Deep Learning for Hardware-Constrained Driverless Cars

Abstract

The field of self-driving cars is a fast-growing one, and numerous companies and organizations are working at the forefront of this technology. One of the major requirements for self-driving cars is the necessity of expensive hardware to run complex models. This project aims to identify a suitable deep learning model under hardware constraints. We obtain the results of a supervised model trained with data from a human driver and compare it to a reinforcement learning-based approach. Both models will be trained and tested on devices with low-end hardware, and their results visualized with the help of a driving simulator. The objective is to demonstrate that even a simple model with enough data augmentation can perform specific tasks and does not require much investment in time and money. We also aim to introduce portability to deep learning models by trying to deploy the model in a mobile device and show that it can work as a standalone module.