SLO-Aware Task Offloading Within Collaborative Vehicle Platoons

Abstract

In the context of autonomous vehicles (AVs), offloading is essential for guaranteeing the execution of perception tasks, e.g., mobile mapping or object detection. While existing work on offloading focused extensively on minimizing inter-vehicle networking latency, vehicle platoons (e.g., heavy-duty transport) present numerous other objectives, such as energy efficiency or data quality. To optimize these Service Level Objectives (SLOs) during operation, this work presents a purely Vehicle-to-Vehicle approach (V2V) for collaborative services offloading within a vehicle platoon. By training and using a Bayesian Network (BN), services can proactively decide to offload whenever this promises to improve platoon-wide SLO fulfillment; therefore, vehicles estimate how both sides would be impacted by offloading a service. In particular, this considers resource heterogeneity within the platoon to avoid overloading more restricted devices. We evaluate our approach in a physical setup, where vehicles in a platoon continuously (i.e., every 500 ms) interpret the SLOs of three perception services. Our probabilistic, predictive method shows promising results in handling large AV platoons; within seconds, it detects and resolves SLO violations through offloading.