Accurate 3D pathfinding for multiple locomotion types is crucial for indoor applications. Many of the existing methods rely on the correct topology or the rich semantics of the indoor space, which restricts their usage in unstructured and semantic-less environments. Though space-decomposition methods, such as voxelization, have been applied to support 3D pathfinding, two major defects remain: the first is that voxelization of a coarse resolution brings about discretization artifacts, which is detrimental to the accessibility of wheeled platforms. The second is that the movement behaviors of different locomotion types were not modeled and integrated with a consistent world representation. This paper employs the octree to reconstruct the indoor spaces in multiresolution, especially to differentiate between slopes and steps. A height-related spatial weighting is proposed for modeling the 3D movement cost. The behaviors of typical locomotion types are modeled by the variable-length search in continuous coordinate space and compose a unified hybrid A* pathfinding algorithm. Experiments show this method is robust and can be applied to both structured and unstructured building models. The results are accurate 3D paths conforming to the movement behaviors of multiple locomotion types.

多种运动类型的准确 3D 寻路对于室内应用至关重要。许多现有方法依赖于正确的拓扑结构或室内空间的丰富语义，这限制了它们在非结构化和无语义环境中的使用。虽然空间分解方法（例如体素化）已用于支持 3D 寻路，但仍然存在两个主要缺陷：第一是粗分辨率的体素化会带来离散化伪影，这不利于轮式平台的可访问性。第二个是不同运动类型的运动行为没有被建模和整合到一致的世界表示中。本文采用八叉树以多分辨率重建室内空间，特别是区分坡度和台阶。提出了一种与高度相关的空间加权来对 3D 移动成本进行建模。通过在连续坐标空间中的变长搜索对典型运动类型的行为进行建模，并组成统一的混合A*寻路算法。实验表明，该方法具有鲁棒性，可应用于结构化和非结构化建筑模型。结果是符合多种运动类型的运动行为的准确 3D 路径。