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

MSc Thesis #9677

Title:Evolving Neurocontrollers and Body Plans of a Lego Robot
Date: 1996
Abstract:In this project we investigated the possibility of evolving neurocontrollers and body plans of a LEGO robot, so as to achieve a simple obstacle avoidance behaviour. A LEGO robot was built, and the response of its infrared sensors and motors in different conditions was measured. A mathematical model of sensor activation was built from such experimental data. A simulator was then implemented, through which the LEGO robot neurocontroller can be evolved by a Genetic Algorithm. The simulator makes use of sensor response measurements, in order to convert the robot's position into a set of sensor readings. The neurocontroller is then fed with such sensor readings, and consequently determines motor activation for the next time-step. The motor response measurements are used in order to calculate the robot's displacement, given the neurocontroller's output. The robots are tested for a given number of time-steps, and their performance is assessed through a suitable fitness function. The best individuals are then allowed to reproduce by parthenogenesis, so as to keep the robot population size constant. Children undergo random mutations, and the process is iterated for a given number of generations. Possible experiments making use of the simulator are suggested in the report.

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