Robots have long been used for completing tasks that are too difficult, dangerous, or distant to be accomplished by humans. In many cases, these robots are highly specialized platforms - often expensive and capable of completing every task related to a mission's objective. An alternative approach is to use multiple platforms, each less capable in terms of number of tasks and thus significantly less complex and less costly. With advancements in embedded computing and wireless communications, multiple such platforms have been shown to work together to accomplish mission objectives. In the extreme, collections of very simple robots have demonstrated emergent behavior akin to that seen in nature (e.g., bee colonies) motivating the moniker of ''swarm robotics'' - a group of robots working collaboratively to accomplish a task. The use of robotic swarms offers the potential to solve complex tasks more efficiently than a single robot by introducing robustness and flexibility to the system. This work investigates localization in heterogeneous and autonomous robotic swarms to improve their ability to carry out exploratory missions in unknown terrain. Collaboratively, these robots can, for example, conduct sensing and mapping of an environment while simultaneously evolving a communication network. For this application, among many others, it is required to determine an accurate knowledge of the robot's pose (i.e., position and orientation). The act of determining the pose of the robot is known as localization. Some low cost robots can provide location estimates using inertial measurements (i.e., odometry), however this method alone is insufficient due to cumulative errors in sensing. Image tracking and wireless localization methods are implemented in this work to increase the accuracy of localization estimates. These localization methods complement each other: image tracking yields higher accuracy than wireless, however a line-of-sight (LOS) with the target is required; wireless localization can operate under LOS or non-LOS conditions, however has issues in multipath conditions. Together, these methods can be used to improve localization results under all sight conditions. The specific contributions of this work are: (1) a concept of 'shared sensing' in which extremely simple and inexpensive robots with unreliable localization estimates are used in a heterogeneous swarm of robots in a way that increases the accuracy of localization for the simple agents and simultaneously extends the sensing capabilities of the more complex robots, (2) a description, evaluation, and discussion of various means to estimate a robot's pose, (3) a method for increasing reliability of RSSI measurements for wireless ranging/localization systems by averaging RSSI measurements over both time and space, (4) a process for developing an in-field model to be used for estimating the location of a robot by leveraging the existing wireless communication system.

In this thesis, we investigate the problem of path formation and prey retrieval in a swarm of robots. We present two swarm intelligence control mechanisms used for distributed robot path formation. In the first, the robots form linear chains. We study three variants of robot chains, which vary in the degree of motion allowed to the chain structure. The second mechanism is called vectorfield. In this case, the robots form a pattern that globally indicates the direction towards a goal or home location. Both algorithms were designed following the swarm robotics control principles: simplicity of control, locality of sensing and communication, homogeneity and distributedness. We test each controller on a task that consists in forming a path between two objects—the prey and the nest—and to retrieve the prey to the nest. The difficulty of the task is given by four constraints. First, the prey requires concurrent, physical handling by multiple robots to be moved. Second, each robot’s perceptual range is small when compared to the distance between the nest and the prey; moreover, perception is unreliable. Third, no robot has any explicit knowledge about the environment beyond its perceptual range. Fourth, communication among robots is unreliable and limited to a small set of simple signals that are locally broadcast. In simulation experiments we test our controllers under a wide range of conditions, changing the distance between nest and prey, varying the number of robots used, and introducing different obstacle configurations in the environment. Furthermore, we tested the controllers for robustness by adding noise to the different sensors, and for fault tolerance by completely removing a sensor or actuator. We validate the chain controller in experiments with up to twelve physical robots. We believe that these experiments are among the most sophisticated examples of self-organisation in robotics to date.

