EcoFusion: Energy-Aware Adaptive Sensor Fusion for Efficient Autonomous Vehicle Perception
TimeWednesday, July 13th10:53am - 11:15am PDT
Location3004, Level 3
Event Type
Research Manuscript
Embedded System Design Methodologies
Embedded Systems
DescriptionCurrently, autonomous vehicles (AVs) use multiple sensors, large deep-learning models, and powerful hardware platforms to perceive the environment and navigate safely. However, the high energy demands of these systems can reduce vehicle range significantly. In many contexts, some sensing modalities can negatively impact perception while increasing energy consumption. In this work, we propose EcoFusion: a context- and energy-aware sensor fusion approach that dynamically switches between different sensor combinations to reduce energy consumption without reducing perception performance in comparison to both early and late fusion methods. We additionally propose several context-identification strategies, analyze the energy-performance trade-off, and present scenario-specific results.