This article proposes a deep learning framework to enhance the accuracy of steering angle predictions for autonomous vehicles (AV). By improving directional control, the model aims to reduce traffic-related incidents, reduce driver workload, and support economic efficiency. Accurate prediction of angle influenced by real-time factors such as surface conditions, weather disturbances, and visual clarity. Safety compliance is embedded within the system to prevent erratic vehicular behavior. This study uses the publicly available driving dataset, consisting of 84, 000 images serves as foundation for training. The solution employs deep learning Convolution Neural Network (3D-Conv) in this research. A transformer-based optimization strategy reduces prediction error by minimizing mean squared deviation between ground truth and estimated angles. Benchmarking results demonstrate that the architecture consistently outperforms conventional models like Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks in both precision and computational efficiency.