blowing in different directions at different altitudes shift the thermals sideways. Birds have to adapt to this, and their position within the updraft also determines the efficiency of their climb. The bird must therefore constantly adapt its behaviour to the environmental stimuli. It can do this on the basis of its own perception of physical stimuli, but it can also gain valuable information from observing the behaviour of other birds in the group.

Of course, every bird must find a balance between copying its mates and making individual decisions. Neither extreme is good: if each bird follows its neighbour, the whole flock is just as smart as the one that is followed by the others. It is also no good if everyone follows his own lead, because then they cannot take advantage of the information benefits of group life.

Researchers have also found in the group behaviour of birds – and rats in other contexts – that there is a balance between individual choice and imitation of the choices of mates. Máté Nagy said that although rats and their search in the maze have been studied since the beginning of behavioural research, no one had done experiments on their group search behaviour before. This is all the more surprising as rats naturally live in groups and forage for food.

In the experiment, they found that when the animals meet at intersections, although they appear to be zigzagging back and forth randomly (i.e. engaging in individual exploratory behaviour), this is not what actually happens. Statistical analysis has shown that the rodent follows an efficient search strategy on its own (for example, it remembers junctions it has already visited), but it also observes its companions and chooses a direction at a junction depending on their location.

From birds to self-driving algorithms

Studies can both reveal the mechanisms of group search and flight, and increase our understanding of the internal structure of thermals. Thirdly, as an applied branch of the studies, the behavioural rules revealed could potentially be incorporated into algorithms for robotic flight.

In theory, this could allow gliding drones to use thermals in the same way as birds, reducing their energy consumption and allowing them to stay in the air for much longer. Some birds take to the skies almost immediately after hatching and live their lives gliding, without flapping their wings. But this is still a long way off, because the internal structure of termites is incredibly complex and difficult to model, and therefore difficult to train the algorithm for. Birds, on the other hand, excel at this task, so we can learn their tricks.

“We’re trying to use the data from the birds to create a simulation in which the self-driving algorithm of the simulated drone can learn how to recognise and use thermals most efficiently. This will allow us to test a relatively well-functioning self-guiding system in the real world later on. This has its risks, as a wrong decision could result in a crash. The drone could even observe the birds flying around it and use their behaviour to learn about the thermals,” the scientist explained.