Rebodiment - Robotswarms Emulating Bees explore 2-Dimensional Temperature fields

Honeybees are social insects which exhibit a wide range of collective behaviour. This leads to the emergence of abilities that single individuals wouldn`t be capable of. For example, groups of bees are able to collaboratively find a spot with optimal temperature while single bees fail at this task. We are currently conducting an in-depth investigating of this special aspect of swarm behavior in an FWF-funded research project (FWF - P 19478-B16). REBODIMENT will relate closely to this project, but the swarm system will be examined from a new point of view. We will exploit the results retrieved from behavioural observations to precisely recreate swarms of bees in simulations (computer models) and emulations (robots) and to observe them "from within". With this approach, we expect to deepen our knowledge about the dependency between individual and collective behaviour and the influence of the physical embodiment on the relationship between them. We will design a mathematical model which describes the behaviour of a swarm of bodi- and physicless individuals (agents) with the help of simple differential equations. A multi-agent model will allow us to deliberately simulate bees as bodiless agents without physical environment or as embodied agents which interact with one another and with a simplified simulated physical environment. For an implementation of the behavioural algorithms under physically realistic conditions we will resort to ePuck robots which we will extend by two temperature sensors located at the ends of two antennae. Similar to the bees in the current project, these ThermoBots will move in an arena on the ground of which we will establish a thermal gradient. Due to the comparable embodiment (antennae), the perception of the thermal gradient will be similar to the one that bees experience. We expect to identify a core algorithm in all forms of embodiment which acts as their common foundation of behaviour, while a set of additional, specific parameters adapt the core algorithm to the ultimate embodiment. Additionally, we will develop a new paradigm for programming robotic swarms based on the variability of individual behaviour. This will allow us to control a swarm`s ultimate behaviour by composing it from individuals with different behavioural traits. The results of the projected experiments and the introduction of the new concept for programming swarms will contribute to the solution of technical problems which robot engineers haven`t been able to solve so far. Additionally, we expect to improve our knowledge about the mechanisms that govern the collective behaviour of biological organisms on the most fundamental level.

The objective of REBODIMENT was to investigate principles of individual and collective behaviour of honeybees and to transfer (to re-embody) it to swarms of robots. The main result was a robotic swarm that emulates behaviors of bees in being able to choose their locations based on temperatures in a thermal gradient. Therefore, the robots were programmed with a simple behavioural model to perform a rich set of behaviours and their intermediate (mixed) forms. To achieve this, we developed a method to track walking bees and to extract their motion principles automatically. Our mathematical model of agents motion was derived as a single stochastic differential equation (motion law) and we applied a process of stochastic optimization (evolutionary algorithm) to parameterize the model so that the resulting behaviours were as close as possible to that of honeybees. These behaviours were implemented in simple physic-less particle models, in sophisticated agent-based simulations (some physics simulated) and in a real robotic swarm (real physics). We found that the type composition of groups has a strong effect on the swarms overall performance. To search for an optimal group composition we again used artificial evolution: Optimal group composition in robots closely resembled the group compositions of natural honeybees. This stresses not only the validity of our model and its embodiment into agents, it also generates new insights into the ultimate reasoning of natural evolution in social insects: Diversity among colony members has profound advantages in performance. We here present a novel way to generate artificial swarm systems and to automatically extract behavioural programs of artificial (robotic) agents from natural organisms. Interestingly, swarms are also affected by social gradients, e.g. agents that pursue other goals and thus select the local optimum over the global one. This allows the system to be controlled from the outside. Analogies of this effect can be seen in human crowd control, markets and other social systems where VIP programs can also affect the human swarm significantly. We found that the swarm is quite insensitive to impaired agents within the group, which helps the swarm to be efficient in its diversity, as an agent that is bad in one functionality aspect but might be good concerning another aspect. The properties we found in the honeybee system by modelling, simulation and physical embodiment opened promising new perspectives on future self-organizing technical systems, e.g., smart traffic with autonomous cars, where diversity, flexibility, robustness and collective intelligence will also be emergent properties. On the one hand, the emergence of unforeseen effects can cause problems if they were not anticipated early enough, thus understanding and predicting them is important. On the other hand, they may also offer benefits in terms of added efficiency and robustness. This requires their understanding by the system designers. Or they extract those features from well-evolved natural organisms. Our project laid the foundation of this approach towards the understanding the emergent behaviours arising from agent diversity in cooperative systems.