**1. Introduction**

Given the recent development of self-driving cars by companies such as Tesla, Google and others, it was of interest to attempt to replicate this on a smaller scale, by implementing a similar method on a small electronic toy car. An issue many developers in this industry are having is the issue of leaving the device connected to a wide area network (WAN) all the time, leaving the vehicles vulnerable to not only hacking but also vulnerable to being unable to make decisions in out-of-reach places such as the countryside where a WAN may not be available or only available intermittently. The decision was made to attempt the replication on a small scale, using a closed network with the computations occurring within the vehicle, as opposed to externally. The study was viewed as a proof of concept to test the possibility and feasibility, which could help lead research on full-scale self-driving cars and potentially also enhance the toy industry.

This study mainly focuses on a proof-of-concept implementation of artificial neural networks (ANNs) to classify road sign images in a real-time scenario

integrated with relevant image processing techniques. The study includes how the processed data outputted by ANN models is translated into defined movements of the car and whether it is suitably efficient to follow a track.

Neural networks are often used for image processing. Once the network is trained and the weighting values/structure is stored, all that would be required on future runs of the program would be to load these stored values/designs for the trained network for it to have the ability to run correctly. This means that once the network is trained, real-time image classification and thus prediction of the movement become very efficient. This is a beneficial method for a project of this sort, as the car is driving in real time and thus computations must be made in real time.

The overall objective was to allow the vehicle to drive autonomously around an unknown track with little to no error, utilising collision avoidance.
