Implicit and Robust Evaluation Methodology for the Evolutionary Design of Feasible Robots

This paper deals with the evolutionary design of feasible and manufacturable robots. Specifically, here we address the problem of defining a methodology for the evaluation of candidate robots that guides the evolution of morphology and control towards a valid design when transferred to reality. We aim to minimize the explicit knowledge introduced by the designer in the fitness function. As a consequence of this higher flexibility, we must include elements to ensure that the obtained robots are feasible. To do it, we propose an extension of the principles proposed by classical authors from traditional evolutionary robotics to brain-body evolution. In this paper we describe this methodology and show its application in a benchmark example of evolutionary robot design. To this end, previously presented elements like the structural definition of the robotic units, the encoding of the morphology and control and the specific evolutionary algorithm applied are also briefly described.