A-OctoMap: An Adaptive OctoMap for Online Path Planning

TL;DR

The paper tackles the challenge of preserving geometric detail during downsampling while enabling efficient online path planning in complex environments. It introduces A-OctoMap, an adaptive multiscale OctoMap framework that preserves obstacle boundaries and supports planning-oriented grid projections. Key contributions include a parallel, convex-hull based downsampling method, a dynamic refinement strategy via a Minimum Controllable Region, and a geometry-aware grid mapping compatible with Jump Point Search, with empirical evidence of improved information retention, reconstruction efficiency, and path planning performance. The work offers a practical pathway toward robust, real-time navigation and lays groundwork for tighter integration with full perception–planning–control loops in robotics.

Abstract

Downsampling and path planning are essential in robotics and autonomous systems, as they enhance computational efficiency and enable effective navigation in complex environments. However, current downsampling methods often fail to preserve crucial geometric information while maintaining computational efficiency, leading to challenges such as information loss during map reconstruction and the need to balance precision with computational demands. Similarly, current graph-based search algorithms for path planning struggle with fixed resolutions in complex environments, resulting in inaccurate obstacle detection and suboptimal or failed pathfinding. To address these issues, we introduce an adaptive OctoMap that utilizes a hierarchical data structure. This innovative approach preserves key geometric information during downsampling and offers a more flexible representation for pathfinding within fixed-resolution maps, all while maintaining high computational efficiency. Simulations validate our method, showing significant improvements in reducing information loss, enhancing precision, and boosting the computational efficiency of map reconstruction compared to state-of-the-art methods. For path planning, our approach enhances Jump Point Search (JPS) by increasing the success rate of pathfinding and reducing path lengths, enabling more reliable navigation in complex scenes.

Figures (8)

• Figure 1: OctoTree map with hierarchical voxels.
• Figure 2: Uniform grid map based on Data Structure \ref{['datastructure: uniform grid map']}.
• Figure 3: Illustration of mapping from OctoTree to uniform grid. The top row shows the fixed uniform grid method, while the bottom row shows our adaptive method. Red lines indicate OctoTree splits at different levels; red dotted lines represent SplitBoundary, and blue dotted lines represent NodeBoundary. The right columns visualize the resulting grid structures under both methods.
• Figure 4: Four experimental scenarios used to evaluate the effectiveness of the proposed framework.
• Figure 5: Illustration of mesh triangulation after downsampling.

Theorems & Definitions (3)

• Theorem 1: Pigeonhole principle 4115264
• Definition 1: Minimum Controllable Region (MCR)
• Definition 2