pith. sign in

arxiv: 2607.01029 · v1 · pith:AKC72EXSnew · submitted 2026-07-01 · 💻 cs.RO

AMBUSH: Collaborative Capture in Complex Environments with Neural Acceleration

Junfeng Chen , YinHang Luo , Xinyi Wang , Junrui Li , Meng Guo This is my paper

Pith reviewed 2026-07-02 11:18 UTC · model grok-4.3

classification 💻 cs.RO
keywords ambush strategycollaborative capturepursuit-evasionHybrid Monte Carlo Tree Searchneural accelerationrobotic teamscomplex environmentsevader intelligence
0
0 comments X

The pith

Ambush alone suffices for multiple slower pursuers to capture one faster evader in complex environments.

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

The paper sets out to show that ambush tactics by themselves let several slower robots catch a single faster target even when the target moves intelligently through cluttered spaces with obstacles. A strategy is defined using workspace topology, limited visibility, speed ratios and capture distance to choose ambush positions in advance. Hybrid Monte Carlo Tree Search optimizes the parameters over long horizons while a neural network learns to rank choices across environments so that online planning stays fast. Validation covers simulations and physical tests against evaders of different speeds and human-level control.

Core claim

The central claim is that a parameterized ambush strategy incorporating topological workspace properties, truncated line-of-sight visibility, relative speed ratio and limited capture range can be optimized by Hybrid Monte Carlo Tree Search and ranked by a neural network to enable multiple slower pursuers to capture one faster evader with varying intelligence levels in complex environments.

What carries the argument

Parameterized ambush strategy whose discrete and continuous parameters are optimized by Hybrid Monte Carlo Tree Search and scored by an offline-trained neural network that replaces rollouts.

If this is right

  • Pursuer teams achieve capture without matching evader velocity in obstacle-rich spaces.
  • The approach succeeds against both algorithmic and human-controlled evaders.
  • Neural ranking accelerates planning while preserving solution quality for online deployment.
  • The method extends to security, surveillance and search-and-rescue tasks in real settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the neural scorer generalizes across map types it could shorten planning time in other multi-robot coordination tasks.
  • Adding three-dimensional topology or moving obstacles would test whether the same ambush parameterization still suffices.
  • Scaling experiments with larger numbers of pursuers or simultaneous evaders would reveal practical limits not addressed in the current trials.

Load-bearing premise

The parameterized ambush strategy can be reliably optimized and ranked to guarantee capture success across varied environments and evader intelligence levels.

What would settle it

A single simulation or hardware trial in a complex environment where the optimized ambush parameters fail to intercept an evader moving at twice the pursuer speed or under direct human control would falsify the claim.

Figures

Figures reproduced from arXiv: 2607.01029 by Junfeng Chen, Junrui Li, Meng Guo, Xinyi Wang, Yinhang Luo.

