Learning of intelligent swarm behavior

Principal Investigators:

Henning Sprekeler
Pawel Romanczuk
David Bierbach (Co-PI)
mail: bierbacd@hu-berlin.de

Team members:

Luis Alberto Gómez Nava (Postdoctoral researcher)
Robert Tjarko Lange (doctoral researcher)

When is it ecologically beneficial to act as a collective, or to develop diversity within a swarm? When is it best to act alone?

Research Unit 1, SCIoI Project 12

A central challenge in both understanding and developing swarm intelligence is the relation between the behavior of a swarm of agents and its ecological niche. The goal of this project is to combine analytical behavioral characterizations of biological swarms (specifically, fish) in different ecological settings with synthetic state-of-the-art machine learning methods that optimize the behavior of a multi-agent system for a given task (e.g., (deep) multi-agent reinforcement learning).

The ideal outcome on the analytical side would be a normative, ecological interpretation of observed swarm behavior as an adaptation to an ecological niche, rather than classical phenomenological models that derive collective behavior from the behavior of the individuals.

For example, the suggested optimization approach to swarm behavior can provide insights regarding when it is ecologically beneficial to act as a collective, when to develop diversity within a swarm, and when to act alone.

The ideal synthetic outcome would be the development of novel learning algorithms for artificial swarms of, e.g., robotic agents, that allow a behavioral optimization for a given ecological niche, i.e., an ensemble of tasks. The project aims at establishing essential methods for the implementation of the “collective shepherding” behavior. The developed methodology will serve as a starting point for incorporating priors for specific ecological niches, as well as the sensory and motor aspects of the agents in the swarm.



Related Publications

Winklmayr, C., Kao, A. B., Bak-Coleman, J. B., & Romanczuk, P. (2020). The wisdom of stalemates: consensus and clustering as filtering mechanisms for improving collective accuracy. Proceedings of the Royal Society B: Biological Sciences, 287(1938), 20201802. https://doi.org/10.1098/rspb.2020.1802
Rahmani, P., Peruani, F., & Romanczuk, P. (2020). Flocking in complex environments—Attention trade-offs in collective information processing. PLOS Computational Biology, 16(4), e1007697. https://doi.org/10.1371/journal.pcbi.1007697
Lukas, J., Romanczuk, P., Klenz, H., Klamser, P., Arias Rodriguez, L., Krause, J., & Bierbach, D. (2021). Acoustic and visual stimuli combined promote stronger responses to aerial predation in fish. Behavioral Ecology, arab043. https://doi.org/10.1093/beheco/arab043
Lukas, J., Auer, F., Goldhammer, T., Krause, J., Romanczuk, P., Klamser, P., Arias-Rodriguez, L., & Bierbach, D. (2021). Diurnal Changes in Hypoxia Shape Predator-Prey Interaction in a Bird-Fish System. Frontiers in Ecology and Evolution, 9. https://doi.org/10.3389/fevo.2021.619193
Jolles, J. W., Weimar, N., Landgraf, T., Romanczuk, P., Krause, J., & Bierbach, D. (2020). Group-level patterns emerge from individual speed as revealed by an extremely social robotic fish. Biology Letters, 16(9), 20200436. https://doi.org/10.1098/rsbl.2020.0436
Bierbach, D., Krause, S., Romanczuk, P., Lukas, J., Arias-Rodriguez, L., & Krause, J. (2020). An interaction mechanism for the maintenance of fission–fusion dynamics under different individual densities. PeerJ, 8, e8974. https://doi.org/10.7717/peerj.8974

User registration

You don't have permission to register

Reset Password