pith. sign in

arxiv: 2606.01170 · v1 · pith:QQQ3PXJJnew · submitted 2026-05-31 · 💻 cs.MA · cs.RO

Coordinating Task Switching in a Robotics Multi-Agent System Using Behavior Trees

Lucas Haug , Anarosa Alves Franco Brand\~ao , Arthur Casals This is my paper

Pith reviewed 2026-06-28 16:19 UTC · model grok-4.3

classification 💻 cs.MA cs.RO
keywords behavior treesmulti-agent roboticstask coordinationVSSS soccerfinite state machinesrobot teamssimulation evaluation
0
0 comments X

The pith

Behavior trees coordinate task switching among three robots in a small soccer game, replacing the team's earlier finite-state-machine controller.

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

The paper introduces a behavior-tree system to manage how three robots on the same team switch between different tasks during a soccer match. In this very small league the robots must constantly adjust their roles in a fast-changing environment, making reliable coordination necessary for success. The authors compare the new behavior-tree controller against the previous finite-state-machine version by running both in the FIRASim simulator to measure differences in game outcomes. They also put the behavior-tree version into an actual academic competition to observe its performance with real robots. This application demonstrates that behavior trees can handle the coordination requirements of multi-robot teams in dynamic competitive settings.

Core claim

The authors present a Behavior-Tree-based approach to support multi-robot coordination within the VSSS team of the ThundeRatz robotics team. A comparison between the proposed approach and the previous one based on a Finite State Machine was conducted using the FIRASim simulator. The performance of this new strategy was further evaluated in an academic robotics competition.

What carries the argument

Behavior trees that define rules for switching tasks and roles among multiple robots on a soccer team.

If this is right

  • The behavior-tree controller can be substituted for the finite-state-machine design to manage robot coordination.
  • Simulator matches between the two controllers produce observable performance differences.
  • The behavior-tree version operates successfully when deployed on real robots in competition.
  • Coordination logic becomes modular and easier to extend for new team behaviors.

Where Pith is reading between the lines

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

  • Behavior trees may simplify adding new coordination rules compared with rewriting states and transitions in a finite-state machine.
  • The approach could transfer to other multi-robot tasks such as warehouse transport or search-and-rescue teams.
  • Simulator validation reduces the number of costly real-robot experiments needed before competition use.

Load-bearing premise

Observed performance differences between the two controllers in simulation will correspond to meaningful advantages when the behavior-tree team plays against other real teams in competition.

What would settle it

A series of real-robot matches in which the behavior-tree team wins no more often than the finite-state-machine team would falsify the claim of practical improvement.

Figures

Figures reproduced from arXiv: 2606.01170 by Anarosa Alves Franco Brand\~ao, Arthur Casals, Lucas Haug.

Figure 1
Figure 1. Figure 1: Mapping of the system structural specification using [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Previous FSM of the coordination strategy. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Base structure of the Coach’s Behavior Tree. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: It has two sub-branches, one for the moment [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 4
Figure 4. Figure 4: Roles Swapper Subtree. • ? → Blackboard Check use two strikers mode == True → Change from Two Strikers to One Striker mode Set Blackboard use two strikers mode = False → Blackboard Check use penalty mode == True → Change Penalty Kicker to Striker Set Blackboard use penalty mode = False Swap Striker and Assistant Always Success [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Attack state internal swap subtree. • ? → Blackboard Check use penalty mode == True → Change Penalty Defender to Goalkeeper Set Blackboard use penalty mode = False Swap Fullback and Wingback Swap Goalkeeper and Wingback Always Success [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Defense state internal swap subtree. for its open-source nature, large community, and wide use in different robotics applications. 5. EVALUATION 5.1 Functional and non-functional comparison To make a comparison between the FSM-based team and the BT-based team, it is necessary to take into account the objectives of the refactoring. The change was mainly aimed at modifying the system’s control architecture t… view at source ↗
Figure 7
Figure 7. Figure 7: Goals difference histogram: µ = 0.2120, σ = 1.1794. totaling 2500 minutes of game. The overall results of the tests are presented in Tables 1a and 1b and in [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
read the original abstract

The application of multi-agent systems in robotics is a very challenging field. Several competitions involving such systems are proposed to foster research and development of strategies and mechanisms using games as the underlying domain. Among them are the ones from the \textit{IEEE Very Small Soccer (VSSS)} category, which is the case study described in this paper. In VSSS, two teams of three robots each compete in a very dynamic environment of a soccer game. Thus, coordination of robots' behavior during the game is crucial to win it. In this paper, we present a Behavior-Tree-based approach to support multi-robot coordination within the VSSS team of the ThundeRatz robotics team from the Universidade de S$\tilde{a}$o Paulo. Moreover, a comparison between the proposed approach and the previous one, which was based on a Finite State Machine (FSM), was conducted using the FIRASim simulator. Besides that, the performance of this new strategy was further evaluated in an academic robotics competition.

