| This thesis is an investigation of whether simple control systems, using few explicit instructions but incorporating biologically-inspired subsystems of motivation and learning, are capable of enabling a robot to navigate a maze.; This "brain" consists of a discrete autocorrelator for recognition of spatial configurations in the environment, a counterpropagation network for learning successful stimulus-response pairs, and a generally distributed architecture of simpler, interconnected modules. It also includes basic principles of drives and reinforcement from psychology to cue and to evaluate action.; The model is designed to enable an autonomous robot bearing sonar, light, and touch sensors to seek out a foodlike goal. It identifies sets of spatial configurations from noisy sonar signals, and learns the best-fitting response to each set. This study has shown the capacity of imprecise mechanisms to guide navigation in a maze, and the emergence of complex behavioural strategy from a design of local, distributed components. |