Figure 1
Figure 1. Figure 1: Top: Illustration of the proposed ambush strategy. The pursuer (in purple) hides in concealed positions and makes a surprise attack on the evader (in red) which is driven by other pursuers into the ambush area (gray region). Bottom: Hardware experiments against human-controlled evaders with limited or even full field-of-view. I. INTRODUCTION Dynamic and collaborative capture refers to the process of coordi… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed ambush framework, which (I) generates the role of pursuers as attacker or evader and their motion parameters (right-top); (II) determines the motion policy under the parameters (right-bottom); and (III) oversees the execution progress and reaction of the evader to trigger the adaptation scheme (middle-top). capture, especially when facing faster evaders with different levels of int… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the goal generation and role assignment proce￾dure in different workspaces. (a) Red points denote obstacle skeleton vertices Vo, and olive segments denote obstacle skeleton edges Eo. (b) Boundary vertices Vb are obtained by projecting skeleton vertices onto the workspace boundary, where red boundary points denote the projected vertices and orange segments denote the projection lines. (c) Ca… view at source ↗
Figure 4
Figure 4. Figure 4: Examples of the capture graphs GA in Def. 4 in different workspaces. The generated capture graph forms a polygonal en￾closure around the evader. Green, purple, and blue circles denote pursuers, and the red triangle denotes the evader. Orange points and segments denote obstacle skeleton vertices and edges, respectively. Cross markers indicate hider goals, while star markers indicate attacker goals. connecti… view at source ↗
Figure 5
Figure 5. Figure 5: The overall ambush strategy under different parameters ξ, including the role assignment A and motion parameters ρ, and their impact on the pursuers’ trajectories. As [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Architecture of the proposed neural accelerator NARE: (a) The Neural Network Ranker (NAR) takes the visibility graph Grr to predict the cumulative Q-value for ambush parameters ξ P Ξ; (b) The Neural Network Evaluator (NAE) requries the state-encoding graph Gre to predict the immediate simulation rewards rpνq for nodes ν in the search tree. Both components share a common three pre-processing steps including… view at source ↗
Figure 7
Figure 7. Figure 7: Illustration of the ambush assignments (top), which are ranked and sorted via the learned NAR (bottom). Each gray enclosure denotes the capture graph induced by a specific assignment, including the goals of hiders and attackers. entirely different scenarios featuring distinct topologies and velocity ratios compared to those in the training set. 4) Execution of H-MCTS via NARE: To maintain the solu￾tion qua… view at source ↗
Figure 8
Figure 8. Figure 8: Snapshots of the AMBUSH strategy for 3 pursuers and 1 evader, under different scenarios in the evaluation: Scenario-I (top), Scenario-II (middle); and Scenario-III (bottom). Note that the local view of the evader is shown in the corners, while the switching of pursuer roles and their relative distance to the evader are shown in the right column. the Alg. 1 can be bypassed, and the rollout procedures in Alg… view at source ↗
Figure 9
Figure 9. Figure 9: The capture rate and average capture time (bottom) under different number of pursuers ranging from 2 to 5 over five trials, along with the snapshots of final capture (top) [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Snapshots of the final capture under complex (top) and simple (middle) strategies of the evader and different velocity ratios; Comparisons of the average capture rate and time of different base￾lines, under different evader strategies and velocity ratios (bottom) over five trials with different initial positions. at t “ 5.2s systematically shrunk via P0-P1 herding, culminat￾ing in P2’s emergence-driven ca… view at source ↗
Figure 13
Figure 13. Figure 13 [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Evolution of the capture process under online changes. Top: the number of pursuers increases from 3 to 4 during execution. Bottom: the environment undergoes three layout changes during online execution. fect situational awareness. (IV) Capture Range: Capture range dc critically influences the effectiveness as shown in [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Comparison of the proposed method with five baselines across three scenarios with varying difficulties, where the final capture time is summarized in the bottom. TABLE I COMPARISON OF BASELINES ACROSS THREE SCENARIOS. Scene Method Success Rate (%) Capture Time (s) Plan Time (ˆ10´3 s) NO. of Replans (#) Avg Std Max Min Avg Std Max Min Avg Std Max Min Scenario I analytic 100.0 10.3 2.1 15.8 8.0 6.7 3.1 10.8… view at source ↗
Figure 16
Figure 16. Figure 16: (a) The average node quality along with the expansion order, by the proposed method is much higher than the vanilla MCTS without acceleration; (b) The average capture time by comparing the proposed method exploring different number of nodes with vanilla MCTS at different speed ratios [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Heterogeneous team of pursuers with different capabilities, including (a) different capture ranges, (b) velocities, and (c) both. fails in cluttered environments, and Shooting approach shows the unstable 40´60% performance. Although MCTS achieves 100% success in simpler scenarios, it declines to 80% in Scenario III with longer capture times (21.8s vs our 14.25s). It should be noted that Scenario III is an… view at source ↗
Figure 21
Figure 21. Figure 21: Capture results under different robot dynamic models. The left subfigure shows the capture process and trajectories under the unicycle model, while the right panel illustrates the results under the double-integrator model [PITH_FULL_IMAGE:figures/full_fig_p015_21.png] view at source ↗
Figure 20
Figure 20. Figure 20: Snapshots of different sparse environments. (left) Sparse environment with large obstacles; (middle) Sparse environment with small scattered obstacles; (right) Largely free space. the evader at t “ 5.0s in Phase Q2. After the evader disappears in Phase Q3, all robots revert to search until Pursuer 2 reac￾quires the evader at 26.5s in Phase Q4, enabling coordinated attacks that achieve the successful captu… view at source ↗
Figure 23
Figure 23. Figure 23: Hardware validation of the proposed ambush strategy in a complex environment with 3 pursuers and 1 evader. (a) The experimental setup features a 5m ˆ 5m office-like layout and an autonomous evader; (b) The evader is tele-operated by an operator with limited first person view, where the capture takes places at t “ 12.85s; (c) The tele-operator is given a global view, where the capture failed within 60s. 6)… view at source ↗
Figure 24
Figure 24. Figure 24: Illustration of various aspects considered to select temporary goals for the evader, including betweenness centrality (left), the stochastic perturbation (middle), and the risk-aware distance (right). APPENDIX A THE TIME COMPLEXITY OF GO A L S&AS S I G N Sp¨q The time complexity of Alg. 1 is dominated by the combi￾natorial assignment procedure. Geometric initialization as in Lines 14–21 computes the verti… view at source ↗
Figure 25
Figure 25. Figure 25: The training loss (left) and validation loss (right) for the proposed NAE and NAR networks, in the simulations of Sec. V. Me o to combine the pursuer repulsion, goal attraction, and obstacle avoidance. Optimal paths are computed via the A ‹ search, and updated at every step. APPENDIX C THE TIME COMPLEXITY OF H-MCTSp¨q The computational complexity of H-MCTSp¨q is domi￾nated by the following components, i.e… view at source ↗
Figure 27
Figure 27. Figure 27: The sweeping strategy of the attackers to herd the evader towards the hider gates, where the hiders wait for a surprise capture. Stage II: As shown in [PITH_FULL_IMAGE:figures/full_fig_p019_27.png] view at source ↗
read the original abstract

