pith. sign in

arxiv: 2601.17017 · v2 · submitted 2026-01-15 · 💻 cs.CY · cs.MA· math.OC

Self-Organizing Railway Traffic Management

Federico Naldini , Fabio Oddi , Leo D'Amato , Gr\'egory Marli\`ere , Vito Trianni , Paola Pellegrini This is my paper

Pith reviewed 2026-05-16 14:20 UTC · model grok-4.3

classification 💻 cs.CY cs.MAmath.OC
keywords self-organizationrailway traffic managementdecentralized controlperturbation managementconsensus mechanismtraffic plandelay reductioninstance decomposition
0
0 comments X

The pith

Trains self-organize traffic management by identifying neighbors, forming hypotheses, checking compatibility, and reaching consensus to produce executable plans that reduce delays more than centralized algorithms.

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

The paper introduces a decentralized process in which trains detect their immediate neighbors, each generates possible ways to resolve local perturbations, the trains test which proposals can coexist without conflict, and a consensus step selects and combines the strongest compatible set into one ready-to-use schedule. Experiments run this process inside a closed-loop simulator of an Italian rail segment and compare the outcomes directly against a state-of-the-art centralized optimizer under the same disruptions. The authors report that the self-organized plans limit delay propagation better than the centralized baseline, and they attribute the gain to the way the method breaks the overall problem into smaller, locally solvable pieces. A reader would care because real railways often suffer cascading delays when a single central solver cannot react fast enough or when communication to the center is slow.

Core claim

The central claim is that a modular self-organization process—trains identifying neighbors, formulating traffic management hypotheses, checking their compatibility, and selecting the best compatible set through consensus—produces a directly applicable traffic plan that outperforms a state-of-the-art centralized algorithm, specifically by defining and exploiting an instance decomposition of the overall traffic-management problem.

What carries the argument

The modular self-organization process of neighbor identification, hypothesis formulation, compatibility checking, and consensus mechanism that merges selected hypotheses into a single executable traffic plan.

If this is right

  • Self-organization reduces delay propagation more effectively than centralized decision-making on the tested network segment.
  • The performance edge arises from the explicit decomposition of the global problem into compatible local instances that trains can resolve without global oversight.
  • The final merged plan can be applied directly by the trains without further central intervention.
  • The advantage is demonstrated in closed-loop microscopic simulation that models realistic train dynamics and communication timing.

Where Pith is reading between the lines

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

  • If train-to-train communication proves reliable at scale, the approach could lower the infrastructure cost of maintaining a always-available central control center.
  • The decomposition technique might allow the method to handle larger or more densely perturbed networks than a single centralized solver can process in time.
  • Real deployment would require testing whether the consensus step remains stable when trains move at different speeds and when some proposed hypotheses become invalid mid-negotiation.

Load-bearing premise

Trains can reliably identify their current neighbors, formulate and exchange hypotheses, and reach consensus in real time without communication failures or excessive latency under actual railway conditions.

What would settle it

A controlled simulation run in which neighbor detection or consensus messages are delayed or corrupted and the resulting self-organized plan produces longer total delays than the centralized algorithm on the same perturbation instances.

Figures

Figures reproduced from arXiv: 2601.17017 by Fabio Oddi, Federico Naldini, Gr\'egory Marli\`ere, Leo D'Amato, Paola Pellegrini, Vito Trianni.

Figure 1
Figure 1. Figure 1: Conceptual representation of the self-organized traffic management [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Top: Four trains traveling in a control area. Each train’s neighborhood [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic representation of the Segrate-Ospitaletto control area [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Space-time diagram representing the timetable considered. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Percentage improvement of total weighted and unweighted delay of [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Representation of traffic and its management evolution across [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Analysis of the consensus process: A: Regret distribution. B: Distribution of the number of decision steps per train. In a thorough experimental analysis covering a portion of the Italian network shared by different railway under￾takings operating mixed traffic, we showed that the traffic self-organization was able to outperform the state-of-the-art centralized decision-making algorithm used as a benchmark… view at source ↗
read the original abstract

