Edge Computing for Predictive Braking in Software-Defined Vehicles

As the automotive industry shifts toward SoftwareDefined Vehicles (SDVs), managing functional safety in real-time becomes increasingly complex, especially in braking systems. This paper introduces a novel architecture utilizing Edge Computing for predictive braking while leveraging the Braking Safety Performance Index (BSPI) and a newly introduced Real-Time Braking Performance Index (RTBPI). These Key Performance Indicators (KPIs) ensure both localized decision-making at the edge and system-wide coordination through central compute, enhancing safety and efficiency. The BSPI focuses on braking safety metrics, while the RTBPI measures edge AI performance, combining latency, resource utilization, and cross-zonal safety consistency. The architecture also integrates supervised and reinforcement learning models, offering a real-time and long-term balanced solution for braking systems in SDVs. The paper demonstrates how edge computing using Renesas R-Car-V3H devices and centralized compute with Nvidia Orin AGX improves real-time performance while reducing system latency, energy consumption, and improving safety-critical functions.