Collaborative capture of dynamic targets is common in nature as an essential strategy for weaker species against the strong. Similar concepts have shown to be useful for numerous robotic applications, such as security and surveillance, search and rescue. However, most existing works focus on analytical and geometric solutions or end-to-end reinforcement learning methods, which are largely constrained to obstacle-free environments or scenarios with sparse, regularly distributed obstacles. This work tackles the problem from a unique perspective: the renowned strategy of``ambush'' alone would suffice for multiple slower pursuers to capture one faster evader with different levels of intelligence efficiently in complex environments. A parameterized strategy of ambush (including discrete and continuous parameters) is designed first, which takes into account the topological properties of the workspace, the truncated line-of-sight visibility, the relative speed ratio and the limited capture range. Then, a Hybrid Monte Carlo Tree Search (H-MCTS) algorithm is proposed to optimize the associated parameters through long-term planning, enabling the identification of highly promising parameters for future capture. Lastly, the neural acceleration is trained offline to learn the ranking of different choices of parameters across various environments, and to directly predict scores, replacing the rollout process in H-MCTS. The neural acceleration is adopted during online H-MCTS to accelerate the planning procedure while guaranteeing the planning quality. Its efficiency and effectiveness are validated in extensive simulations and hardware experiments, against evaders with different capabilities and intelligence levels, including two-times higher velocity and human-controlled behavior.

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.

Referee Report

1 major / 1 minor

Summary. The paper claims that a parameterized ambush strategy—incorporating topological workspace properties, truncated line-of-sight visibility, relative speed ratio, and limited capture range—suffices for multiple slower pursuers to capture one faster evader (including 2x velocity or human-controlled) in complex environments. The strategy parameters are optimized via Hybrid Monte Carlo Tree Search (H-MCTS) for long-term planning, with a neural network trained offline to rank parameter choices across environments and replace rollouts for online acceleration, with claimed validation via extensive simulations and hardware experiments.

