Initially, data was ingested from simulation tool ‘Udacity‘, generated a driving log with over 17,000 images from central, left, and right cameras, along with parameters such as steering angles, throttle, reverse, and speed. The dataset underwent meticulous preprocessing and normalization, including path extraction and histogram equalization to address data imbalance. Utilizing TensorFlow and Keras, a convolutional neural network (CNN) with five convolutional layers and three fully connected layers was architected to extract spatial features and predict steering angles. To enhance the model’s generalization, data augmentation techniques such as 20% brightness adjustments, 15-pixel translations, and horizontal flips were implemented to simulate diverse driving conditions. The model was trained over 25 epochs with a batch size of 32, employing advanced methodologies like batch normalization and a dropout rate of 0.5 to prevent overfitting and ensure stability. Performance was evaluated using mean squared error (MSE), achieving a low loss of 0.02. After testing, the model was able to learn well and adapt on its own, even in unknown driving conditions achieving a well generalized model. This project showcases technical prowess in integrating deep learning techniques for real-time steering command inference, significantly contributing to the field of autonomous vehicle technology and demonstrating the capability to deploy a robust CNN model for self-driving car navigation.