[Paper Review] Graph-Based Topological Exploration Planning in Large-Scale 3D Environments

Fan Yang, Dung-Han Lee+2 more|arXiv (Cornell University)|Mar 31, 2021

Robotics and Sensor-Based Localization15 references4 citations

TL;DR

This paper proposes a graph-based topological exploration planning framework that uses convex polyhedron-based 3D geometry estimation to build a sparse topological map, enabling efficient, high-level exploration in large-scale 3D environments. By grouping frontier regions into distinct topological nodes and using a dual-level planning strategy, the method reduces replanning frequency and improves exploration efficiency by over 40% in simulation and demonstrates robust real-world performance in underground mine environments.

ABSTRACT

Currently, state-of-the-art exploration methods maintain high-resolution map representations in order to optimize exploration goals in each step that maximizes information gain. However, during exploring, those "optimal" selections could quickly become obsolete due to the influx of new information, especially in large-scale environments, and result in high-frequency re-planning that hinders the overall exploration efficiency. In this paper, we propose a graph-based topological planning framework, building a sparse topological map in three-dimensional (3D) space to guide exploration steps with high-level intents so as to render consistent exploration maneuvers. Specifically, this work presents a novel method to estimate 3D space's geometry with convex polyhedrons. Then, the geometry information is utilized to group space into distinctive regions. And those regions are added as nodes into the topological map, directing the exploration process. We compared our method with the state-of-the-art in simulated environments. The proposed method achieves higher space coverage and outperforms exploration efficiency by more than 40% during experiments. Finally, a field experiment was conducted to further evaluate the applicability of our method to empower efficient and robust exploration in real-world environments.

Motivation & Objective

• To address the high computational cost and frequent replanning in state-of-the-art 3D exploration methods due to rapidly changing information in large-scale environments.
• To reduce planning redundancy by constructing a sparse topological map based on geometric contiguity and mutual visibility in 3D space.
• To enable consistent, high-level exploration maneuvers by grouping frontier regions into distinct topological nodes for long-term planning.
• To improve exploration efficiency through a dual-level planning strategy that allows short-cutting through uncertain regions.
• To validate the method’s robustness and efficiency in both simulated and real-world subterranean environments.

Proposed method

• The method estimates 3D space geometry using multiple convex polyhedrons to represent known regions, enabling clean frontier extraction and region grouping.
• Frontier regions are grouped into distinctive exploration regions based on geometric contiguity and mutual visibility, forming nodes in a topological graph.
• A sparse topological map is constructed from these grouped regions, representing high-level exploration destinations for long-term planning.
• A bifurcated planning strategy is employed: local planning guides consistent maneuvers toward nearby topological nodes, while global planning enables short-cutting through partially known or uncertain areas.
• The exploration planner uses a detection range of 15m and coverage range of 10m to define frontier space, ensuring reliable information extraction and path feasibility.
• Real-time execution is achieved via an on-board system with LiDAR, depth cameras, and a motion-primitives planner on an Intel NUC, enabling autonomous flight in complex underground settings.

Experimental results

Research questions

• RQ1 Can a geometry-based topological map built from convex polyhedrons reduce planning redundancy and improve exploration efficiency in large-scale 3D environments?
• RQ2 How does the proposed topological planning framework compare to state-of-the-art methods in terms of space coverage and computational efficiency in simulation?
• RQ3 Can the method enable consistent, high-level exploration maneuvers in real-world subterranean environments with GPS-denied, textureless, and complex structures?
• RQ4 To what extent does the ability to short-cut through uncertain regions reduce backtracking and improve overall exploration speed?
• RQ5 How effective is the integration of geometric region grouping with a dual-level planning strategy in maintaining exploration continuity and coverage?

Key findings

• The proposed method outperforms the state-of-the-art method by more than 40% in exploration efficiency in simulated environments, measured by space coverage over time.
• The method achieved an average exploration speed of 0.8 m/s in real-world flight, approaching the maximum speed limit of 1.0 m/s, indicating consistent and efficient decision-making.
• The field experiment at Brady’s Bend mine demonstrated the planner’s ability to maintain consistent flight patterns along main hallways and redirect to new branches after local dead-ends, ensuring continuous exploration.
• The reconstructed point cloud map from the real-world flight showed faithful coverage of the environment, validating the method’s robustness in complex, GPS-denied underground settings.
• The topological map enabled effective short-cutting through uncertain regions, minimizing backtracking and reducing reliance on high-frequency replanning.
• The method successfully reduced computational cost and planning redundancy by shifting from high-resolution map maintenance to sparse topological representation based on geometric region grouping.