Centralized Multi-UAV Exploration and 3D Reconstruction Using Single-UAV Planners

João Félix Mendes; Meysam Basiri; Rodrigo Ventura

cs.ROSep 30, 2026

Centralized Multi-UAV Exploration and 3D Reconstruction Using Single-UAV Planners

Authors: João Félix Mendes, Meysam Basiri, Rodrigo Ventura

Organizations: Institute for Systems and Robotics, Department of Electrical and Computer Engineering, Instituto Superior Técnico, Lisboa, Portugal

Abstract

Extending single Unmanned Aerial Vehicles (UAVs) exploration methods to multi-UAV teams can improve coverage speed and robustness, but introduces challenges such as consistent mapping, safe navigation, and deployment strategy. In this work, we present a centralized multi-UAV exploration framework that enables the use of existing single-UAV sampling-based planners in a multi-UAV setting. The proposed architecture allows multiple UAVs to collaboratively explore unknown environments using a shared global Truncated Signed Distance Field (TSDF) map and centralized planning. Building on the voxblox library, we adapt its mapping pipeline to support real-time fusion of depth measurements from multiple UAVs into a common TSDF representation. In addition, inter-UAV collision avoidance and robot self-filtering mechanisms are integrated into the system to ensure safe navigation and prevent reconstruction of other UAVs as static obstacles. The framework is evaluated in simulation using four sampling-based exploration planners - RH-NBVP, KRH-NBVP, AEP, and KAEP - whose core sampling logic is preserved, with only system-level adaptations for multi-UAV operation. Experiments are conducted across multiple environments and under two deployment configurations: Joint Start (JS), where UAVs are initialized in close proximity, and Separated Start (SS), where UAVs are initialized in distinct locations. Results show that SS deployments consistently achieve faster exploration and improved coverage across all planners, highlighting the importance of the deployment strategy in multi-UAV exploration performance.

Figures & tables

Panel of Figure 1 — Fig. 1: Map reconstructions generated by a three-UAV team using the KAEP planner in the Police Station environment at $t=$ 160\text{,}\mathrm{s}$$ . (a) In JS, UAVs begin in close proximity, resulting in more overlapping coverage. (b) In SS, UAVs are spatially distributed, leading to broader coverage with reduced overlap.

Fig. 2: Overview of the proposed centralized multi-UAV exploration framework for two UAVs. Color-coded blocks denote hardware boundaries (blue and green for UAVs and yellow for the central server). Each UAV performs state estimation, sensing, and local pointcloud filtering via peer-to-peer pose sharing, integrating measurements into a local sub-map. These local updates are periodically sent to the server and fused into a unified voxblox map. The planner nodes then asynchronously compute new paths upon request, utilizing the unified map and a shared active path cache for inter-UAV collision avoidance. The collision-free paths are then transmitted back to the UAVs.

Panel of Figure 3 — Fig. 3: Effect of pointcloud filtering in a two-UAV mapping scenario. (a) Experimental setup. (b,c) TSDF visualization from the left and right UAV perspectives using unfiltered pointclouds, where UAVs are reconstructed as obstacles. The highlighted orange regions indicate areas of mutual mapping. (d,e) TSDF visualization with pointcloud filtering enabled, where inter-robot reconstruction is removed, and only the static environment remains.

Panel of Figure 4 — Fig. 4: UAV initialization layouts in maze (left), police station (middle), and school (right) environments for a team of three UAVs. The green boxes represent the SS configurations, and the orange boxes represent the JS configurations.

Parameter	Value	Parameter	Value
$v_{xy_{\max}},v_{z_{\max}}$	$1\text{\,}\mathrm{m}\text{\,}{\mathrm{s}}^{-1}$	FoV	[ $87^{\circ}$ , $58^{\circ}$ ]
$\dot{\psi}_{\max}$	$2\text{\,}\mathrm{rad}\text{\,}{\mathrm{s}}^{-1}$	Camera range	$5\text{\,}\mathrm{m}$
$a_{xy_{\max}},a_{z_{\max}}$	$1\text{\,}\mathrm{m}\text{\,}{\mathrm{s}}^{-2}$	Camera pitch	$10\text{\,}\mathrm{\SIUnitSymbolDegree}$
$\ddot{\psi}_{\max}$	$2\text{\,}\mathrm{rad}\text{\,}{\mathrm{s}}^{-2}$	$\mathbf{v}_{size}$	$0.2\text{\,}\mathrm{m}$
$d_{replan}$	$0.8\text{\,}\mathrm{m}$	$d_{collision}$	$1.5\text{\,}\mathrm{m}$

TABLE I: Key Experimental Parameters (unchanged from [ 3 ] ).

Metric	Police Station	School
Local TSDF Integration (ms)	113.0 $\pm$ 15.4	87.0 $\pm$ 13.1
Local ESDF Integration (ms)	9.2 $\pm$ 8.2	10.0 $\pm$ 4.7
Central TSDF Merge (ms)	7.9 $\pm$ 2.6	8.6 $\pm$ 8.5
Central ESDF Merge (ms)	8.9 $\pm$ 1.9	10.1 $\pm$ 9.2
Central Map RMSE (m)	0.24 $\pm$ 0.20	0.19 $\pm$ 0.15

TABLE II: Computational Performance and Map Accuracy for KAEP in three-UAV Team.

