A computational model for learning to navigate in an unknown environment

The ultimate goal of this research is to design a system to automatically learn a visual domain so that it can subsequently execute a controlled navigation from any location to another when instructed to do so. The required system must have: flexibility, autonomy, scalability and robustness, which is defined for clarity. It must be flexible in order to cope with a broad range of problems, for example indoor and outdoor path planning and a large class of visual features i.e. the ability to function in as broad a range of circumstances as possible. We have based our algorithms on producing an architecture which can learn an unknown environment, using a self-generating map. A self-generating map is an extensible neural network method which uses self organising features. Our method is different in that is uses a novel method for feature extraction and a novel neural network architecture-the contextual layered associative memory