The dykes from the Igaliko Nepheline Syenite complex belong to at least 3 individual swarms (i) a Mid-Gardar swarm in the Østfjordsdal valley, (ii) a Late-Gardar, Si-oversaturated swarm associated with the Younger Giant Dykes of Tugtutôq and (iii) a Si-undersaturated swarm intimately associated with the Late Gardar Igaliko Nepheline Syenite Central Complexes. In addition Early Gardar activity is recorded by the presence of some ultramafic lamprophyres which predate the Motzfeldt centre, sparse trachytes which are truncated by intrusions within the Motzfeldt centre and a possible BD(_0) dolerite which is also cut by the Motzfeldt centre. Most dykes however are bracketed between the Early and Late Igdlerfigssalik syenite intrusions. The main oversaturated and undersaturated suites can be separated on their Zr/Nb ratios (≈6.4 and 3.9 respectively). In addition, the undersaturated basic rocks have smooth chondrite normalised incompatible element spidergrams whereas the oversaturated basic rocks are characterised by negative Nb and positive P anomalies. Evolution of both suites can be modelled in terms of fractional crystallisation of feldspar, clinopyroxene, olivine, apatite and opaques from basaltic parents to either phonolitic or rhyolitic minimum compositions. In each instance these evolved composi tions are extremely rich in incompatible trace elements (REE, Nb, Zr, Rb). In some cases a high CO(_2) content in the undersaturated rocks may lead to the formation (by liquid immiscibility) of late stage carbonatite magmas. High CO(_2) also produces high ƒo(_2) in these magmas and it is argued that in some cases this can suppress the development of negative Eu anomalies on feldspar fractionation. The undersaturated swarm may have evolved from lamprophyric parental magmas, eg. camptonites, which are relatively abundant basic dykes. Ultramafic lamprophyres, often early, may have formed as extremely small degree partial melts at the onset of Gardar rifting. In the Late Gardar, magma genesis is related to the different extensional tectonic regimes which were operative at that time. Mineralogical evolution follows paths similar to several other Gardar suites and records a higher ƒo(_2) in the undersaturated rocks. Zr becomes concentrated in interstitial residual liquids in benmoreites and substitutes into amphibole as the newly proposed end-member zirconian-arfvedsonite.

Thesis (M.S.)--University of Wyoming, 2008. / Title from PDF title page (viewed on Apr. 1, 2010). Includes bibliographical references (p. 69-71).

Computer game-playing agents are almost as old as computers themselves, and people have been developing agents since the 1950's. Unfortunately the techniques for game-playing agents have remained basically the same for almost half a century -- an eternity in computer time. Recently developed approaches have shown that it is possible to develop game playing agents with the help of learning algorithms. This study is based on the concept of algorithms that learn how to play board games from zero initial knowledge about playing strategies. A coevolutionary approach, where a neural network is used to assess desirability of leaf nodes in a game tree, and evolutionary algorithms are used to train neural networks in competition, is overviewed. This thesis then presents an alternative approach in which particle swarm optimization (PSO) is used to train the neural networks. Different variations of the PSO are implemented and compared. The results of the PSO approaches are also compared with that of an evolutionary programming approach. The performance of the PSO algorithms is investigated for different values of the PSO control parameters. This study shows that the PSO approach can be applied successfully to train game-playing agents. / Dissertation (MSc)--University of Pretoria, 2007. / Computer Science / Unrestricted

Particle swarm optimization (PSO) is a well known population-based search algorithm, originally developed by Kennedy and Eberhart in 1995. The PSO has been utilized in a variety of application domains, providing a wealth of empirical evidence for its effectiveness as an optimizer. The PSO itself has undergone many alterations subsequent to its inception, some of which are fundamental to the PSO's core behavior, others have been more application specific. The fundamental alterations to the PSO have to a large extent been a result of theoretical analysis of the PSO's particle's long term trajectory. The most obvious example, is the need for velocity clamping in the original PSO. While there were empirical fndings that suggested that each particle's velocity was increasing at a rapid rate, it was only once a solid theoretical study was performed that the reason for the velocity explosion was understood. There has been a large amount of theoretical research done on the PSO, both for the deterministic model, and more recently for the stochastic model. This thesis presents an extension to the theoretical deterministic PSO model. Under the extended model, conditions for particle convergence to a point are derived. At present all theoretical PSO research is done under the stagnation assumption, in some form or another. The analysis done under the stagnation assumption is one where the personal best and neighborhood best are assumed to be non-changing. While analysis under the stagnation assumption is very informative, it could never provide a complete description of a PSO's behavior. Furthermore, the assumption implicitly removes the notion of a social network structure from the analysis. The model used in this thesis greatly weakens the stagnation assumption, by instead assuming that each particle's personal best and neighborhood best can occupy an arbitrarily large number of unique positions. Empirical results are presented to support the theoretical fndings. / Dissertation (MSc)--University of Pretoria, 2013. / gm2014 / Computer Science / Unrestricted

This research explores optimal sequencing of aircraft engine maintenance events. Due to the high ongoing maintenance costs and large capital investments required for supporting an aircraft engine fleet, the timing and associated costs of maintenance events are key to minimizing overall costs for an airline. This paper examines a novel application of particle swarm optimization techniques in order to create a decision tool which may be easily implemented by the practitioner. Numerical experiments demonstrate the quality of this solution method under multiple maintenance pricing structures and operational constraints.

