Energy-Aware Collaborative Perception in HetVNets

Abstract

Environmental perception is essential for autonomous vehicles. Collaborative perception, supported by vehicular communication technologies like C-V2X, aggregates data from nearby sensors to extend sensing range and improve decision accuracy. However, it raises communication overhead and computational energy because vehicles process more data than their own sensor input. The issue is exacerbated in heterogeneous vehicular networks (HetVNets). Our measurements show that, under equal workloads, low-capability vehicles incur 10% higher computational latency and energy, resulting in a lower energy-to-accuracy gain ratio and a higher marginal cost than high-capability peers. In reality, users prioritize driving range over marginal accuracy gains. We therefore advocate that collaborative perception should move beyond pure accuracy maximization to include user-centric sustainability goals. We present GreenFusion, an energy-aware collaborative perception scheme that embeds explicit sustainability and fairness metrics. GreenFusion adapts each vehicle's engagement and role according to information value and capability, enabling selective sharing of noteworthy data. In evaluations, GreenFusion maintains perception performance while reducing energy consumption for low-capability vehicles by 81.0% and 31.5% on average compared with fully connected and information-adaptive centralized baselines, respectively. In a typical driving scenario, these savings correspond to a 65.6% increase in driving range, demonstrating practical sustainability benefits without sacrificing perception. GreenFusion reframes collaborative perception from an accuracy-only objective to a balanced accuracy-energy strategy, fostering a more sustainable, practical vehicular networking framework that improves resilience, longevity, and user experience.