In cooperative and collective robotic systems, a swarm as a whole can handle situations and solve problems that single robots cannot, although each of the single robots of the system is completely autonomous. Single robots can exploit their physical conjunctions like docking elements for example to join together into artificial organisms which are capable of dealing with a changing environment and challenges that are too complex for a single robot. This thesis is part of the Symbrion and Replicator project funded by the European Commission and investigates whether robots can be forced to develop cooperative behavior in a swarm by using an evolutionary approach. For this purpose we consider a task that can only be performed by two or more robots cooperatively, like a rescue scenario, bilateral docking or coordinated locomotion. Because the fitness function according to this task can be very complex, we first break the whole task into simpler ones and examine them by the use of related work and modern approaches as for example CGE, the Common Genetic Encoding or EANT, the Evolutionary Acquisition of Neural Network Topologies. All applied approaches are biologically motivated and increase the level of realism in neural simulations. Afterwards, these approaches will be applied to more complex scenarios. Because a serial artificial evolution of individual robots on a physical robot might require quite a long time, the system is first tested using a simulated scenario. Based on the existing evolutionary framework, the controllers will be evolved and evaluated online and onboard. This thesis concludes with presentations and sample scenarios which illustrate the evolved coordinated behavior in solving a variety of different tasks