William Spears | University of Wyoming (original) (raw)

Books by William Spears

Standard approaches to understanding swarms rely on inspiration from biology and are generally co... more Standard approaches to understanding swarms rely on inspiration from biology and are generally covered by the term “biomimetics”. This book focuses on a different, complementary inspiration, namely physics. The editors have introduced the term "physicomimetics" to refer to physics-based swarm approaches, which offer two advantages. First, they capture the notion that “nature is lazy", meaning that physics-based systems always perform the minimal amount of work necessary, which is an especially important advantage in swarm robotics. Second, physics is the most predictive science, and can reduce complex systems to simple concepts and equations that codify emergent behavior and help us to design and understand swarms.

The editors consolidated over a decade of work on swarm intelligence and swarm robotics, organizing the book into 19 chapters as follows. Part I introduces the concept of swarms and offers the reader a physics tutorial; Part II deals with applications of physicomimetics, in order of increased complexity; Part III examines the hardware requirements of the presented algorithms and demonstrates real robot implementations; Part IV demonstrates how the theory can be used to design swarms from first principles and provides a novel algorithm that handles changing environments; finally, Part V shows that physicomimetics can be used for function optimization, moving the reader from issues of swarm robotics to swarm intelligence. The text is supported with a downloadable package containing simulation code and videos of working robots.

This book is suitable for talented high school and undergraduate students, as well as researchers and graduate students in the areas of artificial intelligence and robotics.

About this book: Despite decades of work in evolutionary algorithms, there remains a lot of uncer... more About this book:
Despite decades of work in evolutionary algorithms, there remains a lot of uncertainty as to when it is beneficial or detrimental to use recombination or mutation. This book provides a characterization of the roles that recombination and mutation play in evolutionary algorithms. It integrates prior theoretical work and introduces new theoretical techniques for studying evolutionary algorithms. An aggregation algorithm for Markov chains is introduced which is useful for studying not only evolutionary algorithms specifically, but also complex systems in general. Practical consequences of the theory are explored and a novel method for comparing search and optimization algorithms is introduced. A focus on discrete rather than real-valued representations allows the book to bridge multiple communities, including evolutionary biologists and population geneticists.

Written for:
Computer scientists and mathematicians specializing in evolutionary algorithms; evolutionary biologists and population geneticists; and practitioners interested in evaluating and comparing different search and optimization algorithms

Papers by William Spears

One of the major goals of most early concept learners was to find hypotheses that were perfectly ... more One of the major goals of most early concept learners was to find hypotheses that were perfectly consistent with the training data. It was believed that this goal would indirectly achieve a high degree of predictive accuracy on a set of test data. Later research has partially disproved this belief. However, the issue of consistency has not yet been resolved completely. We examine the issue of consistency from a new perspective. To avoid overfitting the training data, a considerable number of current systems have sacrificed the goal of learning hypotheses that are perfectly consistent with the training instances by setting a goal of hypothesis simplicity (Occam's razor). Instead of using simplicity as a goal, we have developed a novel approach that addresses consistency directly. In other words, our concept learner has the explicit goal of selecting the most appropriate degree of consistency with the training data. We begin this paper by exploring concept learning with less than...

This paper summarizes recent research on competition-based learning procedures performed by the N... more This paper summarizes recent research on competition-based learning procedures performed by the Navy Center for Applied Research in Artificial Intelligence at the Naval Research Laboratory. We have focused on a particularly interesting class of competition-based techniques called genetic algorithms. Genetic algorithms are adaptive search algorithms based on principles derived from the mechanisms of biological evolution. Recent results on the analysis of the implicit parallelism of alternative selection algorithms are summarized, along with an analysis of alternative crossover operators. Applications of these results in practical learning systems for sequential decision problems and for concept classification are also presented. INTRODUCTION One approach to the design of more flexible computer systems is to extract heuristics from existing adaptive systems. We have focused on a class of learning systems that use competition-based procedures, called genetic algorithms (GAs). GAs are b...

