Data-driven Spatial Classification using Multi-Arm Bandits for Monitoring with Energy-Constrained Mobile Robots

Abraham P. Vinod; Siddharth Nayak; Stefano Di Cairano; Xiaoshan Lin

arxiv: 2501.08222 · v2 · pith:FEWYYMJJnew · submitted 2025-01-14 · 💻 cs.RO

Data-driven Spatial Classification using Multi-Arm Bandits for Monitoring with Energy-Constrained Mobile Robots

Xiaoshan Lin , Siddharth Nayak , Stefano Di Cairano , Abraham P. Vinod This is my paper

Pith reviewed 2026-05-25 08:44 UTC · model grok-4.3

classification 💻 cs.RO

keywords mobileteamapproachclassificationdataregionsrobotscollected

0 comments

The pith

Bi-level multi-armed bandit plus integer programming planner for energy-constrained robot teams performing online spatial classification with theoretical guarantees.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Mobile robots must explore an area and label parts as worth further attention or not, but they have limited battery and their sensors are noisy. The method splits the job into two layers. The top layer treats possible regions like slot-machine arms in a bandit problem and uses incoming data to decide which region to send robots toward next. The bottom layer solves an integer program that assigns actual paths to the robots and any charging stations so they never collide and never run out of energy before returning. The authors state that this combination gives performance bounds that hold even if the system is stopped early and that the total time to finish the classification task can be bounded. They report that computer simulations and tests with physical robots match the expected behavior.

Core claim

We propose a bi-level approach, where a high-level planner leverages a multi-armed bandit framework to determine the potential regions of interest for the drones to visit next based on the data collected online. Then, a low-level path planner based on integer programming coordinates the paths for the team to visit the determined regions subject to the physical constraints. We characterize several theoretical properties of the proposed approach, including anytime guarantees and task completion time.

Load-bearing premise

That the multi-armed bandit layer can reliably identify regions of interest from noisy online data and that the integer program will produce feasible collision-free, energy-respecting paths in time for the high-level decisions to remain useful.

Figures

Figures reproduced from arXiv: 2501.08222 by Abraham P. Vinod, Siddharth Nayak, Stefano Di Cairano, Xiaoshan Lin.

**Figure 2.** Figure 2: (Top) Four types of potential collisions in the paths [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Snapshot of the hardware experiment with two Turtle [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Classification progress of Algorithm 1 in the hardware [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Classification progress of Algorithm 1 when varying [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: Epochs needed to classify all interesting cells when [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

read the original abstract

We consider the spatial classification problem for monitoring using data collected by a coordinated team of mobile robots. Such classification problems arise in several applications including search-and-rescue and precision agriculture. Specifically, we want to classify the regions of a search environment into interesting and uninteresting as quickly as possible using a team of mobile sensors and mobile charging stations. We develop a data-driven strategy that accommodates the noise in sensed data and the limited energy capacity of the sensors, and generates collision-free motion plans for the team. We propose a bi-level approach, where a high-level planner leverages a multi-armed bandit framework to determine the potential regions of interest for the drones to visit next based on the data collected online. Then, a low-level path planner based on integer programming coordinates the paths for the team to visit the determined regions subject to the physical constraints. We characterize several theoretical properties of the proposed approach, including anytime guarantees and task completion time. We show the efficacy of our approach in simulation, and further validate these observations in physical experiments using mobile robots.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The bi-level MAB-plus-IP planner for energy-constrained robot spatial classification is a straightforward engineering combination that adds physical constraints to an online monitoring loop, but the abstract gives almost no detail on how the claimed theoretical properties are derived or whether the low-level solver stays fast enough.

read the letter

The paper's main contribution is a bi-level scheme for teams of mobile robots and chargers that classify spatial regions. The high level runs a multi-armed bandit on incoming noisy measurements to pick the next regions worth visiting. The low level solves an integer program to produce collision-free paths that also respect energy limits. They test the idea in simulation and on physical robots, and they state that the approach comes with anytime guarantees and bounds on task completion time. That integration for this exact setting with mobile chargers is the concrete new piece; prior work on bandits or on constrained multi-robot planning exists separately, but the joint online loop is not standard in the references given in the abstract. The experiments on real hardware are a plus for a robotics paper. The soft spots sit in the missing mechanics. The abstract asserts theoretical properties without showing assumptions on the noise model, without sketching the proof strategy, and without addressing how long the integer program takes to return a plan. If the solver latency grows with the number of robots or regions, the bandit selections can go stale before the paths are even computed, which would undermine the online claim. The stress-test note on solve-time blowup is therefore worth checking against the full text; nothing in the abstract rules it out. This paper is aimed at people who work on constrained multi-robot planning or bandit-driven monitoring. A reader who needs an applied example of the bi-level pattern with energy and collision constraints would get something usable from the experiments. It is coherent enough on its own terms to deserve a serious referee who can verify the derivations and the timing behavior.

