FALCON: Fast Autonomous Aerial Exploration using Coverage Path Guidance

TL;DR

FALCON introduces a fast autonomous aerial exploration framework that leverages coverage path guidance to efficiently map unknown 3D spaces with limited field of view sensors. The method combines online connectivity-aware space decomposition, an incremental connectivity graph, a global coverage path, and a CP-guided local SOP-based planning, followed by minimum-time trajectory generation. Extensive simulations, an onboard real-world demonstration, and ablation studies show that FALCON outperforms state-of-the-art planners in exploration speed, robustness, and map completeness, while maintaining real-time responsiveness. A lightweight exploration planner evaluation environment and objective VECO criteria support fair benchmarking and practical deployment, with open-source release of the code base for community use.

Abstract

This paper introduces FALCON, a novel Fast Autonomous expLoration framework using COverage path guidaNce, which aims at setting a new performance benchmark in the field of autonomous aerial exploration. Despite recent advancements in the domain, existing exploration planners often suffer from inefficiencies such as frequent revisitations of previously explored regions.FALCON effectively harnesses the full potential of online generated coverage paths in enhancing exploration efficiency.The framework begins with an incremental connectivity-aware space decomposition and connectivity graph construction, which facilitate efficient coverage path planning.Subsequently, a hierarchical planner generates a coverage path spanning the entire unexplored space, serving as a global guidance.Then, a local planner optimizes the frontier visitation order, minimizing traversal time while consciously incorporating the intention of the global guidance.Finally, minimum-time smooth and safe trajectories are produced to visit the frontier viewpoints.For fair and comprehensive benchmark experiments, we introduce a lightweight exploration planner evaluation environment that allows for comparing exploration planners across a variety of testing scenarios using an identical quadrotor simulator.Additionally, an in-depth analysis and evaluation is conducted to highlight the significant performance advantages of FALCON in comparison with the state-of-the-art exploration planners based on objective criteria.Extensive ablation studies demonstrate the effectiveness of each component in the proposed framework.Real-world experiments conducted fully onboard further validate FALCON's practical capability in complex and challenging environments.The source code of both the exploration planner FALCON and the exploration planner evaluation environment has been released to benefit the community.

Figures (19)

• Figure 1: A quadrotor with limited sensor field of view (FoV) autonomously explored a challenging cluttered indoor environment of dimension $24 \times 6 \times 2$$m^3$. The online constructed volumetric map and the flight trajectory with the yaw direction as a purple arrow are shown in the top-left image. Two detailed close-ups of the bridges from the volumetric map result are displayed in the right images. Video of the experiments is available at https://youtu.be/BGH5T2kPbWw.
• Figure 2: Overview of the proposed planner FALCON for fast autonomous exploration. The framework comprises two main components: exploration planning preprocessing and CP-guided hierarchical exploration planning. The former part provides fundamental information, including space decomposition and the connectivity graph for coverage path planning, as well as extraction of frontiers and viewpoints for local planning. The latter part performs hierarchical exploration planning, which generates a global coverage path (CP) spanning the entire unexplored space and optimizes frontier visitation order consciously incorporating CP's intention. These modules are consistently updated and replanned until exploration concludes.
• Figure 3: A snapshot illustrating the results of space decomposition and connectivity graph construction during the exploration. The surrounding transparent cells are not intersected with the bounding box of the updated map and kept unchanged. The two zones highlighted in yellow are connected by an inaccessible narrow corridor and consequently grouped into seperate zones. The middle-right rectangle is an inaccessible hollow and excluded from exploration planning utilizing the connectivity graph.
• Figure 4: Examples for the three types of restricted A* performed when constructing (a) the free subgraph $\mathcal{G}_f$, (b) the unknown subgraph $\mathcal{G}_u$ and (c) the portal edges $E_p$ respectively.
• Figure 5: The top image illustrates the viewpoint qualification assessment, where the assessed viewpoint is disqualified due to its observation of an insufficient number of unknown voxels. The blue lines represent raycast samplings within camera FoV used for counting unknown voxels. The bottom image shows an example of the histogram and the distribution of the numbers of unknown voxels at all viewpoints, where the qualification cutoff line $\mu - z\sigma$ is highlighted in a dotted blue line.

Theorems & Definitions (2)

• Lemma 1: Connected Components Labeling
• Lemma 2: Sequential Ordering Problem