Deep-Learning-for-Autonomous-Driving

The purpose of this project is the development of an End-to-End learning model in order to predict the steering angle of an autonomous car. The proposed method uses monocular vision in order to acomplish the prediction task. Specifically, a CNN followed by LSTM units, is trained in order to manage both spatial and temporal information of the image sequence. In addition, a fusion with a second CNN that uses past prediction as inputs, is proposed, in order to improve the temporal information available. Both of the architectures were trained and tested on human driving data, provided by Udacity Challenge 2.