Operating large autonomous fleets demands fast, resilient allocation of scarce resources (such as energy and fuel, charger access and maintenance slots, time windows, and communication bandwidth) under uncertainty. We propose a side-information-aware approach for resource allocation at scale that combines distributional predictions with decentralized coordination. Local side information shapes per-agent risk models for consumption, which are coupled through chance constraints on failures. A lightweight consensus-ADMM routine coordinates agents over a sparse communication graph, enabling near-centralized performance while avoiding single points of failure. We validate the framework on real urban road networks with autonomous vehicles and on a representative satellite constellation, comparing against greedy, no-side-information, and oracle central baselines. Our method reduces failure rates by 30-55% at matched cost and scales to thousands of agents with near-linear runtime, while preserving feasibility with high probability.

翻译：运营大规模自主车队需要在不确定性下快速、弹性地分配稀缺资源（如能源与燃料、充电桩接入与维护时段、时间窗口及通信带宽）。本文提出一种面向大规模资源分配的侧信息感知方法，该方法将分布预测与去中心化协调相结合。局部侧信息构建了各智能体的消耗风险模型，并通过故障机会约束实现耦合。轻量级共识-交替方向乘子法（ADMM）在稀疏通信图上协调智能体，在避免单点故障的同时实现接近集中式调度的性能。我们在真实城市路网的自动驾驶车辆及典型卫星星座场景中验证了该框架，并与贪婪策略、无侧信息基准及理想集中式基准进行对比。结果表明，在同等成本下本方法将故障率降低30-55%，可扩展至数千智能体且运行时间接近线性，同时以高概率保持方案可行性。