Swarms are characterized in nature by a dynamic behaviour which is quite appealing for researchers involved in numerous fields of study, like robotics, computer science, pure mathematics and space sciences. Global group organization acquired in absence of centralized control is the feature of natural swarms which is most interesting to reproduce. This study proposes to make use of some evolutionary robotics findings in order to obtain the autonomous group organization in the framework of a deeper knowledge of the astrodynamics. The main task which will be accomplished is the implementation of the control laws for the single satellite. A careful tuning of the parameters at member level is necessary in order to gain an autonomously evolving global behaviour in a number of space missions of immediate interest. In remote sensing missions, for example, trains of a small number of satellites are already orbiting and integrating their collected data: in near future entire swarms of agents could accomplish this task, and should be controlled in order to acquire and maintain the desired leader-follower configuration. Another example can be seen in deep space exploration of unknown celestial bodies, where the migration of the entire swarm from a reference orbit to a (previously unknown) targeted one is an issue; the same group migration is of interest in Earth orbit, when transferring from parking to operational orbit. Finally, self-assembly of rigid-like virtual structures is also simulated. This paper shows that all these cases are autonomously performed by the swarm by correctly implementing four simple rules at individual level, which assess the primal needs for any satellite: avoid collision, remain grouped, align to the neighbor, reach a goal