Referee Report

2 major / 0 minor

Summary. The paper proposes a bi-level framework for spatial classification in robotic monitoring tasks. A high-level multi-armed bandit selects regions of interest from noisy online sensor data collected by energy-constrained mobile robots, while a low-level integer program generates collision-free paths respecting battery limits and other physical constraints. The approach is claimed to admit anytime guarantees and bounded task completion time, with supporting evidence from simulations and physical experiments on small teams.

Significance. If the claimed theoretical properties can be rigorously established with explicit assumptions and if the bi-level timing is shown to be feasible, the work would offer a practical data-driven method for coordinated monitoring under energy limits. The combination of online bandits with constrained multi-robot path planning is a reasonable construction for the stated applications, but the absence of derivation details and timing analysis currently limits its contribution.

major comments (2)

[Abstract] Abstract: the assertion of 'anytime guarantees and task completion time' is presented without any derivation, explicit noise model assumptions, or statement of how the properties are obtained from the bi-level construction. This is load-bearing for the central online data-driven claim.
[Abstract] Abstract: the bi-level claim requires that low-level IP solve times remain short enough for high-level MAB region selections (made from noisy data) to stay relevant; no bound, assumption, or empirical timing profile is supplied to address this, which directly engages the energy-constrained online loop.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful comments, which highlight important aspects of our presentation that require clarification. We address each major comment below and will revise the manuscript accordingly to strengthen the exposition of our theoretical contributions and practical feasibility.

read point-by-point responses

Referee: [Abstract] Abstract: the assertion of 'anytime guarantees and task completion time' is presented without any derivation, explicit noise model assumptions, or statement of how the properties are obtained from the bi-level construction. This is load-bearing for the central online data-driven claim.

Authors: The manuscript does provide derivations of these properties in Section IV (Theoretical Analysis), where we model the sensor noise as additive Gaussian noise and leverage the regret bounds of the multi-armed bandit combined with the feasibility guarantees of the integer program under energy constraints to establish anytime performance and bounded completion time. However, we agree that the abstract does not sufficiently link these to the bi-level construction or state the assumptions explicitly. We will revise the abstract to include a concise statement of the noise model and a high-level indication of how the guarantees follow from the framework. This revision will be made. revision: yes
Referee: [Abstract] Abstract: the bi-level claim requires that low-level IP solve times remain short enough for high-level MAB region selections (made from noisy data) to stay relevant; no bound, assumption, or empirical timing profile is supplied to address this, which directly engages the energy-constrained online loop.

Authors: We acknowledge the importance of demonstrating that the low-level IP solves are sufficiently fast to maintain relevance in the online setting. The current manuscript includes runtime results in the simulation section but does not provide a dedicated timing profile or explicit assumptions on solver performance in relation to the MAB update cycle. We will add an empirical timing analysis subsection, including average and worst-case solve times from both simulations and physical experiments, along with a discussion of the assumptions (e.g., use of efficient MIP solvers and problem sizes for small teams). If a theoretical bound on solve time is not derivable without further restrictions, we will clearly state the reliance on empirical evidence. This will be incorporated in the revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity in bi-level MAB-IP framework

full rationale

The paper presents a novel bi-level construction: high-level multi-armed bandit for online region selection from noisy data, followed by low-level integer programming for collision-free, energy-constrained paths. No equations, parameters, or claims in the abstract reduce by construction to fitted inputs or self-definitions. Theoretical properties (anytime guarantees, task completion time) are stated as characterizations of the proposed method rather than reductions to prior fitted values. No self-citation chains or ansatzes are invoked as load-bearing. The approach is evaluated externally via simulation and physical robot experiments, rendering the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on standard domain assumptions about bandit algorithms and integer programming solvers; no free parameters, invented entities, or ad-hoc axioms are visible in the abstract.

