I. Introduction
Autonomous driving aims to reduce the number of road accidents occurring due to driver fatigue. Typically multiple sensors are mounted on the vehicle for autonomous driving and ADAS (Advanced Driver Assistance Systems) applications. For environment perception tasks, the camera has been used extensively owing to its low cost and capability to provide a detailed information of the road environment. This has resulted in significant advances in vision-based research problems. An important vision-based research problem is the task of predicting the steering angle from camera. This particular task has received significant attention from the research community in recent years [1] due to the advances in deep learning. However, solving such a problem is not trivial, owing to the difficulties in robustly estimating the steering angle in varying scenarios, environments and illuminations. Another challenge is the non-linearity associated with the vehicle dynamics for sudden changes in trajectory. Finally, the most important challenge is the need for real-time or near real-time computation.