EPIC: A Lightweight LiDAR-Based UAV Exploration Framework for Large-Scale Scenarios

TL;DR

The paper tackles autonomous exploration of large-scale 3D environments with UAVs using LiDAR, addressing the high memory and computational costs of traditional representations. It introduces EPIC, which directly operates on point clouds via an observation map derived from surface observation quality, and an incremental topological graph built on the same data to accelerate path planning. These components feed a hierarchical planner with an incremental global guidance path and a local MINCO-based trajectory generator, achieving fast exploration with low memory and computation. Empirical results in simulation and real-world tests show reduced memory footprint and planning time while maintaining exploration efficiency and trajectory quality, demonstrating practical viability for large-scale unmanned exploration.

Abstract

Autonomous exploration is a fundamental problem for various applications of unmanned aerial vehicles (UAVs). Recently, LiDAR-based exploration has gained significant attention due to its ability to generate high-precision point cloud maps of large-scale environments. While the point clouds are inherently informative for navigation, many existing exploration methods still rely on additional, often expensive, environmental representations. This reliance stems from two main reasons: the need for frontier detection or information gain computation, which typically depends on memory-intensive occupancy grid maps, and the high computational complexity of path planning directly on point clouds, primarily due to costly collision checking. To address these limitations, we present EPIC, a lightweight LiDAR-based UAV exploration framework that directly exploits point cloud data to explore large-scale environments. EPIC introduces a novel observation map derived directly from the quality of point clouds, eliminating the need for global occupancy grid maps while preserving comprehensive exploration capabilities. We also propose an incremental topological graph construction method operating directly on point clouds, enabling real-time path planning in large-scale environments. Leveraging these components, we build a hierarchical planning framework that generates agile and energy-efficient trajectories, achieving significantly reduced memory consumption and computation time compared to most existing methods. Extensive simulations and real-world experiments demonstrate that EPIC achieves faster exploration while significantly reducing memory consumption compared to state-of-the-art methods.

Figures (7)

• Figure 1: Autonomous exploration of a large-scale scene with complex structures, spanning thousands of square meters. The central image displays the exploration result, showing the online-generated point cloud map and the UAV's flight trajectory. The trajectory color intensity indicates the UAV's velocity, with darker hues representing higher speeds. The side images illustrate two snapshots during the exploration process: the UAV's first-person view, current pose, and locally planned trajectories. The color of the local trajectories also denotes velocity, ranging from red (maximum speed) to blue (stationary).
• Figure 2: Evaluation of view direction quality. (a): A 2D example illustrating how adjacent LiDAR rays are used to evaluate view direction quality. (b): A pyramidal-shaped volume is formed by four adjacent LiDAR rays, and the view direction quality of a voxel, whose center falls within this volume, is determined by these four LiDAR rays.
• Figure 3: Illustration of topological graph vertex updating process in a region. (a) and (b): Free space coverage using collision-free spheres. (c): Clustering spheres based on connectivity. (d): Selection representative vertices.
• Figure 4: This figure shows the process of incremental updating of the topological graph as UAV flies along the trajectory. For clarity, we present a 2D example and assume that each region has only four neighbors (up, down, left, and right). However, in the actual implementation, to achieve better connectivity, we consider diagonally adjacent regions as neighbors as well.
• Figure 5: Viewpoint selection. (a): Uniformly sampling candidate viewpoints in a cylindrical coordinate system centered at the frontier cluster. (b): The process of clustering candidate viewpoints based on connectivity.