Significance. If the empirical results hold with proper quantitative support, the work offers a practical engineering approach to multi-robot pursuit-evasion that leverages domain-specific ambush parameterization and neural acceleration rather than pure geometric analysis or end-to-end RL, potentially enabling efficient capture in cluttered settings relevant to security, surveillance, and search-and-rescue.

major comments (1)
  1. [Abstract] Abstract: the central claim of validation 'in extensive simulations and hardware experiments' against evaders with different capabilities is asserted without any quantitative results, performance metrics, error bars, success rates, or comparison baselines, rendering the effectiveness of the H-MCTS + neural acceleration pipeline impossible to assess from the provided text.
minor comments (1)
  1. The abstract states 'two-times higher velocity' for the evader; explicit reporting of the exact speed ratios tested and how they interact with the parameterized capture range would improve clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the abstract. We agree that quantitative support should be included to allow readers to assess the claims directly from the abstract.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of validation 'in extensive simulations and hardware experiments' against evaders with different capabilities is asserted without any quantitative results, performance metrics, error bars, success rates, or comparison baselines, rendering the effectiveness of the H-MCTS + neural acceleration pipeline impossible to assess from the provided text.

    Authors: We acknowledge that the abstract currently states the validation without specific metrics. The full manuscript contains detailed quantitative results (success rates, timing comparisons, and baselines) in the experimental sections. We will revise the abstract to concisely report key quantitative highlights from those sections, such as capture success rates across evader types and planning speedups from the neural acceleration. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper describes an empirical engineering pipeline: a parameterized ambush strategy (incorporating workspace topology, truncated visibility, speed ratio, and capture range) is optimized via H-MCTS, after which a neural network is trained offline on the resulting parameter rankings to accelerate online search by replacing rollouts. No equations, uniqueness theorems, or self-citations appear in the provided abstract or high-level description that would make any claimed prediction or result equivalent to its inputs by construction. The neural acceleration step is a standard supervised approximation of an external planner rather than a fitted input renamed as prediction. Validation occurs via separate simulation and hardware experiments against varied evaders, keeping the method self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

Abstract-only view yields minimal ledger; central claim rests on unstated assumptions that ambush parameters can be discretized/continuous and that neural ranking preserves planning quality.

free parameters (1)
  • ambush discrete and continuous parameters
    Abstract states a parameterized strategy is designed first, implying multiple tunable values for topology, visibility, speed ratio, and capture range.

pith-pipeline@v0.9.1-grok · 5805 in / 1094 out tokens · 27993 ms · 2026-07-02T11:18:01.937652+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages · 1 internal anchor

  1. [1]

    Group hunting within the carnivora: physiological, cognitive and environmental influences on strategy and cooperation,

    I. Bailey, J. P. Myatt, and A. M. Wilson, “Group hunting within the carnivora: physiological, cognitive and environmental influences on strategy and cooperation,”Behavioral ecology and sociobiology, vol. 67, no. 1, pp. 1–17, 2013

    work page 2013

  2. [2]

    Intercepting rogue robots: An algorithm for capturing multiple evaders with multiple pursuers,

    A. Pierson, Z. Wang, and M. Schwager, “Intercepting rogue robots: An algorithm for capturing multiple evaders with multiple pursuers,”IEEE Robotics and Automation Letters, vol. 2, no. 2, pp. 530–537, 2016

    work page 2016

  3. [3]

    Adaptive partitioning for coordinated multi-agent perimeter defense,

    D. G. Macharet, A. K. Chen, D. Shishika, G. J. Pappas, and V . Kumar, “Adaptive partitioning for coordinated multi-agent perimeter defense,” inIEEE/RSJ international conference on intelligent robots and Systems (IROS), 2020, pp. 7971–7977

    work page 2020

  4. [4]

    Multi-usv cooperative chasing strategy based on obstacles assistance and deep reinforcement learning,

    W. Gan, X. Qu, D. Song, and P. Yao, “Multi-usv cooperative chasing strategy based on obstacles assistance and deep reinforcement learning,” IEEE Transactions on Automation Science and Engineering, vol. 21, no. 4, pp. 5895–5910, 2023

    work page 2023

  5. [5]

    Poaching detection technologies—a survey,

    J. Kamminga, E. Ayele, N. Meratnia, and P. Havinga, “Poaching detection technologies—a survey,”Sensors, vol. 18, no. 5, p. 1474, 2018

    work page 2018

  6. [6]

    Uav-ugv-umv multi-swarms for cooperative surveillance,

    D. H. Stolfi, M. R. Brust, G. Danoy, and P. Bouvry, “Uav-ugv-umv multi-swarms for cooperative surveillance,”Frontiers in Robotics and AI, vol. 8, no. 1, p. 616950, 2021

    work page 2021

  7. [7]

