In automation the trend goes away from hierarchically rigidly automated systems to intelligent, distributed and flexible architectures. For this purpose sensor actuator networks must be able to handle cooperative tasks under constraints, which are not known sufficiently in advance and which can change dynamically during the execution. With the aspired cooperation capabilities the system complexity and the requirements regarding performance, robustness and scalability raise. In this context efficient sensor actuator networks are very important where individual nodes can cooperate in a decentralized manner and can communicate in real time with each another. Therefore technologies and methods are required in such a way that actuator networks can configure themselves automatically, organize themselves autonomously and the individual nodes are able to cooperate in a decentralized way for the reaching of a given goal.
State of the art
In the context of a set of research work, strategies were developed so that autonomous sensor actuator knots can perform complex tasks by means of cooperation and interaction. This coordinated handling of tasks can be achieved by three approaches:
- The knots of the sensor actuator network are steered centrally by a control system (central approach).
- The knots of the sensor actuator network cooperate over a fixed protocol from which the global behaviour can be directly obtained. (distributed approach, e.g. Contract Net).
- The knots of the sensor actuator network work together in accordance with their behaviour patterns, whereby the global behaviour is achieved via emerging phenomena by a suitable selection of behaviour primitives (decentralized approach). Cooperation is done by means of sensor technology or over the environment.
All three approaches contain different advantages and disadvantages. Because of the desired flexibility in the context of this project, distributed and decentralized methodologies will be taken up, which are pursued isolated by different research groups at present. Based on given tasks, a goal of the project is on the one hand to understand the advantages and disadvantages of the approaches and on the other hand to exploit synergies between these approaches.
In this subproject, methods for self configuration, self organization and cooperation for sensor actuator networks are to be investigated and developed. Swarm techniques and swarm models are pulled up as basis for self organization and cooperation in sensor actuator networks. Since swarms act without a central control instance, the special attention of the planned research work lies on the investigation of decentralized and distributed approaches. Among other things following questions are to be treated:
It will be examined, how individual knots can take over subtasks and accomplish them autonomously on the basis of their abilities and on the basis of the respective context. The allocation of subtasks to certain knots is derived from a global task and defined firmly for one period or specified by a negotiation process respectively it will be dynamically specified again when constraints and targets change.
Emergence phenomena are to be achieved by collective behaviour with variable distributions of roles of the sensor actuator knots. It will also be examined, under which conditions the autonomous interaction of knots leads to an efficiency increase respectively a performance degradation of the overall system. Investigations to scalability are to point out, to what extent cooperation and self organization mechanisms scale both for small sensor actuator networks with up to 1000 knots and for very large networks with up to 1 million knots.
The loss of an individual sensor actuator knot may have ideally no effect on the overall system, i.e. a "single point of failure" does not exist. For the autonomous treatment of error situations robust and realtime capable methods are developed. In this context there are also methods developed, which prevent possible deadlocks respectively a stagnation of the co-operation and repair it autonomously.
Direct and indirect communication and interaction approaches for complex group behaviour are examined, i.e. how can knots communicate and interact locally with their neighbours over sensor technology or over networks respectively how can knots communicate and interact over the common environment. It will be examined, how sensor actuator networks can learn successively and react adaptively to new tasks by means of experience and interaction, for example by positive and negative feedbacks from the environment. In cooperation with other subprojects suitable architectures and software reference models for the control of sensor actuator networks are examined and developed. A strong emphasis is put on the development of model-based tools and methods that assist and simplify the design-phase of emergent swarm behaviour in systems of cooperating sensor-actuator-nodes. In doing so, new methods for the modelling of sensor-actuator-networks are to be developed and existing methods shall be extended. In addition, an algorithmic treatment will be helpful in understanding the efficiency of such approaches. The practicability of the won theoretical bases is to be evaluated on the basis of self organizing robot swarms. The developed cooperation strategies are evaluated both on the basis of tasks and applications in the macro world (production robots) and in the micro world (micro/nano-robots).
To realize an efficient interaction with the environment, the local controller must be supplied with a good model of it. This gives a strong connection with the subproject I1: "Decentralized reconstruction of continuous distributed phenomena based on discrete measurements". By using the results of the subproject at hand, new methods of realizing the positioning of the sensor nodes can be developed, which are needed in subproject I4: "Sensor Scheduling and Routing in Sensor-Actuator-Networks". To deal with the communication aspects in the modelling it is intended to collaborate with subproject K2: "Algorithmic aspects of clustering, topology control and localization".