Steering angle prediction via neural networks

Methods to calculate the correct steering angle are an important aspect of developing self-driving vehicles. Recently, prediction methods based on deep neural networks have achieved accurate results that outperform other over a range road types. This paper investigates for autonomous vehicle as regression problem and solves it using networks. The proposed approach obtains data in form coordinates lane lines examines if they belong one or more lines. Based that, method locates path line is moving on. It then determines equation represents path. Finally, coefficients this fed into trained network, which predicts video frame. Extensive tests were performed Comma.ai Udacity benchmarks evaluate performance approach. State-of-the-art achieved, with mean absolute error 0.64 root square 0.87 dataset, 1.04 2.33 dataset.