POMP (2D demo)

地图构建是机器人与自主系统中的关键环节,因为它为路径规划提供空间基础。高效的建图能够帮助规划算法在复杂环境中生成可靠路径,同时兼顾安全性与实时适应能力。固定分辨率的建图方法常会产生过于保守的障碍表示,进而在拥挤场景中导致次优路径甚至规划失败。为解决这一问题,我们提出 Parallel OctoMapping (POMP),一种高效的、基于 OctoMap 的建图方法,能够最大化可用自由空间并支持多线程计算。据我们所知,POMP 是首个在不改变底层 occupancy grid 结构的前提下,在固定分辨率下细化自由空间表示、同时保持与现有基于搜索的规划器兼容的方法。因此,它可以无缝集成到现有规划流程中,在拥挤环境下获得更高的寻路成功率和更短的路径长度,并显著提升计算效率。

Tutorial

  1. Brush on the canvas to create the point cloud.
  2. Adjust Safe / unsafe ratio. The amount of free space is updated automatically.
  3. Adjust grid rows and cols. The leaf size updates automatically.
  4. Toggle map bound, occupied cells, and octree leaf nodes when needed.
  5. Set Start / Goal and run A* on the mapped grid.

This browser demo is designed for visualization and interaction. It runs the mapping and planning pipeline serially in JavaScript and does not use the parallel acceleration in the full POMP implementation.

论文与引用

论文(arXiv)
@Misc{POMP.SCC.Mao.Tan.ea2026,
  author       = {Yihui Mao and Tian Tan and Xuehui Shen and Warren E. Dixon and Rushikesh Kamalapurkar},
  title        = {{Parallel OctoMapping}: A Scalable Framework for Enhanced Path Planning in Autonomous Navigation},
  howpublished = {{arXiv}:2603.22508},
  year         = {2026},
}
当前模式:Paint obstacle 鼠标位置:(-, -) Grid 索引:(-, -) Brush:0.22
Grid:40 × 50 Leaf:-
Brush to set the point cloud.
POMP A* path Direct OGM A* path