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

2 major / 2 minor

Summary. The manuscript presents a Behavior-Tree-based approach for coordinating task switching in a three-robot team for the IEEE Very Small Soccer (VSSS) competition. It claims this method was compared against the team's prior Finite State Machine implementation via the FIRASim simulator and that the new strategy was further evaluated through deployment in an academic robotics competition.

Significance. If the comparison results are reproducible and the simulator-to-real gap is addressed, the work would supply a concrete engineering case study of behavior trees for reactive multi-agent coordination in a highly dynamic domain. The combination of simulator benchmarking and real-competition deployment is a strength that could inform similar robotics teams, though the current absence of quantitative evidence prevents a firm assessment of impact.

major comments (2)
  1. [Abstract] Abstract: the claim that 'a comparison between the proposed approach and the previous one... was conducted using the FIRASim simulator' is load-bearing for the central contribution, yet the abstract (and available text) supplies no methods, metrics, trial counts, statistical tests, or performance deltas. Without these, the superiority or viability claim cannot be evaluated.
  2. [Evaluation and results sections] Evaluation and results sections: the manuscript does not report how FIRASim models (or fails to model) robot-specific dynamics, wheel slip, vision latency, or inter-robot communication jitter. These factors directly affect task-switching coordination; their omission leaves the simulator-to-real transfer unverified and undermines the claim that observed differences support real-robot advantages.
minor comments (2)
  1. Add explicit captions and axis labels to any figures or tables that present the FIRASim comparison so that quantitative outcomes are immediately readable.
  2. Clarify the precise definition of 'task switching' and the coordination protocol (e.g., role assignment, conflict resolution) early in the methods section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'a comparison between the proposed approach and the previous one... was conducted using the FIRASim simulator' is load-bearing for the central contribution, yet the abstract (and available text) supplies no methods, metrics, trial counts, statistical tests, or performance deltas. Without these, the superiority or viability claim cannot be evaluated.

    Authors: We agree that the abstract would be strengthened by briefly summarizing the evaluation approach. The manuscript body describes the FIRASim comparison and competition deployment; we will revise the abstract to include a concise statement of the comparison method, key metrics employed, and observed outcomes while remaining within length constraints. revision: yes

  2. Referee: [Evaluation and results sections] Evaluation and results sections: the manuscript does not report how FIRASim models (or fails to model) robot-specific dynamics, wheel slip, vision latency, or inter-robot communication jitter. These factors directly affect task-switching coordination; their omission leaves the simulator-to-real transfer unverified and undermines the claim that observed differences support real-robot advantages.

    Authors: We concur that explicit discussion of simulator fidelity would improve transparency. We will add a short paragraph in the Evaluation section describing FIRASim's modeling assumptions for the VSSS domain and known limitations with respect to dynamics, slip, latency, and communication, drawing on our team's experience with both simulation and hardware. revision: yes

Circularity Check

0 steps flagged

No circularity: direct empirical comparison with no derivations or fitted parameters

full rationale

The paper describes an engineering implementation of Behavior Trees for multi-robot task coordination in VSSS soccer, followed by a head-to-head comparison against a prior FSM baseline in the FIRASim simulator and in live competition. No equations, parameter fitting, uniqueness theorems, or self-citation chains appear in the provided text. The central claim rests on observable performance differences rather than any reduction of outputs to inputs by construction. This is the normal case of a self-contained systems paper whose validity is open to external falsification via replication or real-robot testing.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical model, free parameters, axioms, or invented entities are present; the contribution is an engineering application of existing tools.

pith-pipeline@v0.9.1-grok · 5707 in / 961 out tokens · 20500 ms · 2026-06-28T16:19:56.856445+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

16 extracted references · 8 canonical work pages

  1. [1]

    Agis, R.A., Gottifredi, S., and Garc´ ıa, A.J. (2020). An event-driven behavior trees extension to facilitate non-player multi-agent coordination in video games. Expert Systems with Applications, 155, 113457. doi: https://doi.org/10.1016/j.eswa.2020.113457. URL https://www.sciencedirect.com/science/article/ pii/S0957417420302815. ˚Agotnes, T. and Bulling,...

    work page doi:10.1016/j.eswa.2020.113457 2020

  2. [2]

    Biggar, O., Zamani, M., and Shames, I. (2021). An expressiveness hierarchy of behavior trees and related architectures.IEEE Robotics and Automation Letters, 6(3), 5397–5404. doi:10.1109/LRA.2021.3074337

    work page doi:10.1109/lra.2021.3074337 2021

  3. [3]

    Billington, D., Estivill-Castro, V., Hexel, R., and Rock, A. (2010). Plausible logic facilitates engineering the behav- ior of autonomous robots. InThe IASTED International Conference on Software Engineering, 41–48. CBR (2022). Regras IEEE very small size soccer (VSSS) - s´ erie B.https://cbr.robocup.org.br/wp-content/ uploads/2021/05/vssRules3x321.pdf. A...