    A game of cops and robbers,

    M. A.-M. Fromme and M. Aigner, “A game of cops and robbers,” Discrete Appl. Math, vol. 8, no. 2, pp. 1–12, 1984

    work page 1984

  8. [8]

    Multiple pursuer multiple evader differential games,

    E. Garcia, D. W. Casbeer, A. V on Moll, and M. Pachter, “Multiple pursuer multiple evader differential games,”IEEE Transactions on Automatic Control, vol. 66, no. 5, pp. 2345–2350, 2020

    work page 2020

  9. [9]

    Bonato,The game of cops and robbers on graphs

    A. Bonato,The game of cops and robbers on graphs. American Mathematical Soc., 2011

    work page 2011

  10. [10]

    A visibility roadmap sampling approach for a multi- robot visibility-based pursuit-evasion problem,

    T. Olsenet al., “A visibility roadmap sampling approach for a multi- robot visibility-based pursuit-evasion problem,” inIEEE International Conference on Robotics and Automation (ICRA), 2021, pp. 7957–7964

    work page 2021

  11. [11]

    Distributed encirclement and capture of multiple pursuers with collision avoidance,

    X. Wang, L. Xi, Y . Ding, and B. M. Chen, “Distributed encirclement and capture of multiple pursuers with collision avoidance,”IEEE Trans- actions on Industrial Electronics, vol. 71, no. 7, pp. 7520–7530, 2023

    work page 2023

  12. [12]

    Cooperative pursuit with voronoi partitions,

    Z. Zhou, W. Zhang, J. Ding, H. Huang, D. M. Stipanovi ´c, and C. J. Tom- lin, “Cooperative pursuit with voronoi partitions,”Automatica, vol. 72, no. 10, pp. 64–72, 2016

    work page 2016

  13. [13]

    Group chasing tactics: how to catch a faster prey,

    M. Janosov, C. Vir ´agh, G. V ´as´arhelyi, and T. Vicsek, “Group chasing tactics: how to catch a faster prey,”New Journal of Physics, vol. 19, no. 5, pp. 053–063, 2017

    work page 2017

  14. [14]

    Multi-agent actor-critic for mixed cooperative-competitive environ- ments,

    R. Lowe, Y . I. Wu, A. Tamar, J. Harb, O. Pieter Abbeel, and I. Mordatch, “Multi-agent actor-critic for mixed cooperative-competitive environ- ments,” inAdvances in neural information processing systems, 2017, pp. 30–35

    work page 2017

  15. [15]

    On developing a uav pursuit-evasion policy using reinforcement learning,

    B. Vlahovet al., “On developing a uav pursuit-evasion policy using reinforcement learning,” inIEEE International Conference on Machine Learning and Applications (ICMLA), 2018, pp. 859–864

    work page 2018

  16. [16]

    Emergent behaviors in multiagent pursuit evasion games within a bounded 2d grid world,

    S. Xu and Z. Dang, “Emergent behaviors in multiagent pursuit evasion games within a bounded 2d grid world,”Scientific Reports, vol. 15, no. 1, p. 29376, 2025

    work page 2025

  17. [17]

    Modeling and analysis of cooperative pursuit actions in two-dimensional multi-agent pursuit-evasion games,

    S. Xu, Z. Liu, and Z. Dang, “Modeling and analysis of cooperative pursuit actions in two-dimensional multi-agent pursuit-evasion games,” in2025 40th Youth Academic Annual Conference of Chinese Association of Automation (YAC), 2025, pp. 665–670

    work page 2025

  18. [18]

    A survey of the pursuit– evasion problem in swarm intelligence,

    Z. Mu, J. Pan, Z. Zhou, J. Yu, and L. Cao, “A survey of the pursuit– evasion problem in swarm intelligence,”Frontiers of Information Tech- nology & Electronic Engineering, vol. 24, no. 8, pp. 1093–1116, 2023

    work page 2023

  19. [19]

    Pursuit-evasion of an evader by multiple pur- suers,

    A. V on Mollet al., “Pursuit-evasion of an evader by multiple pur- suers,” inIEEE International Conference on Unmanned Aircraft Systems (ICUAS), 2018, pp. 133–142

    work page 2018

  20. [20]

