Insect inspired visual navigation the advantages and risks of using temporal information

<p dir="ltr">This thesis presents an extensive investigation on the use of temporal information for visual navigation algorithms in the context of route following over different environments and noise conditions. The navigational algorithms developed within this thesis implement techniques inspired by ants and bees but with the aim to deploy them in real autonomous robots. The Introduction and Background provide information on the various types of navigation algorithm used for robotics today. I also suggest a breakdown of the components that make up navigation algorithms as this will be used later to contextualise the work carried out within this thesis. Further topics include bio-inspired algorithms inspired by behavioural navigation studies on ants and bees and their visual capabilities. A family of insect-inspired algorithms is introduced that uses panoramic images collected during a training route to subsequently derive directional information from regions around the training route. The ant-like mechanism of pairing visual information with the process of extracting a heading direction (the visual compass) is introduced which is crucial for an agent’s navigational operations. Bio-inspired image processing and heading extraction mechanisms form the basis of most of the navigation algorithms in this thesis. Nevertheless many insect-inspired navigation algorithms presented in the Introduction and Background do not make any use of temporal information despite there being evidence that ants do utilise it. Chapter 3 introduces the use of temporal information in the context of the antinspired navigation algorithms. The analysis and results show that indeed incorporating temporal information improves the performance of the visual navigation algorithms. It explores the effects of the temporal mechanism in detail and lays the ground work for the next chapter. Chapter 4 expands upon the ideas form chapter 3 and outlines a general framework for the use of temporal information in navigation algorithms. The ideas of a dynamic update of temporal sequences is introduced and tested thoroughly in multiple simulated and real world environments. Further analysis of this family of algorithms and discussion outline the benefits and shortcomings of using spatiotemporal information in the context of navigation. iv Chapter 5 explores the effects of visual processing on the relevant algorithms. This is achieved by exploring the relationship between visual processing and a variety of functions used to compare visual signals to each other. Chapter 6 breaks down the mechanism that uses visual signals to extract heading directions. It introduces certain methods that can be used to measure the quality of that signal and proposes way those metrics can be used to further improve the navigation capabilities of an agent. It further indicates that the quality metric of the signal can vary between different environments and thus be used to categorise their difficulty. In summary this thesis attempts to explore and investigate all aspect of an agent that is designed to navigate autonomously inspired by insects. Each chapter explores a part of the working of such agent, from the visual signals to the processing of those signal, the memory storage and then the use of all that information to navigate any environment. In term of the robotics literature this work bring together lessons from biology and methods from engineering in order to outline an autonomous navigating agent aimed to robustly operate in the real world.</p>