From bird flocks to fish schools, from insect swarms to cell colonies, collective behaviour is a very widespread phenomenon in many biological systems. It is hard to define, but easy to recognize. What are the mechanisms regulating collective behaviour in biological systems?
In many cases collective behaviour is the result of local rules of interactions among the individuals, without any need of a centralized control. Whenever this happens, we are in presence of self-organization. It is then natural to ask: What kind of interaction rules can grant such an extraordinary coordination? The study of collective behaviour is not simply restricted to biological systems, but it has a highly interdisciplinary impact. The spontaneous ordering of spins in ferromagnetic materials is a key paradigm in statistical physics. Achieving self-organization in a group of artificial units with distributed intelligence is a crucial problem in robotics. The emergence of herding behaviour is a common phenomenon in economics, finance and social sciences too.
Our aim is to understand the fundamental mechanisms of collective behaviour in biological systems through a strong interplay between quantitative empirical observations and theories.
- Article in PLOS Computational Biology: Starling Flock Networks Manage Uncertainty in Consensus at Low Cost. We find that the reason why starlings interact with 7 neighbours is that this number optimizes the trade-off between the advantage of consensus (favouring many interacting neighbours) and the cost of keeping track of the others (favouring few interacting neighbours). Check the nice layman discussion of this paper in 'News at Princeton'. The original 7 neighbours result can be found here.
- Article in the Proceedings of the Royal Society B: Diffusion of individual birds in starling flocks. We measure how much birds move and change neighbours in the flock's reference frame and find distinct superdiffusive behaviour. However, the diffusion coefficient is rather low, so that the structure of the interacting network does not change much in time. This may be the reason why purely statistico-mechanical approaches, such as maximum entropy, give such good results.
- Article in Physical Review Letters: Boundary information inflow enhances correlation in flocking. The very long ranged spatial correlation functions that we found in flocks, may be due to an information transfer from border to bulk of the flock. This is a purely dynamical effect: the system seems to be always in a self-excited state, so as to maintain extremely long range correlation.
- Brad Hill (Vice President at AOL) is a fan.
- Starlings as `flying magnets' - Wired
- This is a video illustrating life and research in our group, including AC playing basketball (truly embarrassing).
STARFLAG our first project on collective animal behaviour (2005-2007)
Our focus is on studying the collective behaviour exhibited by flocks of starlings (Sturnus vulgaris) and swarms of non-biting midges (Chironomidae) through the analysis of synchronized high speed image sequences from three cameras. Using stereo matching and other computer vision techniques, we are able to reconstruct, in three dimensions, the trajectories of individual animals within the aggregation. Further analysis of the trajectories should lead to a better understanding of the fundamental interaction rules between individuals. More details about our computer vision and tracking research can be found here.
Having completed three seasons in the field collecting flocking events of starlings (more details about the experimental setup), we have completed processing a significant number of the sessions. Using a novel tracking algorithm, we are able to reconstruct the 3D trajectories for large numbers of individuals for nearly the entire duration of an event. The two videos below illustrate the effectiveness of our tracking method. In the left panel, the raw image sequence of the flocking event. In the right panel, the reconstructed 3D trajectories of the same event. They are both best viewed in full screen mode.
Collecting experimental data of collective animal behaviour in their natural habitat is a difficult and time consuming activity. The aggregations tend not to be stationary and the exhibited behaviour cannot be predicted. This is especially true for midges, as they typically swarm over a natural landmark and they are easily disturbed by wind gusts. We have managed to overcome these issues and have successfully completed two sessions of experimental data gathering (for more details about midges and our experiments). We are in the process of reconstructing the 3D trajectories for many events. Below is an example of a typical midge swarm event (best viewed in full screen mode).
Housed in the Physics Department, Sapienza University, the COBBS Laboratory is the central hub for our research.
Massimiliano Viale - computer vision, tracking, maximum entropy method
Stefania Melillo - experiments, computer vision, tracking, data analysis
Lorenzo Del Castello - experiments, computer vision, tracking, data analysis
Edward Shen - postdoc: experiments, computer vision, tracking
Alessandro Attanasi - computer vision, tracking
Asja Jelic - theory, collective decision making
Silvio Duarte Queiros - theory, diffusion
Supravat Dey - theory, maximum entropy
Edmondo Silvestri - PhD student: modelling using GPUs
Leonardo Parisi - PhD student: experiments, computer vision, tracking
- Italian Institute of Technology - Seed Project: Artswarm (2010-2013) to Irene Giardina.
- European Research Council, IDEAS - ERC Starting Grant n.257126: SWARM (2010-2015) to Irene Giardina.
- Air Force Office of Scientific Research (US) - Grant Z809101 (in collaboration with P.S. Krishnaprasad, University of Maryland - 2011-2014) to Andrea Cavagna.