SEO: Safety-Aware Energy Optimization Framework for Multi-Sensor Neural Controllers at the Edge
DescriptionRuntime energy management is quintessential for multi-sensor autonomous systems at the edge for achieving high performance under platform constraints. However, their controllers are typically designed with certain safety guarantees that precede in priority such optimizations. Therefore, we propose a novel framework for adopting energy optimizations under safety guarantees, where a system's safety properties are characterized as safe dynamic deadlines to govern the operation of the involved multi-sensor processing pipelines. Our experiments using an autonomous driving system model for two optimization techniques -- offloading and gating -- show that energy gains (up to 25%) can still be attained under safety guarantees.
TimeThursday, July 13th4:55pm - 5:10pm PDT
Location3004, 3rd Floor
Design of Cyber-physical Systems and IoT