The design and intelligent control of an autonomous mobile robot
This thesis presents an investigation into the problems of exploration, map building and collision free navigation for intelligent autonomous mobile robots. The project began with an extensive review of currently available literature in the field of mobile robot research, which included intelligent control techniques and their application. It became clear that there was scope for further development with regard to map building and exploration in new and unstructured environments. Animals have an innate propensity to exhibit such abilities, and so the analogous use of artificial neural networks instead of actual neural systems was examined for use as a method of robot mapping. A simulated behaviour based mobile robot was used in conjunction with a growing cell structure neural network to map out new environments. When using the direct application of this algorithm, topological irregularities were observed to be the direct result of correlations within the input data stream. A modification to this basic system was shown to correct the problem, but further developments would be required to produce a generic solution. The mapping algorithms gained through this approach, although more similar to biological systems, are computationally inefficient in comparison to the methods which were subsequently developed. A novel mapping method was proposed based on the robot creating new location vectors, or nodes, when it exceeded a distance threshold from its mapped area. Network parameters were developed to monitor the state of growth of the network and aid the robot search process. In simulation, the combination of the novel mapping and search process were shown to be able to construct maps which could be subsequently used for collision free navigation. To develop greater insights into the control problem and to validate the simulation work the control structures were ported to a prototype mobile robot. The mobile robot was of circular construction, with a synchro-drive wheel configuration, and was equipped with eight ultrasonic distance sensors and an odometric positioning system. It was self-sufficient, incorporating all its power and computational resources. The experiments observed the effects of odometric drift and demonstrated methods of re-correction which were shown to be effective. Both the novel mapping method, and a new algorithm based on an exhaustive mesh search, were shown to be able to explore different environments and subsequently achieve collision free navigation. This was shown in all cases by monitoring the estimates in the positional error which remained within fixed bounds.