Enhancing Multi-Robot Exploration Using Probabilistic Frontier Prioritization with Dirichlet Process Gaussian Mixtures

John Lewis Devassy; Meysam Basiri; Mário A. T. Figueiredo; Pedro U. Lima

cs.ROApr 3, 2026

Enhancing Multi-Robot Exploration Using Probabilistic Frontier Prioritization with Dirichlet Process Gaussian Mixtures

Authors: John Lewis Devassy, Meysam Basiri, Mário A. T. Figueiredo, Pedro U. Lima

Organizations: Institute for Systems and Robotics / LARSyS and Instituto Superior Técnico, Universidade de Lisboa, Lisbon 1049-001, Portugal · Instituto de Telecomunicações and Instituto Superior Técnico, Universidade de Lisboa, Lisbon 1049-001, Portugal

Abstract

Multi-agent autonomous exploration is essential for applications such as environmental monitoring, search and rescue, and industrial-scale surveillance. However, effective coordination under communication constraints remains a significant challenge. Frontier exploration algorithms analyze the boundary between the known and unknown regions to determine the next-best view that maximizes exploratory gain. This article proposes an enhancement to existing frontier-based exploration algorithms by introducing a probabilistic approach to frontier prioritization. By leveraging Dirichlet process Gaussian mixture model (DP-GMM) and a probabilistic formulation of information gain, the method improves the quality of frontier prioritization. The proposed enhancement, integrated into two state-of-the-art multi-agent exploration algorithms, consistently improves performance across environments of varying clutter, communication constraints, and team sizes. Simulations showcase an average gain of $10\%$ and $14\%$ for the two algorithms across all combinations. Successful deployment in real-world experiments with a dual-drone system further corroborates these findings.

Figures & tables

Panel of Figure 1 — Fig. 1 : (a) The white and black spaces represent the known and unknown, respectively. The continuous green boundary represents the frontier, and the green dots represent the discretized frontiers. (b) We can see that because of hard clustering, frontier 1 will be considered separate from frontier 2 and frontier 3 , which will lead to separate allocation after frontier processing. Similarly, despite frontier 4 being closer to drone C than drone B, the K-Means clustered it separately from frontier 5 . (c) The responsibility of green, blue, and red Gaussians in a given frontier is represented by the color saturation of the green, blue, and red respectively.

Fig. 2 : The proposed frontier prioritization module depicted as viewpoint processor processes the incoming frontiers and shares the output with a frontier-based exploration planner.

Panel of Figure 3 — Fig. 3 : An instance from a real-world experiment with $2$ UAVs in an Octomap. The UAV paths are shown in red and green. (a) The merged viewpoints, represented as white blobs, from Sec II-B1 (b) Soft cluster assignments from Sec II-B2 , with color-coded probabilities visualized via smooth transitions between red and green to reflect each viewpoint’s responsibility. (c) The prioritized viewpoint inferred from Sec II-B4 is shown in red and green cubes.

$n_{r}$	Forest	SPARSE FOREST (0.1 TREES / m 2 )		MID-DENSITY FOREST (0.15 TREES / m 2 )		DENSE FOREST (0.2 TREES / m 2 )		MULTI-DENSITY FOREST
	Variant	Standard	+FP	Standard	+FP	Standard	+FP	Standard	+FP
1	FAME	$724.63\pm\underline{25.45}$	$\textbf{687.56}\pm 38.47$	$817.52\pm\underline{26.89}$	$\textbf{753.24}\pm 60.47$	$829.10\pm 70.29$	$\textbf{776.13}\pm\underline{39.44}$	$903.76\pm\underline{18.30}$	$\textbf{845.39}\pm 58.37$
	FroShe	$436.61\pm 3.20$	$\textbf{432.05}\pm\underline{2.88}$	$462.57\pm 3.59$	$\textbf{454.71}\pm\underline{3.25}$	$509.55\pm 2.39$	$\textbf{501.11}\pm\underline{2.18}$	$544.54\pm 2.45$	$\textbf{536.72}\pm\underline{2.30}$
	Racer	$589.07\pm 21.26$	-	$648.46\pm 19.60$	-	$667.14\pm 18.35$	-	$761.59\pm 69.18$	-
3	FAME	$252.33\pm 15.80$	$\textbf{220.59}\pm\underline{13.54}$	$274.13\pm 13.60$	$\textbf{251.84}\pm\underline{10.74}$	$335.82\pm 37.13$	$\textbf{300.85}\pm\underline{12.40}$	$389.63\pm\underline{17.48}$	$\textbf{345.67}\pm 25.52$
	FroShe	$154.79\pm 2.41$	$\textbf{142.21}\pm\underline{2.17}$	$172.17\pm 2.52$	$\textbf{164.04}\pm\underline{2.23}$	$226.01\pm 2.96$	$\textbf{224.98}\pm\underline{2.57}$	$243.02\pm 2.34$	$\textbf{242.44}\pm\underline{2.12}$

