Suppose a school of fishes suddenly turns. Who initated this turn? Was this leading fish acting first because of the topology of the swarm, him being closest to the cue? Or do individuals exist with a higher probability of leading? What properties do leaders have? Is it their looks, or their behavior that make the others follow?

Using a biomimetic fish robot we are able to test various hypotheses: we can make the Robofish thin or big, act risk-averse or adventurous, nervous or calm. Having full control over the hypothesis, experiments are perfectly reproduceable.

In the project RoboFish, as a joint project of Freie Universität Berlin (Rojas group) and Leibniz Institute of Freshwater Ecology (Krause group), we are developing a biomimetic fish swarm for the investigation of swarm intelligence in fish schools.

NeuroCopter is a joint project of the Biorobotics Lab and the Neuroinformatics group at Free University Berlin. Our goal is to control an autonomously flying robot with a brain simulation. We want to develop the neural control architecture for “high-level” tasks such as learning, memory and navigation. The design of this neural architecture will follow an insect model, the honey bee.

The robot will learn to navigate in previously unknown terrain by relying on sensory modalities of bees, i.e. measurements of stereoscopic optical flow, polarization of the sky, color and geometrical terrain features and odor detection. The central brain of the robot is equipped with a spiking neural network of deep architecture where learning is expressed locally by different forms of synaptic plasticity and metaplasticity, reinforced through external and internal rewards. The central network will compute behavioral decisions that drive low level motor plans.

The first version of the robot will rely on simulated neural circuitry; the final version will be equipped with neuromorphic electronics running a network of spiking neurons in hardware structures.

Bees have amazing cognitive capabilities. The scouts explore and learn the terrain and build up a highly detailed neuronal representation of the surrounding environment. Once back in the hive, locations are communicated to others by translating location information into body movements; a behavior known as waggle dance. In our experiments with a honeybee robot we noticed that some foragers seem to prefer certain nestmates in the process of decoding the dance. Do bees form stable “peer groups” throughout their lives? What determines that one bee becomes a “friend”?

In the past, answering those questions was very laborious. Biologists would sit in front of an observation hive, keeping track of single marked bees and take note with whom they dance. We are taking this analysis to the next level. We develop a system which allows tracking every single individual inside the hive. We use unique tags to mark each bee and develop a machine vision system to find every bee in the hive. We can then tell which bees are communicating one with whom and where and how long and so on. The analysis of the complete social network has never been done previously. We are happy to work with our partner Zuse Institut Berlin where all our data (~190 Terabyte) is stored and analyzed on supercomputers.