Organizations: State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, 430079, China · School of Electrical and Electronic Engineering, Nanyang Technological University, 639798, Singapore · Electronic Information School, Wuhan University, Wuhan, 430072, China
Autonomous UAV exploration is increasingly used for large-scale surveying and inspection in inaccessible or disaster-affected environments, where onboard memory and computational resources are often severely constrained. Most existing UAV exploration systems maintain globally queryable geometric map representations for motion planning and exploration decision-making, causing memory consumption and spatial-query costs to increase with the explored area. Robot-centric local mapping bounds dense-map memory and computation but discards global task context needed for long-horizon exploration.We present FLAVOR, a hierarchical spatio-temporal framework that confines dense metric geometry to a robot-centric local map while retaining compact global observation and task context. At the observation level, lightweight LiDAR-keyframe visibility records preserve ray-wise evidence for frontier reasoning. At the task level, unresolved frontiers beyond the local-map extent are retained as compact pending tasks associated with odometry anchors. A sparse topological graph connects these anchors through retained routes for future revisits.Upon returning to a task region, FLAVOR reconstructs the frontier and exploration viewpoint from observations, while the topological route is instantiated and validated against the current local map before execution. Dense-map capacity remains fixed as exploration expands, with exploration-dependent memory growth confined to compact observation and task context.FLAVOR achieves 31.98--41.88% shorter exploration time and a 94--99% smaller map memory footprint than the best baseline that completes all three simulated scenes. In real-world deployments, it reduces this footprint by 92.67% in the forest and 62.42% at the construction site relative to an explicit point-cloud ikd-tree while preserving long-term task continuity under constrained onboard resources.
Figures & tables
Fig. 1. : Overview of the Flavor exploration framework. Local geometry and retained mission context are coupled through cascaded frontier updates, preserving exploration continuity without a persistent global point cloud. Selected viewpoints are ordered into an exploration tour. The planner constructs a corridor along the validated local route and optimizes a trajectory within it.
Fig. 2. : Overall workflow of Flavor . The hierarchical spatio-temporal map representation maintains local geometry, observation context, and task context for exploration planning. Local route validation and trajectory checking precede execution. Section numbers refer to the corresponding method descriptions.
Symbol
Definition
Primary use
FW
Fixed localization frame
Persistent state, graph routes, and trajectories
FBt
Moving UAV body frame
Vehicle state and dynamics
FLt
Moving LiDAR frame
Field of view and visibility rays
MtL
World-aligned sliding grid
Occupancy, distance, and frontier queries
Table 1: Coordinate frames and map convention.
Fig. 3. : Sliding local map and frontier extraction in two dimensions. The overlapping map region is retained, and leaving storage is reused for entering cells. Yellow cells are raw frontiers; white, light gray, and dark gray indicate free, unknown, and occupied cells.
Fig. 4. : LiDAR-keyframe visibility and frontier filtering. The inset shows one sector’s voxel-step encoding of free space (blue), an occupied surface (green), and a guard band (yellow). Historical visibility rejects repeated frontiers (gray crosses), leaving unexplained candidates (orange outlines).
Fig. 5. : Odometry anchors along a continuous flight, with current odometry vertex o(ti) . (a) Inserting a2 stores the recorded route from a1 . (b) A later region update connects a2 to skeleton vertex s , preserving existing region–anchor associations. (c) The current odometry vertex gains a validated shortcut to a1 while retaining its recorded connection to a2 . Dashed boxes denote the local map, and arrows indicate flight direction. Anchor-related edges are bidirectional. Only relevant connections are shown.
Fig. 6. : Historical-task retrieval. (a) Regional skeleton updates provide a route to task anchor a1 . (b) The UAV follows the route as its local map shifts. (c) Re-observation separates unresolved frontier cells (green) from resolved cells (crosses), and regenerated viewpoint v is evaluated for reachability.
Fig. 7. : Simulation scenes: (a) Cave, (b) Forest, and (c) Lava Tube. Dashed boxes and matching numbers locate the rendered views or on-site photographs.
Scene
Local-map window (m)
Map resolution (m)
Visibility resolution/range
Cave
40×40×36
0.4
2∘ , 0.1–16 m
Forest
36×36×10
0.4
2∘ , 0.1–18 m
Lava Tube
48×48×16
0.4
2∘ , 0.1–19 m
Table 2: Simulation configuration of Flavor .
Fig. 8. : Exploration trajectories, coverage progress, and the memory footprint of the map representation in the point-cloud scenes simulated with MARSIM. The columns show Cave, Forest, and Lava Tube from left to right. The top row shows the trajectory of Flavor colored by elapsed exploration time, the middle row shows normalized surface coverage, and the bottom row shows the memory footprint of each planner’s map representation on a logarithmic vertical scale.
Scene
Method
Completion Time (s)
Avg. Vel. (m/s)
Coverage Reached
Cave
TARE
–
–
No
ERRT
–
1.08
No
FALCON
–
–
No
EPIC
1242.50
2.80
Yes
Flavor
845.20
3.26
Yes
Forest
TARE
–
2.06
No
Table 3: Mission-level exploration performance in the point-cloud scenes simulated with MARSIM. Bold indicates the best completion result in each scene.
Scene
EPIC (MB)
Flavor (MB)
Reduction
Cave
1510.69
41.93
97.22%
Forest
1341.21
13.48
99.00%
Lava Tube
242.89
14.56
94.00%
Table 4: Memory footprint of the map representation. Reduction is relative to EPIC, and bold indicates the result of Flavor .
