Machine learning for modular robots self-reconfiguration problem

Self-reconfiguration of modular robots is one of the most challenging problems in the robotics field. The objective is to
determine how a set of identical modular robots, with local knowledge of the system and limited capacities, can reorganize
themselves into a target topology or shape. The problem has received great interest from the research community giving
birth to many centralized and distributed algorithms. However, the lack of comparative study of these algorithms makes it
difficult to choose one when faced with a given configuration. In this paper, we present a kind of high-level hybridization
approach of these algorithms using a neural network technique. The objective is to propose a centralized pre-processing
procedure that allows, according to the self-reconfiguration problem, to determine which algorithm is most suitable. We
applied the Neural Network technique to two self-reconfiguration algorithms: C2SR and TBSR. The obtained results show
that the machine learning tool succeeds 96.67% of the time to determine the suitable algorithm based on the initial and
the final shape. Consequently, using machine learning directly leads to the reduction of the required number of moves for
the reconfiguration.