TABLE I : Mean successful exploration times for FAME, RACER, and FroShe across varying UAV team sizes (

n_{r}

) and forest densities. The Standard and +FP columns represent the default and FP-enabled frameworks, respectively. The lowest mean value is bolded , and the lowest standard deviation is underlined .

Fig. 5 : Exploration time analysis across a dense forest environment with varying communication range and UAV count.

Panel of Figure 7 — Fig. 6 : (a) Two Tarot 650 UAVs, each equipped with Mid-360 Livox and ASUS NUC Pro. (b) A minimal feature $18m\times 36m\times 3m$ volume test environment. (c) The reconstructed point cloud and the path taken across $3$ trials.

Dirichlet Process Mixture Models

See all 8 papers

Decentralized Multi-Robot Exploration with Probabilistic Peer Intent and Multi-hop Plan Propagation
Sep 19, 2026Saurbh Singh Jamwal, Nived Chebrolu, Shivaram KalyanakrishnanMulti-Robot Exploration Multi-Robot Coordination
Department of Computer Science and Engineering Indian Institute of Technology Bombay
Mobility-Budget-Constrained Submodular Maximization for Distributed Semantic Multi-Robot Exploration
Oct 13, 2026Changxuan Yao, Sandilya Sai Garimella, Lu Gan +1Multi-Robot Exploration Informative Path Planning
The authors are with the Institute of Robotics and Intelligent Machines, Georgia Institute of Technology, Atlanta, GA 30332, USA.
Communication-Constrained Multi-Robot Exploration With Adaptive Communication Windows
Sep 11, 2026Ben Rossano, Jaein Lim, Jonathan P. HowMulti-Robot Exploration Multi-Agent Communication
Aerospace Controls Lab, Massachusetts Institute of Technology, Cambridge, MA, USA · Draper Scholar with the Charles Stark Draper Laboratory, Cambridge, MA · Charles Stark Draper Laboratory, Cambridge, MA, USA

Sep 19, 2026J/K move · Enter open · S save

Decentralized Multi-Robot Exploration with Probabilistic Peer Intent and Multi-hop Plan Propagation

Saurbh Singh Jamwal, Nived Chebrolu, Shivaram Kalyanakrishnan

Department of Computer Science and Engineering Indian Institute of Technology Bombay

Multi-Robot Exploration Multi-Robot Coordination Multi-Agent Systems Multi-Agent Communication Monte Carlo Tree Search

Open paper AskPDF arXiv

Efficient coordination under limited communication remains a key challenge in decentralized multi-robot exploration. While centralized approaches benefit from global information sharing, they are often impractical in large-scale or communication-constrained environments. Existing Monte Carlo Tree Search (MCTS)-based approaches, such as Decentralized Monte Carlo Exploration (DMCE), enable decentralized planning by taking peer intent into account. This peer intent is obtained by communicating sequences of planned waypoints with robots within direct communication range. In this work, we extend this idea by introducing Probabilistic Peer Intent (PPI), which converts peer trajectories into a continuous spatial representation of predicted intent and incorporates it into local MCTS action evaluation. We additionally study the effects of sharing peer intent beyond direct communication range by propagating plans over multiple hops. Experiments across multiple simulated environments and team sizes show that PPI and Multi-hop propagation can each improve decentralized exploration, with their relative benefits depending on environment structure and team size. We also demonstrate the real-world deployment of our method on three robots operating in different environment types.

Abstract

Dirichlet Process Mixture Models

Explore similar work

Decentralized Multi-Robot Exploration with Probabilistic Peer Intent and Multi-hop Plan Propagation

Mobility-Budget-Constrained Submodular Maximization for Distributed Semantic Multi-Robot Exploration

Communication-Constrained Multi-Robot Exploration With Adaptive Communication Windows