Paderkicker’s camera, via which the robot observes its environment (pitch, balls and other robots)
As a “Smart Team”, BeBots collaborate with a Paderkicker as self-optimising multiagent system
Organic computing is a sophisticated vision for future information processing systems. It is based on the understanding that we will soon be surrounded by a multitude of autonomous systems, which communicate freely and organise themselves in order to provide actions and services. An “Organic Computing System” is a technical system that dynamically adapts to its environment’s current conditions. It is equipped with the ability to be self-organising, self-configurating, self-optimising, self-healing, self-protecting, self-explaining and context-aware.
The priority programme of the German Research Community (GRC) is concerned with fundamental challenges of the organic computing system’s design; its goal is to gain a deeper understanding for the emergence of global behaviour in selforganising systems and the design of specific concepts and tools, in order to support the construction of organic computing systems for technical applications. The Heinz Nixdorf Institute participates with two projects.
The project "A Modular Approach for Evolving Societies of Learning Autonomous Systems", led by Prof. F. Rammig and Dr. B. Kleinjohann, is about the development of self-organised and self-optimised autonomous systems that exhibit emergent behaviour in groups. Previous approaches are concerned with the question of how single systems treat mistakes and provide first solutions for individual self-adaptation. But it is still a question, how individual selfadaptation influences the behaviour and performance of whole groups. We investigate how a system can learn to adapt to changing environmental conditions and at the same time consider the whole group’s behaviour. A modular approach is pursued, in which a system learns a model of itself and its environment including its group members, in order to decide, which behavioural alternatives are the most promising in specific situations. Among other things, a procedure was developed, which – in reference to the system of the so-called human mirror neurons – enables imitation in robot groups. A mirror neuron is a neuron that fires when man conducts an action, as well as when he observes another human conducting this action. This approach was implemented in robot systems, in which the module responsible for the learning of behaviour, learns to recognise similar behaviour in other robots, at the same time. Currently it is examined how the learning of cooperative behaviour can be supported. The developed modular approach is evaluated simulative and experimentally with the help of the Paderkicker-soccer robotteam, the mini robot BeBot respectively.
In the project “Smart Teams: Local Distributed Strategies for Self-Organizing Robotic Exploration Teams”, led by Prof. F. Meyer auf der Heide and Prof. C. Schindelhauer (University of Freiburg, former Member of the Heinz Nixdorf Institute), we want to lay the algorithmic principles for a scenario, in which an exploration team of robots (smart team) has to organise itself, in order to carry out tasks such as the exploration of unknown terrain and the conduct of assignments in this terrain. Such a smart team’s work has to be guided by special strategies for the exploration, as well as for the detection of important objects and their allocation into subgroups of robots, which collectively possess the ability to process the object. The fact that all these tasks have to be locally conducted through distributed strategies, which operate on a mobile network of moving robots and have to result from a team’s robust, effective self-organisation, poses this project’s central challenge. None of these robots will ever have more than a very limited, local knowledge of the system’s global condition. Besides the exploration quality, the continuous securing of the communicative connectivity and the skillful distribution of tasks, it is also necessary to consider the robots’ energy needs. We measure these strategies’ quality theoretically, with competitive analysis, as well as experimentally, with the help of our simulation platform.
