Beyond Car Sharing: Uncertainty-Aware Pooling of Vehicular Compute at the Network Edge
2026-07-20 • Networking and Internet Architecture
Networking and Internet Architecture
AI summaryⓘ
The authors address the challenge of using computing power from connected vehicles, which constantly change how much compute they can offer. They created SMART, a system that predicts how much vehicle compute will be available and uses this uncertain prediction to decide whether to accept tasks. Their method balances admitting many tasks while rarely exceeding available capacity, performing better than other strategies that ignore prediction uncertainty. They also show that varying compute at the base station affects how well admissions work, emphasizing the need for smart, uncertainty-aware management.
Connected VehiclesAI AcceleratorsMobile Edge Computing (MEC)Bayesian Neural Network (BNN)Predictive UncertaintyChance-Constrained OptimizationSample Average Approximation (SAA)Conditional Value at Risk (CVaR)Resource Admission ControlCompute Capacity Violation
Authors
Wellington Lobato, Nadjib Achir, Aline Carneiro Viana
Abstract
Connected vehicles increasingly embed AI accelerators, offering a substantial yet volatile source of supplemental compute near the network edge. Unlike provisioned MEC hosts, vehicular resources are highly dynamic: vehicles may leave the cell, become locally occupied, or offer heterogeneous compute capacities. Therefore, exploiting vehicular resources requires making admission decisions without knowing the compute capacity that will be available during task execution. We present SMART, an uncertainty-aware admission mechanism that enables an \acs{ETSI} \ac{MEC} orchestrator \textit{to opportunistically exploit vehicular compute under predictive uncertainty.} SMART predicts future vehicular capacity using a \ac{BNN} -- whose uncertainty estimates are the best calibrated among the evaluated forecasters at the nominal 95\% level, and incorporates its calibrated predictive uncertainty into a chance-constrained admission program reformulated through \ac{SAA} and \ac{CVaR} approximations. Under a compute-only admission model, SMART admits 95.6\% of tasks while maintaining a median capacity-violation rate of about 0.81\%. It achieves a favorable admission-violation tradeoff compared with seven reactive, mean-only, and uncertainty-aware baselines, by approaching the performance of a compute-capacity oracle under the modeled assumptions. Finally, the sensitivity analyses show that variability in base-station compute availability is a key determinant of admission performance, highlighting the need for a calibrated admission and resource allocation mechanism tailored to opportunistic base-station compute pooling.