Situative Teams in Cooperative Autonomous Systems

Distributed systems have been established in many areas of IT and will play an even more significant role in the future. Such systems are no longer limited to specific fields of application but interconnect many different domains. They encompass, e. g. Cloud Computing, Internet of Things, service robotics, and autonomous vehicles. The integrated sub-systems communicate in order to exchange information and, if necessary, perform tasks together. Moreover, the number of interconnected sub-systems is constantly growing. That makes it challenging to administrate them without a certain degree of autonomy. Furthermore, these systems should also be able to handle new tasks without previously specified adaptations. The demand for solutions that consider these circumstances is therefore growing at the same rate. It leads to the question of what such a system could look like. The novel approach presented in this thesis focuses on distributed systems composed of idle executing units in the initial state, i. e. they do not necessarily execute a task or pursue a goal. Instead, tasks assign themselves to units under certain conditions at runtime. Therefore, tasks themselves specify these conditions. Furthermore, tasks can decide independently about their assignment to executing units. Thus the task assignment takes place depending on the currently provided properties and capabilities of individual units and by considering the needs of the tasks. As a result, single tasks are not necessarily tailored to specific units. In addition, dependencies between tasks lead to automatic team formations. They dissolve after all tasks have been completed. The provided solution approach defines several models. Therefore, the approach adopts a unique perspective on the organisational structure. The first model specifies relationships and dependencies of properties, abilities, and behaviours. The second model describes the assignment of tasks to suitable executing units. Finally, the third model considers the reusability of behaviours to become more independent of predefined definitions and enable existing executing units to fulfil previously unknown novel tasks. Furthermore, it opens the possibility of evaluating the quality of the execution and, if necessary, suggesting more appropriate alternatives for behavioural implementations, which requires adequate learning mechanisms.