【摘要】 面向应急通信场景下无人机高清视频回传等低时延任务，通感一体化技术可利用环境感知信息辅助毫米波波束追踪，从而提升空地通信链路的稳定性与传输效率。然而，在基站侧，高频雷达感知任务与计算卸载任务之间存在着激烈的算力竞争。基于此，构建了联合边缘服务器资源分配、任务队列调度和无人机轨迹规划的优化框架。首先，显式刻画了感知过程对算力资源的预占代价，针对混合低时延任务流设计了基于优先级的排队调度机制与超时丢弃机制；其次，将多无人机轨迹规划、任务卸载决策和算力分配比例统一建模为马尔可夫决策过程，提出一种基于势能引导奖励的近端策略优化算法，该算法利用集中式Actor-Critic架构评估全局状态价值，并创新性地引入人工势场引导奖励，有效克服了长航时轨迹优化中高维连续动作空间下的稀疏奖励与位置震荡难题。仿真结果表明，所提方法能够在保障感知精度与任务完成率的前提下，有效降低无人机任务的平均计算时延，验证了通感算资源协同调度机制在无人机低时延任务场景中的有效性。更多还原

【Abstract】 To address low-latency tasks such as high-definition video transmission by unmanned aerial vehicles(UAVs) in emergency communication scenarios, integrated sensing and communication technology can utilize environmental sensing information to assist millimeter-wave beam tracking, thereby improving the stability and transmission efficiency of air-to-ground communication links. However, at the base station side, fierce competition for computing resources exists between high-frequency radar sensing tasks and computation offloading tasks. To address the challenge, this paper constructs a joint optimization framework encompassing edge server resource allocation, task queue scheduling, and UAV trajectory planning. First, the computational resource preemption cost of the sensing process is explicitly characterized, and a priority-based queuing scheduling along with timeout dropping mechanism is designed for mixed low-latency task flows. Second, multi-UAV trajectory planning, task offloading decisions, and computing power allocation ratios are uniformly modeled as a Markov decision process, and a proximal policy optimization algorithm based on a potential-guided reward is proposed. This algorithm adopts a centralized actor-critic architecture to evaluate the global state value and innovatively introduces an artificial potential field-guided reward, effectively overcoming the challenges of sparse rewards and position oscillation in high-dimensional continuous action spaces during long-duration trajectory optimization. Simulation results demonstrate that the proposed method effectively reduces the average computation latency of UAV tasks while guaranteeing sensing accuracy and task completion rates, validating the effectiveness of the collaborative communication-sensing-computing resource scheduling mechanism in UAV low-latency task scenarios.更多还原