    A dimension-reduction solution of free-time differential games for spacecraft pursuit-evasion,

    Z.-y. Li, H. Zhu, Z. Yang, and Y .-z. Luo, “A dimension-reduction solution of free-time differential games for spacecraft pursuit-evasion,” Acta Astronautica, vol. 163, no. 19, pp. 201–210, 2019

    work page 2019

  21. [21]

    Differential games. a mathematical theory with applications to warfare and pursuit, control and optimization

    S. Vajda, “Differential games. a mathematical theory with applications to warfare and pursuit, control and optimization.”The Mathematical Gazette, vol. 51, no. 375, pp. 80–81, 1967

    work page 1967

  22. [22]

    Isaacs,Differential games: a mathematical theory with applications to warfare and pursuit, control and optimization

    R. Isaacs,Differential games: a mathematical theory with applications to warfare and pursuit, control and optimization. Courier Corporation, 1999

    work page 1999

  23. [23]

    Pursuit-evasion differential games of players with different speeds in spaces of different dimensions,

    S. Li, C. Wang, and G. Xie, “Pursuit-evasion differential games of players with different speeds in spaces of different dimensions,” inIEEE American Control Conference (ACC), 2022, pp. 1299–1304

    work page 2022

  24. [24]

    Encirclement guaranteed cooperative pursuit with robust model predictive control,

    C. Wang, H. Chen, J. Pan, and W. Zhang, “Encirclement guaranteed cooperative pursuit with robust model predictive control,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021, pp. 1473–1479

    work page 2021

  25. [25]

    Cooperative pursuit with multi-pursuer and one faster free-moving evader,

    X. Fang, C. Wang, L. Xie, and J. Chen, “Cooperative pursuit with multi-pursuer and one faster free-moving evader,”IEEE transactions on cybernetics, vol. 52, no. 3, pp. 1405–1414, 2020

    work page 2020

  26. [26]

    Game-theoretic utility tree for multi-robot cooperative pursuit strategy,

    Q. Yang and R. Parasuraman, “Game-theoretic utility tree for multi-robot cooperative pursuit strategy,” inISR Europe International Symposium on Robotics, 2022, pp. 1–7

    work page 2022

  27. [27]

    Fg-pe: Factor-graph approach for multi-robot pursuit-evasion,

    M. A. Esfahani, A. Bas ¸ar, and S. Saeedi, “Fg-pe: Factor-graph approach for multi-robot pursuit-evasion,”arXiv preprint arXiv:2411.00741, 2024

    work page arXiv 2024

  28. [28]

    Decentralized multi-agent pursuit using deep reinforcement learning,

    C. De Souza, R. Newbury, A. Cosgun, P. Castillo, B. Vidolov, and D. Kuli ´c, “Decentralized multi-agent pursuit using deep reinforcement learning,”IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4552–4559, 2021

    work page 2021

  29. [29]

    Cooperative control for multi-player pursuit-evasion games with reinforcement learning,

    Y . Wang, L. Dong, and C. Sun, “Cooperative control for multi-player pursuit-evasion games with reinforcement learning,”Neurocomputing, vol. 412, no. 1, pp. 101–114, 2020

    work page 2020

  30. [30]

    Learning to play pursuit-evasion with dynamic and sensor constraints,

    B. M. Gonultas and V . Isler, “Learning to play pursuit-evasion with dynamic and sensor constraints,”arXiv preprint arXiv:2405.05372, 2024

    work page arXiv 2024

  31. [31]

    Pursuit-evasion with decen- tralized robotic swarm in continuous state space and action space via deep reinforcement learning

    G. Singh, D. M. Lofaro, and D. Sofge, “Pursuit-evasion with decen- tralized robotic swarm in continuous state space and action space via deep reinforcement learning.” inInternational Conference on Agents and Artificial Intelligence, 2020, pp. 226–233

    work page 2020

  32. [32]

    On discrete-time pursuit-evasion games with sensing limitations,

    S. D. Bopardikaret al., “On discrete-time pursuit-evasion games with sensing limitations,”IEEE Transactions on Robotics, vol. 24, no. 6, pp. 1429–1439, 2008

    work page 2008

  33. [33]

