Light electric quadricycles the EU's L6e/L7e class of compact four-wheeled urban electric vehicles are rapidly growing in European cities as affordable, low-footprint alternatives to conventional cars. However, strict cost, size, and weight limitations prevent many of these vehicles from integrating Advanced Driver Assistance Systems (ADAS),commonly found in full-size vehicles. As a result, safety features such as Lane Keeping Assist Systems (LKAS), which help prevent unintended lane departure, remain largely unavailable for this emerging mobility segment.

This project addresses this gap by developing a low-cost, camera only LKAS designed for compact urban electric vehicles and micromobility platforms. Instead of relying on expensive LIDAR, IMU, GPS, or complex automotive hardware, the proposed system combines computer vision, adaptive control, and affordable embedded computing to achieve autonomous lane keeping capability.

The system was first designed, developed, and validated inside the CARLA autonomous driving simulator, where lane perception algorithms, vehicle dynamics, and control strategies were evaluated under controlled conditions. Following simulation validation, the same core perception and control architecture was transferred with minimal adaptation to a 1/10-scale physical robotic prototype powered by a Raspberry Pi 5 and a monocular camera system. The prototype follows a front-wheel Ackermann steering and rear-wheel drive configuration, representing the fundamental vehicle dynamics used in compact four-wheel electric vehicles.

The perception system processes real-time camera input using an OpenCV-based computer vision pipeline. Road images are analyzed through grayscale conversion, adaptive image enhancement, Region of Interest (ROI) filtering, Canny Edge Detection, Probabilistic Hough Transform, and temporal filtering to extract lane boundaries. From the detected lane geometry, the system calculates cross-track error and heading error, which serve as feedback parameters for autonomous steering control.

The first key innovation is the implementation of a Gain-Scheduled Stanley Controller for adaptive path tracking. Unlike conventional fixed-parameter PID controllers, the proposed controller adjusts its behavior according to vehicle conditions and road geometry. By considering both lateral displacement and heading orientation, the system improves curve handling, reduces overshoot, and minimizes steering oscillations in dynamic driving scenarios.

The second key contribution is an Optical Flow-based virtual speedometer that estimates vehicle speed from camera motion when direct vehicle speed data is unavailable. When onboard speed information is accessible, the Optical Flow estimation can also serve as an independent validation method. This approach reduces sensor dependency and supports low-cost mobility applications.

The complete system operates in real-time closed-loop conditions, where camera perception, virtual speed estimation, and adaptive steering control continuously interact. Physical prototype testing demonstrated reliable lane tracking performance, achieving approximately 4–5 cm lateral accuracy while maintaining stable steering behavior.

Because the architecture relies entirely on commercially available components, including a single-board computer, compact camera module, and standard vehicle actuation hardware, it provides a scalable and cost-effective pathway for future integration into lightweight electric vehicles.The modular software structure also enables adaptation to different embedded platforms and vehicle configurations.

As cities move toward cleaner and smaller transportation solutions, this project demonstrates a practical approach for bringing intelligent safety technologies beyond premium vehicles and making autonomous assistance accessible to affordable urban mobility systems.