FYP25061

Autonomous driving has made significant progress, but reliable performance in complex urban environments remains challenging due to rare and unpredictable “long-tail” events. This project explores how computer vision techniques can be applied to improve autonomous driving performance in simulation. Using the CARLA Town05 environment, we collect expert driving data, train a vision-based control model based on imitation learning, and evaluate its ability to perform steering and brake control. Initial results show promising direction-following capability, while also revealing limitations such as lane drift and inaccurate turning. These findings motivate our next stage of work on robustness and object detection for safer autonomous driving.