Improving traffic management in case of perturbation is one of the main challenges in today's railway research. The great majority of the existing literature proposes approaches to make centralized decisions to minimize delay propagation. In this paper, we propose a new paradigm to the same aim: we design and implement a modular process to allow trains to self-organize. This process consists in having trains identifying their neighbors, formulating traffic management hypotheses, checking their compatibility and selecting the best ones through a consensus mechanism. Finally, these hypotheses are merged into a directly applicable traffic plan. In a thorough experimental analysis on a portion of the Italian network, we compare the results of self-organization with those of a state-of-the-art centralized approach. In particular, we make this comparison mimicking a realistic deployment thanks to a closed-loop framework including a microscopic railway simulator. The results indicate that self-organization achieves better results than the centralized algorithm, specifically thanks to the definition and exploitation of the instance decomposition allowed by the proposed approach.

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 / 1 minor

Summary. The paper proposes a decentralized self-organizing approach to railway traffic management under perturbations. Trains identify neighbors, formulate traffic hypotheses, check compatibility, reach consensus via a mechanism, and merge results into an executable plan. On a portion of the Italian network, a closed-loop microscopic simulator comparison shows the self-organizing method outperforming a state-of-the-art centralized algorithm, with the gain attributed to explicit instance decomposition.

Significance. If the performance advantage holds under realistic communication constraints, the work offers a scalable alternative to centralized control that could improve resilience in disrupted railway networks. The closed-loop evaluation framework is a positive step toward deployment realism, though the absence of communication modeling limits immediate applicability.

major comments (2)
  1. [Experimental analysis / closed-loop framework] The experimental comparison (described in the abstract and experimental analysis) attributes the performance gain to instance decomposition and real-time consensus, yet the closed-loop simulator provides no explicit model, latency bounds, packet-loss rates, or failure modes for neighbor identification and hypothesis exchange. This assumption is load-bearing for the central claim that self-organization works under realistic conditions.
  2. [Experimental analysis] No sensitivity analysis or ablation is reported on communication parameters (bandwidth, latency, or consensus timeout), so it is unclear whether the measured advantage over the centralized baseline survives when these constraints are introduced at levels typical of railway radio systems.
minor comments (1)
  1. [Abstract] The abstract states that self-organization 'achieves better results' but supplies no numerical deltas, confidence intervals, or number of perturbation scenarios; these should be added to the abstract and results summary for immediate assessment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We agree that incorporating communication constraints is essential for strengthening the claims about realistic deployment. Below we address each major comment point by point.

read point-by-point responses

  1. Referee: [Experimental analysis / closed-loop framework] The experimental comparison (described in the abstract and experimental analysis) attributes the performance gain to instance decomposition and real-time consensus, yet the closed-loop simulator provides no explicit model, latency bounds, packet-loss rates, or failure modes for neighbor identification and hypothesis exchange. This assumption is load-bearing for the central claim that self-organization works under realistic conditions.

    Authors: We acknowledge that our closed-loop framework models the train dynamics and perturbation scenarios but assumes perfect communication for the self-organization process, including instantaneous neighbor identification and hypothesis exchange. This allows us to focus on the benefits of decentralized decision-making and instance decomposition. However, we recognize that this is a limitation for claiming full realism. In the revised manuscript, we will add a discussion of communication assumptions and include initial experiments with modeled delays and losses to test robustness. revision: yes

  2. Referee: [Experimental analysis] No sensitivity analysis or ablation is reported on communication parameters (bandwidth, latency, or consensus timeout), so it is unclear whether the measured advantage over the centralized baseline survives when these constraints are introduced at levels typical of railway radio systems.

    Authors: The original experiments did not include sensitivity analysis on communication parameters as the primary goal was to compare the self-organizing approach against centralized control under ideal conditions to highlight the potential of the method. We agree that this analysis is necessary. We will perform and report an ablation study varying latency, bandwidth, and consensus timeout using realistic railway communication parameters (such as those from GSM-R systems) in the revised paper. revision: yes

