Comparative Transfer Learning Models for End-to-End Self-Driving Car

Self-driving automobiles are prominent in science and technology, which affect social economic development. Deep learning (DL) is the most common area of study artificial intelligence (AI). In recent years, deep learning-based solutions have been presented field self-driving cars achieved outstanding results. Different studies investigated a variety significant technologies for autonomous vehicles, including car navigation systems, path planning, environmental perception, as well control. End-to-end control directly converts sensory data into commands driving. This research aims to identify accurate pre-trained Neural Network (DNN) predicting steering angle vehicle that suitable be applied embedded automotive with limited performance. Three well-known models were compared this study: AlexNet, ResNet18, DenseNet121. Transfer was utilized by modifying final layer order predict vehicle. Experiments conducted on dataset collected from two different tracks. According study's findings, ResNet18 DenseNet121 lowest error percentage values. Furthermore, performance modified evaluated predetermined outperformed terms accuracy, less deviation center track, while demonstrated greater adaptability across multiple tracks, resulting better consistency.