Panel of Figure 7 — Fig. 5: Exploration progress results for maze, police station, and school environments using a three-UAV team under SS (top) and JS (bottom) deployment strategies.

Environment	Cfg	Planner	E25%	E50%	E95%	FE	PL	AV
			Time	Time	Time	[ $\text{\,}\mathrm{\%}$ ]	[ $\text{\,}\mathrm{m}$ ]	[ $\text{\,}\mathrm{m}\text{\,}{\mathrm{s}}^{-1}$ ]
Maze (Time in [ $\text{\,}\mathrm{s}$ ])	SS	RH-NBVP	$14.32\pm 0.88$	$41.67\pm 8.48$	$105.11\pm 7.72$	$\boldsymbol{100.0\pm 0.0}$	$\boldsymbol{236.75\pm 3.52}$	$0.27\pm 0.01$
		AEP	$10.51\pm 0.62$	$28.09\pm 4.68$	$83.02\pm 5.71$	$\boldsymbol{100.0\pm 0.0}$	$275.58\pm 29.90$	$0.29\pm 0.03$
		KRH-NBVP	$\boldsymbol{10.42\pm 0.08}$	$26.51\pm 3.19$	$77.04\pm 5.54$	$\boldsymbol{100.0\pm 0.0}$	$306.48\pm 1.99$	$0.35\pm 0.01$
		KAEP	$11.15\pm 0.04$	$\boldsymbol{24.26\pm 0.50}$	$\boldsymbol{59.04\pm 2.81}$	$\boldsymbol{100.0\pm 0.0}$	$348.92\pm 29.46$	$\boldsymbol{0.42\pm 0.02}$
Police Station (Time in [ $\text{\,}\mathrm{min}$ ])	JS	RH-NBVP	$1.35\pm 0.25$	$2.53\pm 0.55$	–	$87.24\pm 16.97$	$\boldsymbol{433.93\pm 92.69}$	$0.45\pm 0.10$

TABLE III: Combined Multi-Robot Experiment Results. Note: Time units vary by environment. The dashes (–) indicate that E95% exploration was not reached before the simulation time limit.

Explore similar work

Quality-Adaptive Multi-UAV 3D Reconstruction with Sparse Workload Redistribution
Jul 27, 2026Benjamin Sportich, Kenza Boubakri, Olivier Simonin +1Multi-Uav 3D Reconstruction
Inria, INSA Lyon, CITI, 69621 Villeurbanne, France.
A Distributed Multi-UGV Exploration Framework With Loop-Aware Planning and Descriptor-Aided Localization in Resource-Limited Environments
Jun 9, 2026Zhiwei Li, Haiou Liu, Xijun Zhao +3Multi-Uav Exploration
School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China · China North Artificial Intelligence & Innovation Research Institute, Collective Intelligence & Collaboration Laboratory (CIC), Beijing 100081, China
Explore From Sketch: Accelerating UAV Exploration in Large-scale Environments with Prior Maps
Jun 10, 2026Tiancheng Lai, Yuman Gao, Xiangyu Li +8Path Planning Point Cloud Registration
Institute of Cyber-Systems and Control, College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China · Huzhou Institute, Zhejiang University, and Huzhou Key Laboratory of Autonomous System, Huzhou 313000, China. · Differential Robotics Technology Company, Hangzhou 311100, China. +1

Jul 27, 2026, cs.ROJ/K move · Enter open · S save

Quality-Adaptive Multi-UAV 3D Reconstruction with Sparse Workload Redistribution

Benjamin Sportich, Kenza Boubakri, Olivier Simonin, Alessandro Renzaglia

Inria, INSA Lyon, CITI, 69621 Villeurbanne, France.

Multi-Uav 3D Reconstruction 3D Mapping Path Planning Global Coordination

Open paper AskPDF arXiv

3D reconstruction of unknown environments is a key application in robotics but is severely limited by the computational and energy capabilities of current aerial platforms. Deploying multiple UAVs and providing efficient and scalable path planning strategies are common approaches, but effective online coordination among UAVs remains a significant challenge. To address this problem, we propose a quality-adaptive decentralized decision-making strategy to build a 3D map with user-defined degrees of fidelity. The approach integrates a quality-oriented criterion based on TSDF confidence into view generation and information gain estimation to produce viewpoints consistent with the desired fidelity target. Additionally, we employ two levels of coordination: a penalty factor in the viewpoint evaluation to encourage local dispersion among the UAVs and a global imbalance correction mechanism. The latter, based on regularized clustering and optimal task assignment, is only triggered when an unbalanced configuration relative to high-information regions is detected. Simulation results demonstrate that the proposed method improves path efficiency compared to state-of-the-art multi-UAV exploration approaches, while also achieving higher-fidelity reconstructions in terms of coverage and accuracy. We make our code publicly available to the community.

Abstract

Explore similar work

Quality-Adaptive Multi-UAV 3D Reconstruction with Sparse Workload Redistribution

A Distributed Multi-UGV Exploration Framework With Loop-Aware Planning and Descriptor-Aided Localization in Resource-Limited Environments

Explore From Sketch: Accelerating UAV Exploration in Large-scale Environments with Prior Maps