Simulating Human Driving Expertise in Autonomous Campus Shuttle for Enhanced Navigation Using Behavioral Cloning

Despite remarkable advancements in Autonomous Campus Shuttle (ACS) development, achieving human-like driving expertise remains a significant challenge. This study presents Behavioral Cloning (BC) as a strategic approach within an autonomous campus shuttle system to address this challenge. The methodology involves extensively collecting diverse driving dataset from human operators in campus environments using the Udacity simulation platform. Utilizing the dataset, a Convolutional Neural Network (CNN) model was trained through BC to mimic human driving patterns, and the model was seamlessly integrated into the shuttle’s navigation system. Rigorous iterative epochs and meticulous hyperparameter tuning of model parameters demonstrate the efficacy of the BC-based approach in significantly improving the shuttle’s adaptability and responsiveness in campus settings with training and validation losses approaching 0.07. The BC model proves instrumental in facilitating safe and efficient navigation, and accurately handling unforeseen circumstances and dynamic traffic situations. This study underscores the potential of BC as a potent tool for endowing ACSs with human-like driving expertise, thereby enhancing safety, comfort, and overall acceptance among passengers in diverse and challenging environments.