Biologically inspired flight : the autonomous control of a quadcopter using spiking neural networks

Date
2015-12
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: With the recent increase in the availability and popularity of micro aerial vehicles new methods for their autonomous control should be investigated. In this project the autonomous control of a flying drone using biologically inspired models is investigated. The models that are investigated for this project are those of the various neurological structures which form the insect's visual system. The models for the processes which take place in an insect’s lamina, medulla and lobula and are responsible for contrast enhancement, motion detection and collision detection are used to build a system which allows the drone to navigate an enclosed testing environment without collision. These models are implemented and tested in iqr, a spiking neural network simulator able to interface with external hardware. Additional models for colour extraction and tracking using spiking neural networks are also developed and tested using iqr. The models are combined to form two systems. The first of these systems uses the models from the insect’s visual system to create a system that allows the drone to navigate a constrained testing environment without collision. The second system uses the models developed for colour tracking and following to allow the drone to follow a colour object that appears in its visual environment. The results show that the system developed for autonomous navigation works successfully every time and that the system developed for tracking a coloured object worked as long as the coloured object is present in the system's field of view. The results show promise for the use of biologically inspired models for autonomous control of vehicles, especially now with the advances being made in parallel computing and neuromorphic computing. Suggestions for topics of investigation that would improve the systems created during this project as well as in this field in general are also given.
AFRIKAANSE OPSOMMING: Die huidige toename in die gewildheid en beskikbaarheid van mikrolugvaartuie noop die ondersoek na nuwe metodes van outonome beheer. Hierdie projek ondersoek die outonome beheer van ’n onbemande vliegtuig wat biologies geinspireerde modelle gebruik. Die modelle wat ondersoek word is die verskeie neurologiese strukture wat deel uitmaak van ’n insek se visuele of sig-sisteem. Modelle om die prosesse wat in die lamina, medulla en lobula van 'n insek plaasvind en verantwoordelik is vir kontrasverhoging, beweging en botsing word gebruik om ’n sisteem te bou wat die vliegtuig toelaat om ’n geslote eksperimentele area te navigeer sonder dat botsings plaasvind. Die modelle is geimplementeer en getoets in iqr, ’n "spiking neural network"nabootser met koppelvlakke aan eksterne hardeware. Addisionele modelle vir kleurekstraksie en volging met die gebruik van "spiking neural networksïs ook ontwikkel en met behulp van iqr getoets. Die modelle is gekombineer om twee sisteme te vorm. Die eerste sisteem gebruik modelle van die insek se visuele sisteem om die vliegtuig toe te laat om in 'n beperkte eksperimentele omgewing te beweeg sonder om botsings toe te laat. Die tweede sisteem maak gebruik van die modelle ontwikkel vir kleuridentifikasie en volging om die vliegtuig toe te laat om ’n voorwerp te volg wat in sy sigveld verskyn. Die resultate toon aan dat die sisteem ontwikkel vir outonome navigering keer op keer suksesvol is. Die opsporing van 'n gekleurde voorwerp is suksesvol solank die gekleurde voorwerp in die sisteem se sigveld is. Die resultate toon die potensiaal vir die gebruik van biologies geinspireerde modelle vir outonome beheer van voertuie, veral nou met parallelle vooruitgang in rekenaar en neuromorfiese rekenaar ontwikkeling. Voorstelle vir onderwerpe vir verdere ondersoeke en wat ook die sisteem sal verbeter word gemaak.
Description
Thesis (MEng)--Stellenbosch University, 2015.
Keywords
Quadcopter -- Autonomous control, Spiking neural network, Insect visual system, UCTD
Citation