    Distributed pursuit of an evader with collision and obstacle avoidance,

    B. Tian, P. Li, H. Lu, Q. Zong, and L. He, “Distributed pursuit of an evader with collision and obstacle avoidance,”IEEE Transactions on Cybernetics, vol. 52, no. 12, pp. 13 512–13 520, 2021

    work page 2021

  34. [34]

    Decentralized multi-robot pursuit of an evader in obstacle environments,

    K. Rao, H. Yan, P. Yang, and Y . Lv, “Decentralized multi-robot pursuit of an evader in obstacle environments,” inIEEE Asian Control Conference (ASCC), 2024, pp. 321–326

    work page 2024

  35. [35]

    Pursuit winning strategies for reach-avoid games with polygonal obstacles,

    R. Yan, S. Mi, X. Duan, J. Chen, and X. Ji, “Pursuit winning strategies for reach-avoid games with polygonal obstacles,”IEEE Transactions on Automatic Control, vol. 70, no. 2, pp. 814–829, 2024

    work page 2024

  36. [36]

    Pursuit-evasion problems of multi- agent systems in cluttered environments,

    J. Ericsson and M. Bock Agerman, “Pursuit-evasion problems of multi- agent systems in cluttered environments,” 2024

    work page 2024

  37. [37]

    Probabilistic strategies for pursuit in cluttered environments with multiple robots,

    G. Hollingeret al., “Probabilistic strategies for pursuit in cluttered environments with multiple robots,” inIEEE International Conference on Robotics and Automation, 2007, pp. 3870–3876

    work page 2007

  38. [38]

    Game of drones: Multi-uav pursuit-evasion game with online motion planning by deep reinforcement learning,

    R. Zhang, Q. Zong, X. Zhang, L. Dou, and B. Tian, “Game of drones: Multi-uav pursuit-evasion game with online motion planning by deep reinforcement learning,”IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 10, pp. 7900–7909, 2022

    work page 2022

  39. [39]

    Multi-uav pursuit-evasion with online planning in unknown environments by deep reinforcement learning,

    J. Chen, C. Yu, G. Li, W. Tang, X. Yang, B. Xu, H. Yang, and Y . Wang, “Multi-uav pursuit-evasion with online planning in unknown environments by deep reinforcement learning,”IEEE Robotics and Automation Letters, vol. 10, no. 8, pp. 8196–8203, 2025

    work page 2025

  40. [40]

    Multi-target pursuit by a decentralized heterogeneous uav swarm using deep multi- agent reinforcement learning,

    M. Kouzeghar, Y . Song, M. Meghjani, and R. Bouffanais, “Multi-target pursuit by a decentralized heterogeneous uav swarm using deep multi- agent reinforcement learning,” inIEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 3289–3295

    work page 2023

  41. [41]

    Multi-robot cooperative pursuit via potential field-enhanced reinforcement learning,

    Z. Zhang, X. Wang, Q. Zhang, and T. Hu, “Multi-robot cooperative pursuit via potential field-enhanced reinforcement learning,” inIEEE International Conference on Robotics and Automation (ICRA), 2022, pp. 8808–8814

    work page 2022

  42. [42]

    Pursuit-evasion game strategy of usv based on deep reinforcement learning in complex multi-obstacle environment,

    X. Qu, W. Gan, D. Song, and L. Zhou, “Pursuit-evasion game strategy of usv based on deep reinforcement learning in complex multi-obstacle environment,”Ocean Engineering, vol. 273, no. 19, pp. 114–125, 2023

    work page 2023

  43. [43]

    Viper: Visibility-based pursuit-evasion via reinforcement learning,

    Y . Wang, Y . Cao, J. Chiun, S. Koley, M. Pham, and G. A. Sartoretti, “Viper: Visibility-based pursuit-evasion via reinforcement learning,” in 8th Annual Conference on Robot Learning, 2024

    work page 2024

  44. [44]

    Toward multi-target self-organizing pursuit in a partially observable markov game,

    L. Sun, Y .-C. Chang, C. Lyu, Y . Shi, Y . Shi, and C.-T. Lin, “Toward multi-target self-organizing pursuit in a partially observable markov game,”Information Sciences, vol. 648, p. 119475, 2023

    work page 2023

  45. [45]

