Utilizing Transfer Learning in the Udacity Simulator to Train a Self-Driving Car for Steering Angle Prediction

Recently, the study of self-driving autonomous vehicles has gained great popularity. The development of automated driving will change humans’ lives which can provide a fully safe transportation system for them. Such a system requires huge computing power with a high cost, So the simulator is a good choice to train and test self-driving cars as an initial step before implementing them in real. This paper provides a study on how a self-driving car is trained for steering angle prediction and following the road lane in the Udacity simulator via the input data from the car's front-center camera. It also explores the utilization of pre-trained Keras transfer learning models in self-driving cars. The methodology involves collecting the necessary driving data and preprocessing it. Then the transfer learning models are used as base models. The models are trained on a part of the lake track road of the Udacity simulator and tested on the rest part of the road. Also, the trained models have been evaluated by mean square error, the coefficient of determination R square, and Root mean square error. The car on the simulator acts autonomously very well and follows the road’s lane using the center camera which shows promising performance. The findings contribute to the advancement of autonomous driving technology and provide insights for further research in this field.