Collective Behaviour︱ Guy Theraulaz & Lei Liu A spontaneous change of swimming direction by a single fish (right). Schooling fish only

A group of two rummy-nose tetras pay attention to (Hemigrammus rhodostomus). Photo credit: David Villa ScienceImage/CBI/CNRS, Toulouse.

Group coordination (schooling behaviour) in a school few neighbours for of rummy-nose tetras (Hemigrammus rhodostomus). coordinating their Initiation of a collective U-turn in a school of rummy-nose tetras (Hemigrammus collective movements rhodostomus). these brief accelerations, a fish filters the This method may sound simple, but STRATEGIES FOR INTERACTION How do fish combine hat do flocks of birds, Researchers recognise that fish would information coming from its environment the team determined it was enough Over the past few years, and after information from multiple swarms of insects, herds need an incredible cognitive ability to and picks its resulting new heading”, to explain how slight changes in many experiments following swimming neighbours when swimming Wof wildebeest, schools of continually monitor the movements of explain Dr Liu and Dr Theraulaz. direction in a small number of fish could patterns in fish, Dr Liu, Dr Theraulaz in a school? Dr Lei Liu and Dr fish and even crowds of pedestrians a large number of neighbours. It’s easy drastically change the whole group. For and their team developed a general Guy Theraulaz based at the have in common? They can all show to see that it’s a lot simpler for these After hours of observing fish swim example, in the wild, fish may be forced method to extract from tracking data Research Centre on Animal collective behaviours that can be quite animals to only pay attention to a small either alone or in pairs, the team to move from travelling in a specific the interactions between fish that are Cognition, in Toulouse, France, complex. Individuals can respond number of neighbours. characterised what clues fish may use direction (schooling) to adapting a involved in the coordination of their used both experiments to opportunities or threats almost and measured how their behaviour more protective circular swimming collective movements. They also used with live animals, as well as simultaneously giving the impression of Dr Lei Liu and Dr Guy Theraulaz based changed. On one hand, fish avoid each (milling) when a predator is detected. this method to develop a computer computational and robotic operating as a single super-organism. at the Research Centre on Animal other when the distance was less than The researchers also detected what model to predict the collective modelling to show that Such coordinated response needs Cognition, in Toulouse, France, are two body lengths, behaviours of these using information about two specific interaction rules together keen to disentangle these neighbourly but they also animals. Using neighbouring fish is sufficient with efficient and fast transfer of interactions, particularly in fish schools. didn’t like it when this model, the to establish appropriate They can limit their attention to a small information among all individuals. Recently, the team has proved that fish the distance was researchers tested collective behaviour. It’s However, understanding exactly how only need to follow one or two other higher than six to three different important, however, to pick the set of the most influential neighbours, this information is shared between fish to perform collective U-turns. Now, seven body lengths strategies to assess most influential neighbours, i.e. neighbours is proving a little elusive. the researchers want to find out if the and actively tried immediately identifying which ones to how each individual the ones that show the biggest impact when choosing the same pattern is valid when fish are to get closer. For avoid and move away and which ones fish decides which school’s direction of travel. For example, in a group of hundreds moving together and not performing most situations, neighbours to pay of wild fish, how does each animal joint manoeuvres. fish aligned their to follow. attention to. The determine which neighbours to pay direction with three strategies attention to? For researchers, this is SWIMMING AROUND their group mates, especially when is known in the fish world as counter- included selecting the nearest fish, a

CK Ma/Shutterstock.com CK the crucial element to understand how AND AROUND the distance between the fish was a milling behaviour, where occasionally random fish or the fish that exerted the these groups of animals can coordinate For the researchers, this journey started comfortable three body lengths. The some fish started swimming in the largest influence on their behaviour. their movements and how information in a circular tank looking at rummy-nose influence of a fish on another fish opposite direction to the rest of the propagates within these groups. tetra fish (Hemigrammus rhodostomus) also depends on the direction it is group, to ensure that individuals Dr Liu and Dr Theraulaz were impressed swimming around and around. This moving in, and especially the “angle swapped their positions at the front. with the results. “The simulation results The thinking has changed over the years species of tropical fish is fascinating to of vision” with which it perceives the clearly indicate that group behaviours to answer this question. The earlier study because they swim in a highly other fish. Finally, the intensity of social In addition, fish reacted to obstacles can be reproduced by our model, not models considered that each fish was synchronised manner alternating periods interactions between two fish depends in their environment. In this case, the only qualitatively but also quantitatively, influenced by all the neighbours located of bursts in which each fish accelerates on the direction of their movement with tank wall was the obvious barrier. The provided that individuals interact with within a certain distance. In contrast, and changes direction with more respect to each other. This perception researchers found that fish quickly at least two of their neighbours at each more modern approaches recognise subtle gliding phases. This intermittent anisotropy leads to asymmetry in what changed their direction if heading decision time”. Somewhat counter- that the movement of each individual in movement is ideal for analysing is known as the “social force” exerted to the wall at a 45-degree angle, but intuitively, the researchers found that the group is more likely influenced by a trajectory as a series of decisions about on fish A by fish B and that exerted on these animals were happy to swim adding more fish to follow didn’t really small number of neighbours. which direction to take. “Just before fish B by fish A. parallel to the edge. improve the results. In fact, it seems