    Matrixworld: A pursuit-evasion platform for safe multi-agent coordination and autocur- ricula,

    L. Sun, Y .-C. Chang, C. Lyu, C.-T. Lin, and Y . Shi, “Matrixworld: A pursuit-evasion platform for safe multi-agent coordination and autocur- ricula,”arXiv preprint arXiv:2307.14854, 2023

    work page arXiv 2023

  46. [46]

    Distributed pursuit-evasion without mapping or global localization via local frontiers,

    J. W. Durham, A. Franchi, and F. Bullo, “Distributed pursuit-evasion without mapping or global localization via local frontiers,”Autonomous Robots, vol. 32, no. 1, pp. 81–95, 2012

    work page 2012

  47. [47]

    Gurobi Optimizer Reference Manual

    Gurobi Optimization, LLC, “Gurobi Optimizer Reference Manual.” [Online]. Available: https://www.gurobi.com

    work page

  48. [48]

    Belief state monte carlo planning for multi-agent visibility-based pursuit-evasion,

    X. Xu, S. Guo, D. Li, P. Yu, X. Liu, and Y . Ren, “Belief state monte carlo planning for multi-agent visibility-based pursuit-evasion,” inAsian Simulation Conference. Springer, 2022, pp. 431–445

    work page 2022

  49. [49]

    Holistically guided monte carlo tree search for intricate information seeking,

    R. Ren, Y . Wang, J. Li, J. Jiang, W. X. Zhao, W. Wang, and T.-S. Chua, “Holistically guided monte carlo tree search for intricate information seeking,”arXiv preprint arXiv:2502.04751, 2025

    work page arXiv 2025

  50. [50]

    Decentralized control strategies for unmanned aircraft system pursuit and evasion,

    D. W. King and G. L. Peterson, “Decentralized control strategies for unmanned aircraft system pursuit and evasion,” inIEEE Vehicular Technology Conference, 2019, pp. 1–5

    work page 2019

  51. [51]

    Bandit based monte-carlo planning,

    L. Kocsis and C. Szepesv ´ari, “Bandit based monte-carlo planning,” in European conference on machine learning. Springer, 2006, pp. 282– 293

    work page 2006

  52. [52]

    How powerful are graph neural networks?

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph neural networks?” inInternational Conference on Learning Represen- tations, 2019, pp. 120–130

    work page 2019

  53. [53]

    Graph Attention Networks

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y . Ben- gio, “Graph attention networks,”arXiv preprint arXiv:1710.10903, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  54. [54]

    Traversing mars: Cooperative informative path planning to efficiently navigate unknown scenes,

    F. M. Rockenbaueret al., “Traversing mars: Cooperative informative path planning to efficiently navigate unknown scenes,”IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 1776–1783, 2024

    work page 2024

  55. [55]

    Novel implementation of multi-robot space exploration utilizing coordinated multi-robot exploration and frequency modified whale optimization algorithm,

    F. Gulet al., “Novel implementation of multi-robot space exploration utilizing coordinated multi-robot exploration and frequency modified whale optimization algorithm,”IEEE access, vol. 9, no. 12, pp. 22 774– 22 787, 2021

    work page 2021

  56. [56]

    Accelerated k-serial stable coalition for dynamic capture and resource defense,

    J. Chen, Z. Tang, and M. Guo, “Accelerated k-serial stable coalition for dynamic capture and resource defense,”IEEE Robotics and Automation Letters, vol. 9, no. 1, pp. 443–450, 2023

    work page 2023

  57. [57]

    A visibility graph approach for path planning and real-time collision avoidance on maritime unmanned systems,

    E. D’Amatoet al., “A visibility graph approach for path planning and real-time collision avoidance on maritime unmanned systems,” inIEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea), 2021, pp. 400–405

    work page 2021

  58. [58]

    Optimal reciprocal collision avoidance for multi-agent navigation,

    J. Van Den Berg, S. J. Guy, M. Lin, and D. Manocha, “Optimal reciprocal collision avoidance for multi-agent navigation,” inProc. of the IEEE International Conference on Robotics and Automation (ICRA), Anchorage (AK), USA, 2010

    work page 2010