Techniques for Controlling Swarms of Robots

Abstract

This thesis addresses the problem of controlling very large groups of robots, refereed as swarms. Scalable solutions in which there is no need for labelling the robots are proposed. All the robots run the same software and the success of the task execution does not depend on specific members of the group. Robustness to dynamic addition and deletion of agents is also an advantage of our approaches. In the first methodology, we model the swarm as a fluid immersed in a region where a field of external forces, which is free of local minima, is defined. In this case, the Smoothed Particle Hydrodynamics (SPH) method is applied to model the robotic fluid', more specifically, to model the interactions among the robots of the group. The Finite Element Method (FEM) is also used in this work to compute the fields that determine external forces. This approach is instantiated in a pattern generation task and also in a coverage task. In the second methodology, a problem of optimal environment coverage using robots equipped with sensors is addressed by means of tools from the Locational Optimization theory. Three important extensions of well-known results in the literature are presented: (i) sensors with different footprints, (ii) disk-shaped robots, and (iii) nonconvex polygonal environments. Both approaches are verified in simulations. The first technique is also implemented and tested in actual robots.