axioms (2)

domain assumption Multi-armed bandit algorithms can select regions of interest from noisy sequential observations
Invoked by the high-level planner
domain assumption Integer linear programs can encode and solve collision-free, energy-constrained multi-robot path planning
Invoked by the low-level planner

pith-pipeline@v0.9.0 · 5725 in / 1316 out tokens · 38081 ms · 2026-05-25T08:44:44.059480+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages

[1]

Bandit-based multi-agent search under noisy observations,

P. Thaker, S. Di Cairano, and A. Vinod, “Bandit-based multi-agent search under noisy observations,” IFAC-PapersOnLine, pp. 2780–2785, 2023

work page 2023
[2]

Multi-agent systems for search and rescue applications,

D. Drew, “Multi-agent systems for search and rescue applications,” Curr. Rob. Rep., vol. 2, pp. 189–200, 2021

work page 2021
[3]

Collaborative multi- robot search and rescue: Planning, coordination, perception, and active vision,

J. Queralta, J. Taipalmaa, B. Pullinen, V . Sarker, T. Gia, H. Tenhunen, M. Gabbouj, J. Raitoharju, and T. Westerlund, “Collaborative multi- robot search and rescue: Planning, coordination, perception, and active vision,” IEEE Access, vol. 8, pp. 191 617–191 643, 2020

work page 2020
[4]

Robots for environmental monitoring: Significant advancements and applications,

M. Dunbabin and L. Marques, “Robots for environmental monitoring: Significant advancements and applications,” IEEE Rob. & Autom. Mag. , vol. 19, no. 1, pp. 24–39, 2012

work page 2012
[5]

Data-driven monitoring with mobile sensors and charging stations using multi-arm bandits and coordinated motion planners,

S. Nayak, M. Greiff, A. Raghunathan, S. D. Cairano, and A. Vinod, “Data-driven monitoring with mobile sensors and charging stations using multi-arm bandits and coordinated motion planners,” in Amer. Ctrl. Conf., 2024, pp. 299–305

work page 2024
[6]

Near-optimal observation selection using submodular functions,

A. Krause and C. Guestrin, “Near-optimal observation selection using submodular functions,” in AAAI, vol. 7, 2007, pp. 1650–1654

work page 2007
[7]

Bullo, J

F. Bullo, J. Cort ´es, and S. Martinez, Distributed control of robotic networks. Princeton Univ. Press, 2009

work page 2009
[8]

Robust adaptive coverage control for robotic sensor networks,

M. Schwager, M. Vitus, S. Powers, D. Rus, and C. Tomlin, “Robust adaptive coverage control for robotic sensor networks,” IEEE Tran. Ctrl. Netw. Syst., vol. 4, no. 3, pp. 462–476, 2015

work page 2015
[9]

Distributed environmental modeling and adaptive sampling for multi-robot sensor coverage,

W. Luo, C. Nam, G. Kantor, and K. Sycara, “Distributed environmental modeling and adaptive sampling for multi-robot sensor coverage,” in Proc. Intl. Conf. Auto. Agents Multi-Agent Syst. , 2019, pp. 1488–1496

work page 2019
[10]

Revisiting active perception,

R. Bajcsy, Y . Aloimonos, and J. Tsotsos, “Revisiting active perception,” A. Robots, vol. 42, no. 2, pp. 177–196, 2018

work page 2018
[11]

DARP: Di- vide areas algorithm for optimal multi-robot coverage path planning,

A. Kapoutsis, S. Chatzichristofis, and E. Kosmatopoulos, “DARP: Di- vide areas algorithm for optimal multi-robot coverage path planning,” J. Intelli. Robotic Syst. , vol. 86, no. 3, pp. 663–680, 2017

work page 2017
[12]

Online planning for multi-robot active perception with self-organising maps,

G. Best, J. Faigl, and R. Fitch, “Online planning for multi-robot active perception with self-organising maps,” A. Robots , vol. 42, no. 4, pp. 715–738, 2018

work page 2018
[13]

Bayesian optimisation for intelligent environmental monitoring,

R. Marchant and F. Ramos, “Bayesian optimisation for intelligent environmental monitoring,” in IEEE Int’l Conf. Intelli. Robots Syst. , 2012, pp. 2242–2249

work page 2012
[14]

Decentralized multi-agent active search for sparse signals,

R. Ghods, A. Banerjee, and J. Schneider, “Decentralized multi-agent active search for sparse signals,” inProc. Conf. Uncertainty Artif. Intelli., vol. 161, 2021, pp. 696–706

work page 2021
[15]

Lattimore and C

T. Lattimore and C. Szepesv ´ari, Bandit algorithms. Cambridge Univ. Press, 2020

work page 2020
[16]

