EASE-E: Edge-AI based System for Energy-Efficiency in Autonomous Driving (ADAS/AD)

The rise of Software-Defined Vehicles (SDVs) has rapidly advanced the development of Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicle (AV) technology. However, as compute and sensing architectures for SAE Level 2 vehicles increasingly lean towards fully centralized systems, significant concerns arise regarding their energy demands. This shift may have a negative impact on one of the most critical purchasing factors for Battery Electric Vehicles (BEVs): electric driving range. This thesis investigates the potential benefits of decentralization in automotive Electrical/Electronic (E/E) architecture, powered by System-on-Module (SoM) Edge-AI boards. By facilitating efficient deep learning processing locally, the proposed EASE-E (Edge-AI based System for Energy Efficiency) solution achieves up to a 5x reduction in power consumption while maintaining high processing performance. Through a combination of bench testing and Software-in-the-Loop (SiL) simulations, this research demonstrates that EASE-E enhances energy efficiency by 32.8% in highway driving, and 10.8% in urban environments. EASE-E also offers greater scalability and resilience when compared to the existing E/E architectures: distributed, domain, and zonal. The findings underscore the potential of this solution to preserve and extend the driving range of BEVs, presenting a compelling alternative to a fully centralized approach. These insights are crucial for the future design of scalable, energy efficient, and autonomous software-defined vehicles.