Implementation of a navigational neural network on a parallel DSP board

This work presents a neural network architecture that is motivated by the learning and memory characteristics of a part of the brain known as hippocampus, which is important in navigational behavior in humans and animals. Neural networks perform nonlinear transformations on data to yield suitable classification or control actions. In our case, the navigation network takes the distance information as data and maps it to control actions by the mobile robot. Navigation is a very important engineering problem for unknown or hazardous environments to ensure the safety of equipment and human life. Hardware implementation can benefit applications in real time where speed is the major concern. Our objective is to implement such a navigational neural network in parallel so that real time performance can be achieved by using a parallel DSP board system. Supplementary studies are also being carried out on the IBM SP2 supercomputer to understand the design and scaling properties of the parallel algorithm.