    2010

  4. [4]

    and Shi, D

    Chen, J. and Shi, D. (2018). Development and compo- sition of robot architecture in dynamic environment. InProceedings of the 2018 International Conference on

    2018

  5. [5]

    Association for Computing Machinery, New York, NY, USA

    Robotics, Control and Automation Engineering, RCAE 2018, 96–101. Association for Computing Machinery, New York, NY, USA. doi:10.1145/3303714.3303716. URLhttps://doi.org/10.1145/3303714.3303716

    work page doi:10.1145/3303714.3303716 2018

  6. [6]

    Colledanchise, M., Marzinotto, A., Dimarogonas, D.V., and Oegren, P. (2016). The advantages of using behavior trees in mult-robot systems. InProceedings of ISR 2016: 47st International Symposium on Robotics, 1–8

    2016

  7. [7]

    Colledanchise, M., Parasuraman, R., and ¨Ogren, P. (2019). Learning of behavior trees for autonomous agents.IEEE Transactions on Games, 11(2), 183–189. doi:10.1109/ TG.2018.2816806

    arXiv 2019

  8. [8]

    and ¨Ogren, P

    Colledanchise, M. and ¨Ogren, P. (2018).Behavior Trees in Robotics and AI. CRC Press. doi:10.1201/ 9780429489105. URLhttps://doi.org/10.1201% 2F9780429489105

    2018

  9. [9]

    and Colledanchise, M

    Faconti, D. and Colledanchise, M. (2022). Be- haviourtree.cpp.https://github.com/BehaviorTree/ BehaviorTree.CPP/. Accessed: 2023-06-28

    2022

  10. [10]

    Dragule, S. (2023). Behavior trees and state machines in robotics applications.IEEE Transactions on Software Engineering, 1–24. doi:10.1109/TSE.2023.3269081

    work page doi:10.1109/tse.2023.3269081 2023

  11. [11]

    Gillies, M. (2009). Learning finite-state machine con- trollers from motion capture data.IEEE Transactions on Computational Intelligence and AI in Games, 1(1), 63–72. doi:10.1109/TCIAIG.2009.2019630

    work page doi:10.1109/tciaig.2009.2019630 2009

  12. [12]

    Iovino, M., F ¨orster, J., Falco, P., Chung, J.J., Sieg- wart, R., and Smith, C. (2022a). On the program- ming effort required to generate behavior trees and finite state machines for robotic applications.arXiv preprint arXiv:2209.07392. URLhttps://arxiv.org/ abs/2209.07392

    arXiv

  13. [13]

    Smith, C. (2022b). A survey of behavior trees in robotics and ai.Robotics and Autonomous Sys- tems, 154, 104096. doi:https://doi.org/10.1016/j.robot. 2022.104096. URLhttps://www.sciencedirect.com/ science/article/pii/S0921889022000513

    work page doi:10.1016/j.robot 2022

  14. [14]

    League, V.S.S. (2023). Very small size league.https: //vsssleague.github.io/vss/index.html. Accessed: 2026-03-16. Microsoft (2022). Blackboard design pattern.https:// social.technet.microsoft.com/wiki/contents/ articles/13215.blackboard-design-pattern.aspx. Accessed: 2026-03-17. RoboCore (2023). Simulado iron 2023 – tabela VSSS.https://docs.google.com/spr...

    2023

  15. [15]

    Wang, F., Kyriakopoulos, K., Tsolkas, A., and Saridis, G. (1991). A petri-net coordination model for an intelligent mobile robot.IEEE Transactions on Systems, Man, and Cybernetics, 21(4), 777–789. doi:10.1109/21.108296

    work page doi:10.1109/21.108296 1991

  16. [16]

    Yang, Q., Luo, Z., Song, W., and Parasuraman, R. (2019). Self-reactive planning of multi-robots with dynamic task assignments. In2019 International Symposium on Multi-Robot and Multi-Agent Systems (MRS), 89–91. doi:10.1109/MRS.2019.8901075

    work page doi:10.1109/mrs.2019.8901075 2019