A Connectionism Approach to the Emulation of Multi-Agent Mobile Robots

The work presented in this report deals with the modelling and subsequent neural network emulation of autonomous mobile robots, moving in pre-defined environments in response to given control signals. This work was undertaken with the intention of training neural network based controllers for the vehicles, in order to control them whilst they performed required navigational tasks, whilst avoiding collisions with each other and with environmental obstacles. The required modelling was carried out using a combination of trigonometry and geometry, the corresponding neural network emulators were trained using an algorithm based on back propagation and the resulting emulator networks performed well enough for work to now begin on controller training.