Machine learning for energy efficient modular robots self-reconfiguration

While Modular Robots Systems (MRSs) are getting more popular, the challenge of effective self-reconfiguration of the robots persists. Actually, many distributed algorithms were proposed for the Modular Robots Self-Reconfiguration problem (MRSR), however, the efficiency of each algorithm according to the different cases is rarely studied. In a precedent work, we proved the relevance of using an Artificial Neural Network (ANN) component to select the most adapted distributed algorithm according to the current scenario. The proposed approach suffers, as all the other approaches in the literature, ignore the state of charge of the robots. Consequently, an MRSR algorithm may not achieve the MRS reconfiguration if the robots’ batteries are exhausted before the end of the algorithm. In this paper, we propose an ANN system that selects the most suited MRSR algorithm according to the MRSR scenario and the initial state of charge of the robots. The results of the tests show an accuracy of 97%.