Stephen Wayne Soliday (Research Activities)

Analysis of the Fisher Iris Data

Principle Thesis Research while at NC A&T SU was

Inexpensive Autonomous Vehicle Research

My thesis research was in the area of preemptive non-linear controls, specifically, the design and modeling of an hybrid fuzzy-neural Preemptive Navigation System for a two-wheel pivot-steering vehicle capable of target tracking and obstacle negotiation and collision avoidance.

The thesis title and abstract read:

A SUBSYSTEM APPROACH TO DEVELOPING A BEHAVIOR BASED HYBRID NAVIGATION SYSTEM FOR AUTONOMOUS VEHICLES

There does not exist one paradigm in machine intelligence that can function as a "black-box" for complex tasks. This is especially true for behavior based controllers. By implementing a subsystem based organization, each element of the behavior controller may be constructed using the machine intelligence paradigm best suited for that task. In this thesis a Behavior Based Intelligent Navigations System will be developed for use in mobile robotics. This thesis will also contain a review, history, and enhancement of each of the basic paradigms and training methods. Topics such as back-propagation neural networks, fuzzy associative maps, and genetic algorithms will be presented. Enhancements to the traing speed of back-propagation, and a new method for fuzzy clustering will be discussed. Also, presented in this work will be methods for constructing computational models from physical ones. All of the programming code in this work is based on a C++ Class Library that I developed as an independent project.

Side Research

Three Dimensional Object Recognition

In addition to my thesis research I am participating in research under a grant from The Advanced Research Projects Agency - ARPA. We are studying three dimensional object recognition. I have developed a new method of fuzzy clustering. I am currently comparing this method to traditional Neural-Network classifiers as well as Fuzzy Associative Maps using Identification Training.

Machine Intelligence Class Library

During the last several years I have been compiling my code, both personal and research oriented, into an Object Oriented Class Library, written in C++. This Library enables programmers to use many OOP tools for creating Neural Network and Fuzzy Logic applications. Also provided in the library are several prepackaged tools for configuring, training and implementing both MIMO Back-propagation Networks and MIMO Fuzzy Associative Maps[1]. The library also contains a virtual blocking system that allows for creation of very large arrays on systems with paged memory[2]. Additionally, provided is a device independent graphics object. The same graphics object is used regardless of the display driver. Currently the drivers available are (Postscript, HPGL, VGA, and XWindows).

The code and all simulations used in my thesis are built entirely on top of this library. The library in its current version is guaranteed to compile under GNU C/C++ on a Unix system[3]. I have also tested some of the library code in past years on VAX/VMS using GNU C/C++ and on MSDos machines using Borland's C/C++ compiler. When My Thesis is finished I will begin work on portability testing.

1. MIMO - Many Input Many Output

2. Both VMS and MSDOS support paged memory. VMS supports it as a luxury, using disk swapping to create virtually unlimited memory allocation. MSDOS supports it as an architectural oversight: MSDOS relies on page and offset addressing, thereby restricting individual allocations to under 64k

The bulk of my thesis work is being done on a 60mhz Pentium using Linux. When I work on campus or provide demonstrations I use a SparcServer 690MP running Solaris.