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

MSc Thesis #9556

Title:Conceptual Design of Robots Using Evolutionary Algorithms
Date: 1995
Abstract:The development of robots which perform tasks reliably and easily has proven to be a greater stumbling block than was originally anticipated. Traditional hardware and software design methods wither fail to address or over-compensate for problems which occur in the mechanical and behavioural design of robots, such as uncertainty, task representation and planning. Recent work has suggested that by directly connecting sensors to actuators, we can evoke the behaviours we require, rather than meticulously planning them. This work often uses evolutionary algorithms to optimise the connections, the number and type of sensors/actuators. Evolutionary algorithms have also used in the latter stages of mechanical design, for part selection, and parameter optimisation. In this project it is suggested that by using genetic algorithms to evolve the morphology, or initial conceptual design of the robot, we may obtain more powerful results both in the mechanical and behavioural design domains. In addition the design and partial development of a software tool suitable for investigating this conjecture is described.

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