Abstract. This paper introduces a novel framework for designing multiagent systems, called “Distr... more Abstract. This paper introduces a novel framework for designing multiagent systems, called “Distributed Agent Evolution with Dynamic Adaptation

In nature, a species is defined as a collection of phenotypically similar individuals. Many biolo... more In nature, a species is defined as a collection of phenotypically similar individuals. Many biologists believe that individuals in a sexually reproductive species can be created and maintained by allowing restrictive mating only among individuals from the same species. The connection between the formation of multiple species in nature and in search and optimization problems lies in solving multimodal problems, where the objective is not only to find one optimal solution, but to find a number of optimal solutions. In those problems, each optimal solution may be assumed to constitute a species. Since evolutionary algorithms work with a population of solutions, the concept of natural speciation techniques can be implemented to allow formation of multiple subpopulations, each focusing its search for one optimal solution. This way, multiple optimal solutions can be discovered simultaneously. In this section, a number of speciation techniques are discussed.

International Conference on Genetic Algorithms, 1991

Traditionally, genetic algorithms have relied upon 1 and 2-point crossover operators. Many recen... more Traditionally, genetic algorithms have relied upon 1
and 2-point crossover operators. Many recent empir-
ical studies, however, have shown the benefits of
higher numbers of crossover points. Some of the
most intriguing recent work has focused on uniform
crossover, which involves on the average L/2 cross-
over points for strings of length L. Theoretical
results suggest that, from the view of hyperplane
sampling disruption, uniform crossover has few
redeeming features. However, a growing body of
experimental evidence suggests otherwise. In this
paper, we attempt to reconcile these opposing views
of uniform crossover and present a framework for
understanding its virtues.

The ability of robots to quickly and accurately localize their neighbors is extremely important f... more The ability of robots to quickly and accurately localize their neighbors is extremely important for robotic teams. Prior approaches typically rely either on global information provided by GPS, beacons and landmarks, or on complex local information provided by vision systems. In this paper we describe our trilateration approach to multi-robot localization, which is fully distributed, inexpensive, and scalable [15]. Our prior research [14] focused on maintaining multi-robot formations indoors using trilateration. This paper pushes the limits of our trilateration technology by testing formations of robots in an outdoor setting at larger inter-robot distances and higher speeds.

Traditionally, genetic algorithms have relied upon 1 and 2-point crossover operators. Many recent... more Traditionally, genetic algorithms have relied upon 1 and 2-point crossover operators. Many recent empirical studies, however, have shown the benefits of higher numbers of crossover points. Some of the most intriguing recent work has focused on uniform crossover, which involves on the average L/2 crossover points for strings of length L. Despite theoretical analysis, however, it appears difficult to predict when a particular crossover form will be optimal for a given problem. This paper describes an adaptive genetic algorithm that decides, as it runs, which form is optimal.

Foundations of Genetic Algorithms 6, 2001

2009 IEEE Conference on Technologies for Homeland Security, 2009

The research work presented herein is part of a large effort in R&D of a physics-based approach t... more The research work presented herein is part of a large effort in R&D of a physics-based approach to develop networks of mobile sensing agents for monitoring, tracking, reporting and responding to hazardous conditions such as those resulting from the release of a WMD. We present the development of an efficient and robust distributed collaborative search algorithm for a team of unmanned robots that must locate the emitter of a toxic plume in an urban setting. We aim to provide a scientific, yet practical, approach to the design and analysis of rapidly deployable, scalable, adaptive, cost-effective groups of autonomous robots to replace or complement humans in the hazardous task of localizing the source of toxic chemical, biological or radiological plumes. In our approach, the robot group coordination and control are based on a physics-based framework called physicomimetics, and the collaborative search algorithm, called fluxotaxis, is based on fluid mechanics.

The Springer International Series in Engineering and Computer Science, 1993

Lecture Notes in Computer Science, 2009

This paper is concerned with assuring the safety of a swarm of agents (simulated robots). Such be... more This paper is concerned with assuring the safety of a swarm of agents (simulated robots). Such behavioral assurance is provided with the physics method called kinetic theory. Kinetic theory formulas are used to predict the macroscopic behavior of a simulated swarm of individually controlled agents. Kinetic theory is also the method for controlling the agents. In particular, the agents behave like particles in a moving gas. The coverage task addressed here involves a dynamic search through a bounded region, while avoiding multiple large ...