A successive-elimination approach to adaptive robotic source seeking,

E. Rolf, D. Fridovich-Keil, M. Simchowitz, B. Recht, and C. Tomlin, “A successive-elimination approach to adaptive robotic source seeking,” IEEE Tran. Robotics , vol. 37, no. 1, pp. 34–47, 2021

work page 2021
[17]

Multi-robot dynamical source seeking in unknown environments,

B. Du, K. Qian, H. Iqbal, C. Claudel, and D. Sun, “Multi-robot dynamical source seeking in unknown environments,” in Intl Conf. Robotics and Autom. , 2021, pp. 9036–9042

work page 2021
[18]

An optimal algorithm for the thresholding bandit problem,

A. Locatelli, M. Gutzeit, and A. Carpentier, “An optimal algorithm for the thresholding bandit problem,” in Intl. Conf. Machine Learning, 2016, pp. 1690–1698

work page 2016
[19]

Top arm identification in multi-armed bandits with batch arm pulls,

K.-S. Jun, K. Jamieson, R. Nowak, and X. Zhu, “Top arm identification in multi-armed bandits with batch arm pulls,” in Proc. Intl Conf. Artifi. Intelli. Stats., vol. 51, 2016, pp. 139–148

work page 2016
[20]

Finding all ϵ-good arms in stochastic bandits,

B. Mason, L. Jain, A. Tripathy, and R. Nowak, “Finding all ϵ-good arms in stochastic bandits,” Adv. Neural Info. Process. Syst. , 2020

work page 2020
[21]

Mobile recharger path planning and recharge scheduling in a multi-robot environment,

T. Kundu and I. Saha, “Mobile recharger path planning and recharge scheduling in a multi-robot environment,” in IEEE Int’l Conf. Intell. Rob. Syst., 2021, pp. 3635–3642

work page 2021
[22]

Robust planning for persistent surveillance with energy-constrained uavs and mobile charging stations,

X. Lin, Y . Yazıcıo˘glu, and D. Aksaray, “Robust planning for persistent surveillance with energy-constrained uavs and mobile charging stations,” IEEE Rob. Autom. Lett. , vol. 7, no. 2, pp. 4157–4164, 2022

work page 2022
[23]

Algorithms and experiments on routing of unmanned aerial vehicles with mobile recharging stations,

K. Yu, A. K. Budhiraja, S. Buebel, and P. Tokekar, “Algorithms and experiments on routing of unmanned aerial vehicles with mobile recharging stations,” J. Field Rob., vol. 36, no. 3, 2018

work page 2018
[24]

Gurobi Optimizer Reference Manual,

Gurobi Opt., LLC, “Gurobi Optimizer Reference Manual,” https://www. gurobi.com (Last accessed: 2025)

work page 2025
[25]

Distributed multi- robot task assignment and formation control,

N. Michael, M. Zavlanos, V . Kumar, and G. Pappas, “Distributed multi- robot task assignment and formation control,” in Int’l Conf. Rob. Autom., 2008, pp. 128–133

work page 2008
[26]

Burkard, M

R. Burkard, M. Dell’Amico, and S. Martello, Assignment Problems . USA: Society for Industrial and Applied Mathematics, 2009

work page 2009
[27]

Crazyflie 2.1,

Bitcraze, “Crazyflie 2.1,” https://www.bitcraze.io/products/crazyflie-2-1/ (Last accessed: 2025)

work page 2025
[28]

Turtlebot 4,

Clearpath Robotics, “Turtlebot 4,” https://clearpathrobotics.com/ turtlebot-4/ (Last accessed: 2025)

work page 2025
[29]

Crazyswarm2,

W. Honig, K. McGuire et al. , “Crazyswarm2,” https://github.com/ IMRCLab/crazyswarm2 (Last accessed: 2025)

work page 2025
[30]

The OpenCV Library,

G. Bradski, “The OpenCV Library,” Dr. Dobb’s J. Soft. Tools, 2000. APPENDIX A. Proof of Proposition 1 Consider two events — EK(p) = {K(p) \ Sθ−ϵ ̸= ∅} and ER(p) = {R(p) ∩ Sθ+ϵ ̸= ∅}. Specifically, at any epoch p, EK(p) and ER(p) are the undesirable events that a cell in C \ Sθ−ϵ is included in the keep set K(p) and a cell in Sθ+ϵ is included in the reject...

work page 2000

[1] [1]

