AutoEdge: Autonomous Driving Services Orchestration in Unified Aerial Computing
Authors: Awada, U., Gu, Y., Zhang, J., Chen, S., Li, S., Ma, Y., Li, Z., Yang, S.
Journal: IEEE Internet of Things Journal
Publication Date: 01/01/2026
eISSN: 2327-4662
DOI: 10.1109/JIOT.2026.3737705
Abstract:The safety of autonomous vehicles (AVs) depends on their ability to execute a plethora of applications in a timely manner. These applications are complex, they require a substantial amount of resources, and their execution results are often fused to achieve the desired objectives. However, the invehicle edge computing (EC) deployment is often insufficient to provide the required resource capacity. In response, the Internet of autonomous vehicles (IoAV) is being introduced to enable AVs to cooperate with other entities equipped with larger EC resources, such as roadside units (RSUs), and to offload their computationally intensive tasks for swift execution. Nevertheless, fixed-site EC deployments, like RSUs, are incapable of adapting to the mobility of AVs. This can cause delays in the transmission of execution results and increased latency, which could jeopardize the safety of AVs in real-time operations. This paper presents AutoEdge, an intelligent framework for dispatching and scheduling autonomous driving applications within a unified aerial EC system that aims to overcome the limitations of fixed-site EC deployments. Specifically, for a fleet of drones and AVs operating across disjoint locations in a city, we derive a machine learning (ML) linear regression model that estimates each AV’s task resource requirements and execution time. Using these estimated values, we jointly consider each drone’s flight time and EC resources availability to formulate a novel mission scheduling problem (MSP) that selects the most suitable drones to accomplish the missions. Furthermore, we propose a novel fragmentation-aware bin-packing optimization approach that gang-schedules and co-locates applications to fully utilize available resources. Extensive experiments using the Alibaba Cluster Trace demonstrate that AutoEdge achieves a 100% task deployment ratio, reduces response time by up to 48× compared to baseline schemes, and maintains near-zero resource fragmentation, confirming its effectiveness for latency-sensitive autonomous driving applications.
Source: Scopus