HEROS: Hierarchical Exploration with Online Subregion Updating for 3D Environment Coverage

TL;DR

The paper addresses efficient exploration of unknown 3D environments under computational constraints. It introduces HEROS, a framework built on rapid environmental preprocessing, a variable-resolution subregion division with an online Subregion Information Structure, and a hierarchical planning strategy that performs global coverage path planning over subregions followed by local viewpoint optimization guided by the global plan. By representing unknown space with online adaptive subregions and solving an open-loop asymmetric TSP for region visitation plus constrained local optimization, the approach balances exploration efficiency and computational cost. Experimental results in simulation and a real-world underground garage demonstrate faster exploration and lower resource usage compared with state-of-the-art baselines, validating the method's practicality for large-scale and dynamic environments.

Abstract

We present an autonomous exploration system for efficient coverage of unknown environments. First, a rapid environment preprocessing method is introduced to provide environmental information for subsequent exploration planning. Then, the whole exploration space is divided into multiple subregion cells, each with varying levels of detail. The subregion cells are capable of decomposition and updating online, effectively characterizing dynamic unknown regions with variable resolution. Finally, the hierarchical planning strategy treats subregions as basic planning units and computes an efficient global coverage path. Guided by the global path, the local path that sequentially visits the viewpoint set is refined to provide an executable path for the robot. This hierarchical planning from coarse to fine steps reduces the complexity of the planning scheme while improving exploration efficiency. The proposed method is compared with state-of-art methods in benchmark environments. Our approach demonstrates superior efficiency in completing exploration while using lower computational resources.

Figures (8)

• Figure 1: The real-world experiment conducted in an underground garage using real ground platforms. The yellow dot denotes the start position, and the white line indicates the robot's trajectory. Video of the experiment is available at: https://youtu.be/5shy7c1Faf4.
• Figure 2: System overview. The environmental preprocessing is first conducted, which includes frontier detection, building sparse graph, and viewpoint generation. Then, the entire exploration environment is divided into several subregion cells, and the SIS of each cell is stored for global coverage path planning and local path optimization. Finally, the robot follows the local path to explore.
• Figure 3: Visualization effect of environmental preprocessing. During the exploration process, new nodes are continuously sampled to expand the sparse graph $\mathcal{G}$. Viewpoints are obtained by sampling from $\mathcal{G}$ through generating collision-free spheres. All frontier points within the sphere can be observed from the viewpoint.
• Figure 4: An example of the regional division with two levels. The original partitioned regions will be further subdivided if the known voxel ratio inside exceeds a threshold. Regions with no viewpoints inside will be considered as being in inactive status.
• Figure 5: Illustration of hierarchical planning. The red arrows represent the global coverage path $\mathcal{G}_{\text{global}}$ that sequentially enters all subregions in active status, while the dashed lines indicate the optimized local path $\mathcal{T}_{\text{local}}$ guided by the global path $\mathcal{G}_{\text{global}}$. The local path visits all viewpoints within $r_{\text{max}}$.