Bandit-based multi-agent search under noisy observations,

P. Thaker, S. Di Cairano, and A. Vinod, “Bandit-based multi-agent search under noisy observations,” IFAC-PapersOnLine, pp. 2780–2785, 2023

work page 2023

[2] [2]

Multi-agent systems for search and rescue applications,

D. Drew, “Multi-agent systems for search and rescue applications,” Curr. Rob. Rep., vol. 2, pp. 189–200, 2021

work page 2021

[3] [3]

Collaborative multi- robot search and rescue: Planning, coordination, perception, and active vision,

J. Queralta, J. Taipalmaa, B. Pullinen, V . Sarker, T. Gia, H. Tenhunen, M. Gabbouj, J. Raitoharju, and T. Westerlund, “Collaborative multi- robot search and rescue: Planning, coordination, perception, and active vision,” IEEE Access, vol. 8, pp. 191 617–191 643, 2020

work page 2020

[4] [4]

Robots for environmental monitoring: Significant advancements and applications,

M. Dunbabin and L. Marques, “Robots for environmental monitoring: Significant advancements and applications,” IEEE Rob. & Autom. Mag. , vol. 19, no. 1, pp. 24–39, 2012

work page 2012

[5] [5]

Data-driven monitoring with mobile sensors and charging stations using multi-arm bandits and coordinated motion planners,

S. Nayak, M. Greiff, A. Raghunathan, S. D. Cairano, and A. Vinod, “Data-driven monitoring with mobile sensors and charging stations using multi-arm bandits and coordinated motion planners,” in Amer. Ctrl. Conf., 2024, pp. 299–305

work page 2024

[6] [6]

Near-optimal observation selection using submodular functions,

A. Krause and C. Guestrin, “Near-optimal observation selection using submodular functions,” in AAAI, vol. 7, 2007, pp. 1650–1654

work page 2007

[7] [7]

Bullo, J

F. Bullo, J. Cort ´es, and S. Martinez, Distributed control of robotic networks. Princeton Univ. Press, 2009

work page 2009

[8] [8]

Robust adaptive coverage control for robotic sensor networks,

M. Schwager, M. Vitus, S. Powers, D. Rus, and C. Tomlin, “Robust adaptive coverage control for robotic sensor networks,” IEEE Tran. Ctrl. Netw. Syst., vol. 4, no. 3, pp. 462–476, 2015

work page 2015

[9] [9]

Distributed environmental modeling and adaptive sampling for multi-robot sensor coverage,

W. Luo, C. Nam, G. Kantor, and K. Sycara, “Distributed environmental modeling and adaptive sampling for multi-robot sensor coverage,” in Proc. Intl. Conf. Auto. Agents Multi-Agent Syst. , 2019, pp. 1488–1496

work page 2019

[10] [10]

Revisiting active perception,

R. Bajcsy, Y . Aloimonos, and J. Tsotsos, “Revisiting active perception,” A. Robots, vol. 42, no. 2, pp. 177–196, 2018

work page 2018

[11] [11]

DARP: Di- vide areas algorithm for optimal multi-robot coverage path planning,

A. Kapoutsis, S. Chatzichristofis, and E. Kosmatopoulos, “DARP: Di- vide areas algorithm for optimal multi-robot coverage path planning,” J. Intelli. Robotic Syst. , vol. 86, no. 3, pp. 663–680, 2017

work page 2017

[12] [12]

Online planning for multi-robot active perception with self-organising maps,

G. Best, J. Faigl, and R. Fitch, “Online planning for multi-robot active perception with self-organising maps,” A. Robots , vol. 42, no. 4, pp. 715–738, 2018

work page 2018

[13] [13]

Bayesian optimisation for intelligent environmental monitoring,

R. Marchant and F. Ramos, “Bayesian optimisation for intelligent environmental monitoring,” in IEEE Int’l Conf. Intelli. Robots Syst. , 2012, pp. 2242–2249

work page 2012

[14] [14]

Decentralized multi-agent active search for sparse signals,

R. Ghods, A. Banerjee, and J. Schneider, “Decentralized multi-agent active search for sparse signals,” inProc. Conf. Uncertainty Artif. Intelli., vol. 161, 2021, pp. 696–706

work page 2021

[15] [15]

Lattimore and C

T. Lattimore and C. Szepesv ´ari, Bandit algorithms. Cambridge Univ. Press, 2020

work page 2020

[16] [16]

