AI-INTEGRATED AUTONOMOUS GROUND VEHICLE WITH EMBEDDED LANE TRACKING, OBSTACLE AVOIDANCE, AND VISION-BASED SIGNAL INTERPRETATION

Autonomous navigation has emerged as a transformative domain in modern intelligent systems, finding applications across transportation, logistics, surveillance, and robotics. This paper presents the design, integration, and experimental evaluation of a low-cost AI-driven autonomous ground vehicle capable of performing real-time lane tracking, obstacle avoidance, and traffic signal interpretation. The system employs a Raspberry Pi 4B as its central processing unit executing a Python-based control stack. Lane boundaries are identified via a Gaussian–Canny–Hough pipeline operating on live camera frames at 30 FPS with a latency under 35 ms. Obstacle proximity is continuously monitored using a VL53L0X time-of-flight sensor; objects detected within a 150 mm threshold trigger a deterministic four-step overtaking routine. Traffic signal recognition is powered by the GPT-4o-mini multimodal Vision API, which classifies junction images as RED or GREEN with 100% accuracy across 40 test cases. Obstacle avoidance achieved an 85% success rate under controlled indoor conditions. The total prototype cost remains below Rs 7,500 (~USD 90), demonstrating a viable, reproducible, and affordable framework for academic research in autonomous robotics.