ColAG: A Collaborative Air-Ground Framework for Perception-Limited UGVs' Navigation ColAG：面向感知受限无人车的空地协同导航框架

Equipping every ground robot with a full perception and SLAM stack can be unnecessarily expensive. ColAG instead gives the team one perception-equipped UAV and lets UGVs operate with wheel odometry and relative-localization hardware only. The UAV builds and shares the map, while each UGV fuses relative observations with wheel odometry, plans in the shared map and propagates its localization uncertainty to predict collisions with obstacles or unknown space. Each predicted risk becomes a support request with a location and time window. ColAG formulates the UAV schedule as a dynamic Vehicle Routing Problem with Time Windows, so one aerial robot can update maps and relative poses for the UGVs before predicted failures while limiting their waiting time. The framework is evaluated with up to seven UGVs in simulation and with one UAV and three UGVs in real-world experiments.

为每台无人车配置完整的环境感知与 SLAM 系统成本较高。ColAG 只为集群配置一架具备完整感知能力的无人机，无人车侧仅保留轮式里程计与相对定位硬件。无人机负责建图并共享地图；每台无人车融合机间相对观测与轮式里程计，在共享地图中规划轨迹，同时传播定位不确定性，预测其与障碍物或未知区域发生碰撞的风险。支援请求同时给出需要帮助的位置与时间窗口。ColAG 将无人机调度建模为动态带时间窗车辆路径问题，使一架无人机能够在预测风险发生前为多台无人车更新地图和相对位姿，并尽量减少等待时间。系统在仿真中验证至 7 台无人车，并完成 1 架无人机与 3 台无人车的真机实验。