A successive-elimination approach to adaptive robotic source seeking,

E. Rolf, D. Fridovich-Keil, M. Simchowitz, B. Recht, and C. Tomlin, “A successive-elimination approach to adaptive robotic source seeking,” IEEE Tran. Robotics , vol. 37, no. 1, pp. 34–47, 2021

work page 2021

[17] [17]

Multi-robot dynamical source seeking in unknown environments,

B. Du, K. Qian, H. Iqbal, C. Claudel, and D. Sun, “Multi-robot dynamical source seeking in unknown environments,” in Intl Conf. Robotics and Autom. , 2021, pp. 9036–9042

work page 2021

[18] [18]

An optimal algorithm for the thresholding bandit problem,

A. Locatelli, M. Gutzeit, and A. Carpentier, “An optimal algorithm for the thresholding bandit problem,” in Intl. Conf. Machine Learning, 2016, pp. 1690–1698

work page 2016

[19] [19]

Top arm identification in multi-armed bandits with batch arm pulls,

K.-S. Jun, K. Jamieson, R. Nowak, and X. Zhu, “Top arm identification in multi-armed bandits with batch arm pulls,” in Proc. Intl Conf. Artifi. Intelli. Stats., vol. 51, 2016, pp. 139–148

work page 2016

[20] [20]

Finding all ϵ-good arms in stochastic bandits,

B. Mason, L. Jain, A. Tripathy, and R. Nowak, “Finding all ϵ-good arms in stochastic bandits,” Adv. Neural Info. Process. Syst. , 2020

work page 2020

[21] [21]

Mobile recharger path planning and recharge scheduling in a multi-robot environment,

T. Kundu and I. Saha, “Mobile recharger path planning and recharge scheduling in a multi-robot environment,” in IEEE Int’l Conf. Intell. Rob. Syst., 2021, pp. 3635–3642

work page 2021

[22] [22]

Robust planning for persistent surveillance with energy-constrained uavs and mobile charging stations,

X. Lin, Y . Yazıcıo˘glu, and D. Aksaray, “Robust planning for persistent surveillance with energy-constrained uavs and mobile charging stations,” IEEE Rob. Autom. Lett. , vol. 7, no. 2, pp. 4157–4164, 2022

work page 2022

[23] [23]

Algorithms and experiments on routing of unmanned aerial vehicles with mobile recharging stations,

K. Yu, A. K. Budhiraja, S. Buebel, and P. Tokekar, “Algorithms and experiments on routing of unmanned aerial vehicles with mobile recharging stations,” J. Field Rob., vol. 36, no. 3, 2018

work page 2018

[24] [24]

Gurobi Optimizer Reference Manual,

Gurobi Opt., LLC, “Gurobi Optimizer Reference Manual,” https://www. gurobi.com (Last accessed: 2025)

work page 2025

[25] [25]

Distributed multi- robot task assignment and formation control,

N. Michael, M. Zavlanos, V . Kumar, and G. Pappas, “Distributed multi- robot task assignment and formation control,” in Int’l Conf. Rob. Autom., 2008, pp. 128–133

work page 2008

[26] [26]

Burkard, M

R. Burkard, M. Dell’Amico, and S. Martello, Assignment Problems . USA: Society for Industrial and Applied Mathematics, 2009

work page 2009

[27] [27]

Crazyflie 2.1,

Bitcraze, “Crazyflie 2.1,” https://www.bitcraze.io/products/crazyflie-2-1/ (Last accessed: 2025)

work page 2025

[28] [28]

Turtlebot 4,

Clearpath Robotics, “Turtlebot 4,” https://clearpathrobotics.com/ turtlebot-4/ (Last accessed: 2025)

work page 2025

[29] [29]

Crazyswarm2,

W. Honig, K. McGuire et al. , “Crazyswarm2,” https://github.com/ IMRCLab/crazyswarm2 (Last accessed: 2025)

work page 2025

[30] [30]

The OpenCV Library,

G. Bradski, “The OpenCV Library,” Dr. Dobb’s J. Soft. Tools, 2000. APPENDIX A. Proof of Proposition 1 Consider two events — EK(p) = {K(p) \ Sθ−ϵ ̸= ∅} and ER(p) = {R(p) ∩ Sθ+ϵ ̸= ∅}. Specifically, at any epoch p, EK(p) and ER(p) are the undesirable events that a cell in C \ Sθ−ϵ is included in the keep set K(p) and a cell in Sθ+ϵ is included in the reject...

work page 2000