Integration of AI-based autonomous systems in real-time environments is advancing rapidly. Technologies enhance ADAS and broaden autonomous vehicle capabilities, emphasizing trust and user acceptance. Achieving safe lane-keeping performance requires demonstrators that operate in real time. This paper introduces “RaspiCar”, research platform tailored to evaluate control algorithms within embedded systems. Traditional research emphasizes accuracy, while “RaspiCar” explores AI safety and real-time performance. To ensure safety and real-time decisions, architecture with ZeroMQ is integrated, enabling implementation on resource-limited hardware. The platform’s real-time performance is evaluated using PilotNet for steering inference and LaneNet and SegNet for lane segmentation, selected for their complementary strengths. Additionally, a safety mechanism is introduced to enforce emergency stops, mitigating hardware faults and model errors. We demonstrate that real-time AI-driven lane detection runs reliably and safely on low-power platforms. “RaspiCar” provides a scalable framework for advancing real-time autonomy in robotics and industry, enhancing the reliability of lane detection.