The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9809

Title:Mens sana in corpore sano: Co-evolving robot controllers and body plans
Date: 1998
Abstract:Evolutionary robotics seeks to automate the design of robots by utilising the processes of natural selection. The majority of the reported work in the field concentrates on evolving the control mechanism alone. In natural systems, however, the control system and morphology co-evolve. in [Lee et al. 96b] the autors develop a hybrid GP/GA approach for co-evolving the controllers and morphologies of robots for fitness-specified tasks in simulation. [Lee 97] extends this work to evolve hierarchical controllers for more complex tasks and tests co-evolved controllers and morphologies on real robots. In this project the approach is extended to include sequential as well as reactive controllers and a simulator constructed in such a way that arbitrary classes of mobile robots may be easily accommodated. A series of experiments intended to culminate in a football playing capability is begun; due to implementation problems however the experiments are incomplete.

[Search These Pages] [DAI Home Page] [Comment]