An Offline Fully Autonomous Driving Navigation System for Road Networks in Extremely Complex Mountain Areas Based on Hierarchical Gradient Matching Geometry 基于层级梯度匹配几何的极端复杂山区路网离线全自主驾驶导航系统

Engineering notes

Key topics: autonomous driving, path planning, radar, perception, planning. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

摘要 Abstract 当前主流自动驾驶方案高度依赖卫星导航与联网高精地图,在极端复杂山区、偏远峡谷、矿区林场等无信号、弱覆盖区域易出现定位失效、地图缺失等问题,难以支撑连续自主行驶。现有地形辅助导航方案多采用一维高程剖面匹配,特征维度低、区分度不足,复杂路网场景下匹配鲁棒性有限。本文以三维降维层级梯度匹配几何理论为核心底座,将山地无人机突防的地形指纹技术体系平滑迁移至地面车辆场景,构建路网带状地形指纹 + 车载多源感知 + 本地全链路计算的全离线自主导航架构。系统任务前离线构建道路纵向高程梯度与两侧山体横向截面梯度复合的带状地形指纹库;车载端以量产毫米波雷达为常态主感知设备,短时复用可见光 / 红外摄像头补全轮廓细节,全程零外源信号输入,依靠本地特征匹配实现车道级自主定位与路径规划。整套方案不依赖卫星信号、蜂窝网络与基站定位,完全免疫信号干扰与坐标诱骗,可在极端复杂山区路网实现全天候离线全自主驾驶,硬件可直接复用量产车载传感器,改造成本低,是无信号偏远区域自动驾驶的核心技术路径,在山区客运、应急救援、矿区作业等场景具备显著应用价值。 Current mainstream autonomous driving solutions rely heavily on satellite navigation and networked high-definition maps. In areas with no or weak signal coverage, such as extremely complex mountainous regions, remote canyons, mining areas and forest farms, problems such as positioning failure and map absence frequently occur, making it difficult to support continuous autonomous driving. Most existing terrain-aided navigation schemes adopt one-dimensional elevation profile matching, which suffers from low feature dimensionality and insufficient discriminability, resulting in limited matching robustness in complex road network scenarios.Taking the three-dimensional dimensionality-reduction hierarchical gradient matching geometry theory as the core foundation, this paper smoothly migrates the terrain fingerprint technology system for mountain UAV penetration to ground vehicle scenarios, and constructs a fully offline autonomous navigation architecture featuring road network strip terrain fingerprint + vehicle-mounted multi-source perception + local full-link computing. The system offline builds a strip terrain fingerprint database compounded by road longitudinal elevation gradients and lateral cross-section gradients of mountains on both sides before missions. On the vehicle side, mass-produced millimeter-wave radars serve as the normal main perception equipment, and visible/infrared cameras are temporarily reused to supplement contour details. With zero external signal input throughout the process, the system realizes lane-level autonomous positioning and path planning relying on local feature matching.The entire solution is independent of satellite signals, cellular networks and base station positioning, and is completely immune to signal interference and coordinate spoofing. It can realize all-weather offline fully autonomous driving on road networks in extremely complex mountain areas. The hardware can directly reuse mass-produced vehicle-mounted sensors with low modification costs. As a core technical path for autonomous driving in remote areas without signals, it has remarkable application value in scenarios such as mountain passenger transport, emergency rescue and mining operations. 关键词:层级梯度匹配几何;带状地形指纹;离线自主导航;无卫星定位;山区自动驾驶;多源感知融合Keywords: Hierarchical Gradient Matching Geometry; strip terrain fingerprint; offline autonomous navigation; satellite-free positioning; mountain area autonomous driving; multi-source perception fusion

Links and sources

Comments