www.researchfeatures.com www.researchfeatures.com coherent collective motion over long periods of time even when the robots only interact with their most influential Behind the Research neighbour. On the other end of the scale, picking the nearest fish didn’t boost group coordination for only one neighbour, and had only marginal Dr Liu Dr Guy improvements for two neighbours. Lei Theraulaz GETTING INSIDE THE FISH BRAIN E: [email protected] W: https://crca.cbi-toulouse.fr/en/ In all vertebrates – and in particular in guytheraulaz/ fish – the midbrain and the forebrain @GTheraulaz Photo credit: David Villa ScienceImage/CBI/CNRS, Toulouse. are heavily involved in processing visual information and selecting which external stimulus should be the focus of attention. Research Objectives The midbrain continuously monitors the Lei Liu and Guy Theraulaz’s research interests include swarm intelligence in natural and artificial systems, self-organisation environment for clues. This is the primary in biological and robotic systems, collective behaviours and collective intelligence in animal and human societies, computational and location where information coming from systems biology. neighbours is collected and then passed on to the forebrain, which is responsible for picking the best stimuli on which the Detail fish must focus their attention. Liu Lei He is specialised in complex systems Foundation of Shanghai under Grant University of Shanghai for Science and control, such as collective motion No.17ZR1419000 and Visiting Fund of Using this cognitive mechanism, fish Technology (USST), Shanghai, China in swarms of robots, intelligent Shanghai Education Commission. can be very good at filtering relevant transportation systems. information from their surroundings. Guy Theraulaz Guy Theraulaz gratefully acknowledges They can limit their attention to a small Centre de Recherches sur la Cognition Guy Theraulaz is a research director the Indian Institute of Science to serve as set of the most influential neighbours, Animale, Centre de Biologie at CNRS. He is an expert in the study Infosys visiting professor at the Centre immediately identifying which ones to Intégrative, Centre National de la of collective animal behaviours and for Ecological Sciences in Bengaluru. Combining computational and robotic approaches to investigate the impact of different strategies for a avoid and move away and which ones to fish to interact with its neighbours on collective swimming in groups of rummy-nose tetra fish. Recherche Scientifique (CNRS), a leading researcher in the field of follow. These neighbours which trigger Université de Toulouse—Paul Sabatier swarm intelligence. Collaborators that quite the opposite was true. Most but it was only when robots interacted and immediate action – can set in motion (UPS), Toulouse, France Ramon Escobedo (CRCA/CBI/ interactions needed two neighbours, but with three of their nearest neighbours a larger response than other neighbours, Funding CNRS), Clément Sire (Laboratoire for the most influential neighbour, one that this strategy produced a highly making the concept of most influential Bio Liu Lei was supported by a de Physique Théorique, CNRS and interaction was enough. cohesive and coordinated group. neighbours easy to understand. Lei Liu is a USST associate professor. grant from the Natural Science Université de Toulouse). Remarkably, it is possible to obtain ROBOTS VS FISH group cohesion and coherent collective All combined, Dr Liu’s and Dr Theraulaz’s To complement this computational motion over long periods of time even results show that fish typically interact References Personal Response approach, Dr Liu and Dr Theraulaz when swarm robots only interact with with their two most influential neighbours. also used robots to check if they could one most influential neighbour. This selection reduces the amount of Lei L, Escobedo R, Sire C, Theraulaz G. (2020). Can the same methods be applied to other animals that get robust collective movements as information that needs to be processed in Computational and robotic modeling reveal parsimonious study their collective behaviour, like insects or birds? the model predicted, when physical Finally, Dr Liu and Dr Theraulaz wanted the brain and avoids cognitive overload. combinations of interactions between individuals in The general methodology and the procedures constraints are present and when each to compare the predictions of the “Our study thus suggests that each schooling fish.PLoS Comput Biol 16: e1007194. that we used in fish can be similarly applied on any robot must control their speed to avoid computer models and robot swarms with fish must acquire a minimal amount of set of trajectories of other organisms, including information about the behaviour of its Escobedo R, Lecheval V, Papaspyros V, Bonnet F, Mondada humans to measure social interactions between neighbours for coordination to emerge F, Sire C, & Theraulaz G. (2020). A data-driven method for individuals. Understanding how the interactions …each fish must acquire a minimal at the group level”, concluded the reconstructing and modelling social interactions in animal between individuals in swarms of insects, schools researchers, “thus allowing fish to avoid groups. Philosophical Transactions of the Royal Society of of fish, flocks of birds, herds of ungulates, or human crowds give rise to the ‘collective level’ properties amount of information about the information overload when they move London - Serie B., 375, 20190380. requires the development of mathematical models. in large groups. Besides our findings will behaviour of its neighbours. Our methodology ultimately leads to concise and benefit to the design of autonomous Calovi DS, Litchinko A, Lecheval V, Lopez U, Pérez explicit models that can be exploited to understand collision with other robots. Similarly the experiments conducted under the swarms of micro-robots”. Escudero A, Chaté H, Sire C & Theraulaz G. (2018). and explain diverse experimental features and to the computer model, the group of same conditions with groups of live fish. Disentangling and modeling interactions in fish with burst various forms of collective behaviour and that have robots remained close and cohesive by “Overall, and even more convincingly Excitingly, these findings can be exploited and coast swimming a predictive power. interacting with their most influential than in the case of the fish model”, said as a source of inspiration to coordinate reveal distinct neighbour. In contrast, the group lost the researchers, “the most influential the actions of artificial systems, such as alignment and all kind of coordinated behaviour strategy leads to the best overall swarms of drones that in the future might attraction behaviors. by picking the nearest neighbour agreement when fish focused on one become increasingly used for search Plos Computational instead. In this case, using at least two or two neighbours”. Remarkably, it was and rescue operations, environmental Biology 14: neighbours improved their behaviour, possible to obtain group cohesion and and wildlife monitoring. e1005933.

www.researchfeatures.com www.researchfeatures.com