Vision System: Lane detection was achieved through a minimalist, compute-efficient vision pipeline. A vertical Sobel operator combined with a tightly defined region of interest allowed the system to isolate and track lane lines without relying on fragile edge-detection methods like Canny or Hough transforms. Binary thresholding and morphological operations further stabilized output. This approach produced accurate lane centroids in real time, even under variable lighting and slight occlusion, without overwhelming the Jetson Nano’s limited resources. Object Detection: Traffic sign recognition was handled using a custom-trained YOLOv5 model, selected after early experiments with TensorFlow proved unstable at runtime. Over 7,000 annotated images were collected and labeled to build a task-specific dataset. The model’s accuracy was enhanced through careful preprocessing, including color correction, confidence thresholds, and bounding box size filters. The result was a detection system capable of recognizing turn signs, stops, and intersections with over 95% precision and minimal latency, allowing the robot to react in real time during autonomous runs.