Bringing reality to evolution of modular robots: bio-inspired techniques for building a simulation environment in the SYMBRION project

Abstract — Two neural network (NN) based techniques for building a simulation environment, which is employed for evolution of modular robots in the Symbrion project, are presented in this paper. The methods are able to build models of real world with variable accuracy and amount of stored information depending upon the performed tasks and evolu-tionary processes. In this paper, the presented algorithms are verified via experiments in scenarios designed to demonstrate the process of autonomous creation of complex artificial organ-isms. Performance of the methods is compared in experiments with real data and their employability in modular robotics is discussed. Beside these, the entire process of environment data acquisition and pre-processing during the real evolutionary experiments in the Symbrion project will be briefly described. Finally, a sketch of integration of gained models of environment into a simulator, which enables to fasten the evolution of a real complex modular robot, will be introduced. I