Abstract: Aiming at problems of slow update of three-dimensional aerial map, poor dynamic obstacle avoidance ability and lack of global perception in UAV navigation, this study proposes a low-altitude dynamic navigation method based on visual AI and three-dimensional grid computing. The three-dimensional spatial grid representation system is constructed, and grid index is used to efficiently represent spatial information association. Combined with the airborne visual AI model, the environmental semantic information is perceived in real time, and the perceptual results are mapped to the three-dimensional spatial grid in real time through the semantic feature embedding algorithm to support the autonomous obstacle avoidance flight of the UAV. In local grid corresponding to the same sub-level airspace, a self-organizing communication network is constructed to share position, speed, height and other states of UAV in real time, forming a low-altitude dynamic information pool, so that UAV can accurately grasp the neighboring state in high-density areas such as take-off and landing areas. Recursive subdivision of three-dimensional grid, hierarchical search path, through the local grid interactive negotiation protocol to achieve overload grid intelligent shunt. Taking the low-altitude scene of 50 square kilometers of city as the experimental object, three kinds of scenes, simple, medium and complex, are set up to compare the proposed method with GNSS pure positioning navigation and monocular visual feature point navigation. The results show that in the three scenarios, the average obstacle avoidance efficiency of the proposed method is 96.7% ± 1.2%, 95.3% ± 1.1% and 93.8 % ± 1.5%, respectively, which is 10.5, 10.8 and 11.7 percentage points higher than that of GNSS pure positioning navigation, and 26.9, 27.8 and 28.6 percentage points higher than that of monocular visual feature point navigation. The experimental results show that the low-altitude dynamic navigation method based on visual AI and three-dimensional grid computing can effectively solve the core pain points of UAV navigation, and significantly improve navigation efficiency and safety in complex low-altitude environment.

Key words: grid agent, visual AI, low-altitude route planning, UAV navigation, automatic obstacle avoidance

摘要： 针对无人机导航中三维航图更新慢、动态避障能力差、全局感知缺乏等问题，本研究提出一种基于视觉AI与三维网格计算的低空动态导航方法。构建立体三维空域网格表意体系，利用网格索引高效表示空间信息关联，结合机载视觉AI模型实时感知环境语义信息，并通过语义特征嵌入算法，将感知结果实时映射到三维空域网格中，支持无人机自主避障飞行。在同一子级空域对应的局部网格内，构建自组织通信网络，实时共享无人机位置、速度、高度等状态，形成低空动态信息池，使无人机能够在类似起降区等的高密度区域精准掌握邻机状态。采用递归剖分立体网格，层级搜路径，通过局部网格交互协商协议实现过载网格智能分流。以50平方公里城市低空场景为实验对象，设简单、中等、复杂三类场景，对比所提方法与GNSS纯定位导航、单目视觉特征点导航。结果显示：三类场景中，所提方法平均避障有效率分别为96.7%±1.2%、95.3%±1.1%、93.8%±1.5%，较GNSS纯定位导航分别提升10.5、10.8、11.7个百分点，较单目视觉特征点导航分别提升26.9、27.8、28.6个百分点。实验结果表明，基于视觉AI与三维网格计算的低空动态导航方法，能有效解决无人机导航核心痛点，显著提升复杂低空环境下导航效率与安全性。

关键词: 网格智能体, 视觉AI, 低空路线规划, 无人机导航, 自动避障