Robust Swarm Robotics System Using CMA-NeuroES with Incremental Evolution

Kazuhiro Ohkura, Tian Yu, Toshiyuki Yasuda, Yoshiyuki Matsumura and Masanori Goka

Swarm robotics (SR) is a novel approach to the coordination of large numbers of homogeneous robots; SR takes inspiration from social insects. Each individual robot in an SR system (SRS) is relatively simple and physically embodied. Researchers aim to design robust, scalable, and flexible collective behaviours through local interactions between robots and their environment. In this study, a simulated robot controller evolved by a recurrent artificial neural network with the covariance matrix adaptation evolution strategy, i.e., CMANeuroES is adopted for incremental artificial evolution. Cooperative food foraging is conducted by our proposed controller as one of the most complex simulation applications. Since a high level of robustness is expected in an SRS, several tests are conducted to verify that incremental artificial evolution with CMANeuroES generates the most robust robot controller among the ones tested in simulation experiments.