Autonomous agents and multi-agent systems (MASs) represent one of the most exciting and challenging areas of robotics research during the last two decades. In recent years, they have been proposed for several applications, such as telecommunications, air traffic mangement, planetary exploration, surveillance etc.. MASs offer many potential advantages with respect to single-agent systems such as speedup in task execution, robustness with respect to failure of one or more agents, scalability and modularity. On the other hand, MASs introduce challenging issues such as the handling of distributed information data, the coordination among agents, the choice of the control framework and of communication protocols. This thesis investigates some problems that arise in the management of MASs. More specifically it investigates problems of designing decentralized control schemes to manage collections of vehicles
cooperating to reach common goals, while simultaneously avoiding collisions.
An existing decentralized policy for collisions avoidance, already proved safe for a system with three agents, has been extended up to five agents.
A new decentralized policy, the Generalized Roundabout Policy, has been
designed and its properties analyzed. Specifically safety and liveness properties have been studied. The first one has been proved formally, while the second has been addressed by means of probabilistic approaches. Moreover, it is addressed the problem of optimization of autonomous robotic exploration. The problem is clearly of great relevance to many tasks, such as e.g. surveillance or exploration. However, it is in general a difficult problem, as several quantities have to be traded off, such as the expected gain in map information, the time and energy it takes to gain this information, the possible loss of pose information along the way, and so on.
Finally, software and hardware simulation tools have been developed for the analysis and the verification of the decentralized control policies.
Such instruments are particularly useful for the verification of multi-agent systems which could be overwhelmingly complex to be addressed purely by a theoretical approach.