Fig. 9. : Planning latency of EPIC and Flavor in Cave, Forest, and Lava Tube. Boxes show the interquartile range, and horizontal lines mark the medians. The vertical axis is logarithmic.
Scene
EPIC (ms)
Flavor (ms)
Reduction
Cave
85.76±16.52
47.94±11.73
44.10%
Forest
30.36±8.23
21.43±11.59
29.42%
Lava Tube
64.61±13.24
59.50±17.64
7.92%
Table 5: Mean ± standard deviation of planning latency. Bold indicates the lower mean latency.
Fig. 10. : Real-world experiment sites: construction site (top) and forest (bottom). Satellite views on the left show the geographic coordinates and experiment areas marked by yellow dashed boxes. Photographs on the right show the conditions at each site.
Fig. 11. : UAV platform used in the real-world deployments, showing the Intel N305 edge computer, Livox MID360 LiDAR, NxtPX4v2 flight controller, and STM32-based communication module.
Fig. 12. : Offline visualization of the real-world forest (top) and construction-site (bottom) deployments. The left panels overlay the executed UAV trajectory (cyan) and the top-down exploration bounding box (orange) on the reconstructed point-cloud map. Dashed boxes mark selected trajectory locations, and the snapshots on the right show the corresponding third-person and onboard first-person views. The global point clouds are shown only for visualization and were not used by the online planner.
Scene
ikd-tree (MB)
Flavor (MB)
Reduction
Forest
50.64
3.71
92.67%
Construction Site
19.80
7.44
62.42%
Table 6: Memory footprint of the map representation in real-world deployment. The measured components of Flavor are the historical visibility representation and local ROG buffers. Reduction is relative to the explicit point-cloud ikd-tree, and bold indicates the result of Flavor .
Classic exploration methods often rely on dense occupancy maps or high-resolution point clouds for frontier detection and path planning, resulting in substantial memory consumption and computational overhead. Moreover, micro UAVs under size, weight, and power (SWaP) constraints are not practical to be equipped with sensors like LiDAR to obtain accurate environmental geometric measurements. This paper presents a lightweight autonomous exploration system that leverages omnidirectional vision and sparse topological map guidance. Specifically, we utilize a multi-fisheye camera setup to achieve omnidirectional Field of View (FoV) and perform depth estimation. To address the limited depth estimation accuracy, frontiers are represented as potential unexplored regions characterized by topological nodes instead of explicit boundaries, enabling efficient identification of frontier regions without maintaining occupancy grids or global point clouds. Unlike classic dense representations, our approach abstracts the environment using a sparse topological map composed of key nodes and their descriptors, reducing memory consumption and computational demands. Global path planning is performed directly on the sparse graph. The proposed method is validated in both simulation and on a palm-sized vision-based UAV with an 11 cm wheelbase and a 400 g weight in real-world experiments, demonstrating that our method can achieve efficient exploration with extremely low computational consumption.
Zirui Wang, Xinjia Luo, Haotian Sun +3
Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen, 518055, China
Autonomous 3D exploration with UAVs requires efficient target selection under limited flight time and onboard computation. In conventional Euclidean-distance-based frontier pipelines, geometric distance may poorly represent the actual collision-free travel cost in cluttered environments, leading to inefficient frontier choices. This paper presents a lightweight path-aware frontier re-ranking method for LiDAR-based UAV exploration. The method first retains a small top-N set of promising frontier candidates using a lightweight utility function and then applies time-bounded RRT* planning to estimate their feasible path costs for re-ranking. The path generated for the selected frontier is further reused for navigation to reduce redundant global planning. Experiments in three warehouse-like environments show that the proposed method consistently produces shorter exploration trajectories than conventional frontier baselines, with traveled-distance reductions of up to 14.1%, while reducing exploration time by up to 9.5% in two environments and remaining competitive in the third. The method maintains map detail comparable to the conventional OctoMap-based baseline, demonstrating an effective trade-off between exploration efficiency and mapping quality.
Duc Manh Do, Thanh Nguyen Canh, Ngoc Minh Do +2
Vietnam National University, University of Engineering and Technology, Hanoi, Viet Nam. · College of Engineering and Computer Science, VinUniversity, Hanoi 100000, Viet Nam. · School of Information Science, Japan Advanced Institute of Science and Technology, Nomi 923-1292, Japan. +1
Autonomous exploration in unknown environments is key for mobile robots, helping them perceive, map, and make decisions in complex areas. However, current methods often rely on frequent global optimization, suffering from high computational latency and trajectory oscillation, especially on resource-constrained edge devices. To address these limitations, we propose SCOPE, a novel framework that incrementally constructs a real-time skeletal graph and introduces Implicit Unknown Region Analysis for efficient spatial reasoning. The planning layer adopts a hierarchical on-demand strategy: the Proximal Planner generates smooth, high-frequency local trajectories, while the Region-Sequence Planner is activated only when necessary to optimize global visitation order. Comparative evaluations in simulation demonstrate that SCOPE achieves competitive exploration performance comparable to state-of-the-art global planners, while reducing computational cost by an average of 86.9%. Real-world experiments further validate the system's robustness and low latency in practical scenarios.
Kai Li, Shengtao Zheng, Linkun Xiu +4
Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China · School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639956