Publication

Abstract : Steering angle prediction plays a vital role in the control of Autonomous Vehicles (AVs) and has attracted significant interest from technology firms, transportation authorities, and researchers in the automotive industry. Various Deep Learning (DL) architectures have been employed to predict the steering angle under various driving conditions. An accurate prediction of steering angles is important in ensuring the vehicle is maintained in the corresponding lane. The major challenge in predicting the steering angle is various lane marking styles, heterogeneous road types, texture, color, lighting conditions, and so on. This issue can be solved by developing an effective DL model and maintaining the vehicle in the designated lane. In this paper, an African Fire Hawk Optimization-based hybrid Deep Learning (AFHO-hybrid DL) model is proposed, where the steering angle is predicted by a novel hybrid DL model, named as Multi-Input Control with Hierarchy Network (MICH-Net). For predicting the steering angle, this research considers the Driving footage under different road, weather, and illumination conditions as an input. The proposed MICH-Net is designed by combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) through the MICH layer. Moreover, the layer size is determined by the AFHO, which combines the African Vultures Optimization Algorithm (AVOA) and the Fire Hawk Optimizer (FHO). Additionally, AFHO-hybrid DL outperformed the existing techniques with a maximal accuracy of 93.1%, a maximal Positive Predictive Value (PPV) of 92.5%, a minimal Root Mean Square Error (RMSE) of 0.324, and a maximal f1-score of 92.2%, respectively.