This thesis compares the performance of heterogeneous and homogenous swarms. Swarms are defined as particles or agents which react to their environment and fellow particles or agents according to social rules. The weights of three attributes of an individual agent were varied for these experiments: Collision Avoidance with individual agents in the swarm, Center of Mass of the swarm and the parameter that controls Velocity Matching in the swarm. In homogenous swarms, all individuals had the same attribute weights while in heterogeneous swarms weights for one attribute were taken from a normal distribution for the population. These swarms were then given goals on a map to pursue. The maps were two-dimensional grid-surfaces with terrains of open, mountain and swamps. Performance was defined as the number of steps it took for 90% of the swarm to reach its final goal. The results show that heterogeneous swarms outperformed homogenous swarms if the weights for the Center of Mass Weight attribute were heterogeneous in the population. The Collision Avoidance and Matched Velocity attributes showed little performance difference for heterogeneous and homogenous swarms for the parameter weights tested. However, swarms heterogeneous in the Matched Velocity parameter showed substantial performance improvements for the most difficult map.

In the post-9/11 world, new and improved surveillance and information-gathering technologies have become a high-priority problem to be solved. Surveillance systems are often needed in areas too hostile or dangerous for a direct human presence. The field of robotics is being looked to for an autonomous mobile surveillance system. One major problem is the control and coordination of multiple cooperating robots. Researchers have looked to the distributed control strategies found in nature in the form of social insects as an inspiration for new control schemes. Swarm intelligence research centers around the interactions of such systems and how they can be applied to contemporary problems. In this thesis, a surveillance system of mobile autonomous robots based on the principles of swarm intelligence is presented. / Master of Science

Swarm robotics is a young multidisciplinary research field at the<p>intersection of disciplines such as distributed systems, robotics,<p>artificial intelligence, and complex systems. Considerable research<p>effort has been dedicated to the study of algorithms targeted to<p>specific problems. Nonetheless, implementation and comparison remain difficult due to the lack of shared tools and benchmarks. Among the tools necessary to enable experimentation, the most fundamental is a simulator that offers an adequate level of accuracy and flexibility to suit the diverse needs of the swarm robotics<p>community. The very nature of swarm robotics, in which systems may comprise large numbers of robots, forces the design to provide<p>runtimes that increase gracefully with increasing swarm sizes.<p><p>In this thesis, I argue that none of the existing simulators offers<p>satisfactory levels of accuracy, flexibility, and efficiency, due to<p>fundamental limitations of their design. To overcome these<p>limitations, I present ARGoS---a general, multi-robot simulator that<p>currently benchmarks as the fastest in the literature.<p><p>In the design of ARGoS, I faced a number of unsolved issues. First, in existing simulators, accuracy is an intrinsic feature of the<p>design. For single-robot applications this choice is reasonable, but<p>for the large number of robots typically involved in a swarm, it<p>results in an unacceptable trade-off between accuracy and<p>efficiency. Second, the prospect of swarm robotics spans diverse<p>potential applications, such as space exploration, ocean restoration,<p>deep-underground mining, and construction of large structures. These applications differ in terms of physics (e.g. motion dynamics) and available communication means. The existing general-purpose simulators are not suitable to simulate such diverse environments accurately and efficiently.<p><p>To design ARGoS I introduced novel concepts. First, in ARGoS accuracy is framed as a property of the experimental setup, and is tunable to the requirements of the experiment. To achieve this, I designed the architecture of ARGoS to offer unprecedented levels of modularity. The user can provide customized versions of individual modules, thus assigning computational resources to the relevant aspects. This feature enhances efficiency, since the user can lower the computational cost of unnecessary aspects of a simulation.<p><p>To further decrease runtimes, the architecture of ARGoS exploits the computational resources of modern multi-core systems. In contrast to existing designs with comparable features, ARGoS allows the user to define both the granularity and the scheduling strategy of the parallel tasks, attaining unmatched levels of scalability and efficiency in resource usage.<p><p>A further unique feature of ARGoS is the possibility to partition the<p>simulated space in regions managed by dedicated physics engines<p>running in parallel. This feature, besides enhancing parallelism,<p>enables experiments in which multiple regions with different features are simulated. For instance, ARGoS can perform accurate and efficient simulations of scenarios in which amphibian robots act both underwater and on sandy shores.<p><p>ARGoS is listed among the major results of the Swarmanoid<p>project. It is currently<p>the official simulator of 4 European projects<p>(ASCENS, H2SWARM, E-SWARM, Swarmix) and is used by 15<p>universities worldwide. While the core architecture of ARGoS is<p>complete, extensions are continually added by a community of<p>contributors. In particular, ARGoS was the first robot simulator to be<p>integrated with the ns3 network simulator, yielding a software<p>able to simulate both the physics and the network aspects of a<p>swarm. Further extensions under development include support for<p>large-scale modular robots, construction of 3D structures with<p>deformable material, and integration with advanced statistical<p>analysis tools such as MultiVeStA. / Doctorat en Sciences de l'ingénieur / info:eu-repo/semantics/nonPublished