Print Email Facebook Twitter Automated Robot Design With Artificial Evolution Title Automated Robot Design With Artificial Evolution Author Kuppens, P.R. Contributor Wolfslag, W.J. (mentor) Wisse, M. (mentor) Herder, J.L. (mentor) Faculty Mechanical, Maritime and Materials Engineering Department BioMechanical Design Programme BioRobotics Date 2016-02-03 Abstract Robots usually do a single job and would perform better when their mechanical structure is particularly designed for their designated task. However, manually designing robots with mechanical properties that implicitly compute and execute a specific task is very hard and time consuming. To assist designers, a platform that automatically designs dynamical mechanisms is needed. Evolutionary algorithms (EAs) are suggested as invention machines for similar domains and seem promising for automatic mechanism design. However key obstacles are the representation of dynamical mechanisms, operators for sexual reproduction and maintaining population diversity. We present a robot genome based on graph theory that is as compact as possible, has no genetic multiplicity and is completely closed. We define a sexual operator that is able to merge both topology and parameter data from two parent-robots to form a child-robot that is essentially a mixture of both parents. And we investigate which of the most commonly used methods for diversity maintenance improve optimization performance for dynamical mechanism design best. We demonstrate that our new representation and sexual operator enables automatic design of dynamical mechanisms. Specifically, we showcase automatic design of two-dimensional mechanisms (with a single degree of freedom) that track a straight line (Roberts mechanism) as well as an ellipse by virtue of their kinematic and dynamical properties. The best optimization results are obtained by maintaining diversity with a combination of the island and the diffusion model. We demonstrate the practical value of the evolved mechanisms by extracting design principles and by redesigning a mechanism to include mathematically hard-to-formulate objectives such as aesthetics and symmetry. Subject dynamicrobottask specificmechanism designgeneticevolutionaryalgorithm To reference this document use: http://resolver.tudelft.nl/uuid:6e4ce276-00c4-47ef-8ec3-2b03120bfd46 Embargo date 2018-02-03 Part of collection Student theses Document type master thesis Rights (c) 2016 Kuppens, P.R. Files PDF prkuppens_msc_thesis.pdf 5.24 MB Close viewer /islandora/object/uuid:6e4ce276-00c4-47ef-8ec3-2b03120bfd46/datastream/OBJ/view