The current work focused on enhancing models for self-driving cars. Initially, the work involved training models and adjusting parameters for optimal performance on specific tracks, aiming to generalize this performance across various tracks. Implementing image augmentation and processing was essential for real-time generalization due to the superior performance of models on a particular track. This approach incorporated the utilization of Convolutional Neural Networks (CNN) to capture spatial structures and Recurrent Neural Networks (RNN) to address temporal aspects within the image data-set. This combination was chosen for its ability to build efficient and faster neural networks while reducing computational requirements. A proposal to substitute recurrent layers for pooling layers was suggested to potentially reduce information loss, offering a promising direction for future exploration. The work highlights the importance of adapting models across different tracks and environments, the potential benefits of specific neural network architectures (CNN for spatial and RNN for temporal features), and the significance of combining real and simulated data for training and generalization. The work's focus on self-driving cars aligns with the ongoing developments and advancements in autonomous vehicle technology.