Circularity Check

0 steps flagged

No circularity: central claim rests on direct empirical comparison in closed-loop simulation

full rationale

The paper presents a modular self-organization process (neighbor identification, hypothesis formulation, compatibility check, consensus) and validates it via experimental comparison against a state-of-the-art centralized algorithm on a portion of the Italian network using a microscopic simulator. No equations, fitted parameters, or derivations are described that reduce by construction to the inputs. The claimed advantage from instance decomposition is presented as an empirical outcome, not a definitional or self-citation reduction. No load-bearing self-citations or ansatzes are evident in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach depends on domain assumptions about reliable local communication and computation; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (2)
  • domain assumption Trains can accurately and timely identify their current neighbors
    Required for the first step of the modular process.
  • domain assumption Local hypotheses can be formulated and checked for compatibility without global knowledge
    Core premise enabling instance decomposition.

pith-pipeline@v0.9.0 · 5482 in / 1253 out tokens · 165947 ms · 2026-05-16T14:20:37.507855+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

42 extracted references · 42 canonical work pages

  1. [1]

    Shift2rail,

    S2R, “Shift2rail,” 2015, accessed November 5th, 2024. [Online]. Available: https://rail-research.europa.eu/about-shift2rail/

    work page 2015

  2. [2]

    Europe’s rail,

    ERJU, “Europe’s rail,” 2024, accessed November 5th, 2024. [Online]. Available: https://rail-research.europa.eu/

    work page 2024

  3. [3]

    A review of online dynamic models and algorithms for railway traffic management,

    F. Corman and L. Meng, “A review of online dynamic models and algorithms for railway traffic management,”IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 3, pp. 1274–1284, 2015

    work page 2015

  4. [4]

    An overview of recovery models and algorithms for real-time railway rescheduling,

    V . Cacchiani, D. Huisman, M. Kidd, L. Kroon, P. Toth, L. Veelenturf, and J. Wagenaar, “An overview of recovery models and algorithms for real-time railway rescheduling,”Transportation Research Part B: Methodological, vol. 63, pp. 15 – 37, 2014

    work page 2014

  5. [5]

    A survey on problem models and solu- tion approaches to rescheduling in railway networks,

    W. Fang, S. Yang, and X. Yao, “A survey on problem models and solu- tion approaches to rescheduling in railway networks,”IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 6, pp. 2997–3016, 2015

    work page 2015

  6. [6]

    Lamorgese, C

    L. Lamorgese, C. Mannino, D. Pacciarelli, and J. T ¨ornquist Krasemann, Train Dispatching. Cham: Springer International Publishing, 2018, pp. 265–283

    work page 2018

  7. [7]

    A scalable reinforcement learning algorithm for schedul- ing railway lines,

    H. Khadilkar, “A scalable reinforcement learning algorithm for schedul- ing railway lines,”IEEE Transactions on Intelligent Transportation Systems, vol. 20, no. 2, pp. 727–736, 2019

    work page 2019

  8. [8]

    Distributed approximate dynamic control for trafc management of busy railway networks,

    T. Ghasempour, G. Nicholson, D. Kirkwood, T. Fujiyama, and B. Hey- decker, “Distributed approximate dynamic control for trafc management of busy railway networks,”IEEE Transactions on Intelligent Transporta- tion Systems, vol. 21, no. 9, pp. 3788–3798, 2020

    work page 2020

  9. [9]

    Artificial intelligence applied on traffic planning and management for rail transport: A review and perspective,

    J. Zhang and J. Zhango, “Artificial intelligence applied on traffic planning and management for rail transport: A review and perspective,” Discrete Dynamics in Nature and Society, vol. 2023, p. 1832501, 2023

    work page 2023

  10. [10]

    Literature review toward decentralized railway traffic management,

    E. Marcelli and P. Pellegrini, “Literature review toward decentralized railway traffic management,”IEEE Intelligent Transportation Systems Magazine, vol. 13, no. 3, pp. 234–252, 2020

    work page 2020

  11. [11]

    Large scale mathematical methods evaluation,

    X2Rail4 consortium, “Large scale mathematical methods evaluation,” deliverable 8.3 of the X2Rail-4 project, S2R-CFM-IP2-01-2019, 881806. Accessed 11/05/2024. [Online]. Available: https://cordis. europa.eu/project/id/881806/results

    work page 2019

  12. [12]

    Distributed model predictive control for train regulation in urban metro transportation,

    F. Shang, J. Zhan, and Y . Chen, “Distributed model predictive control for train regulation in urban metro transportation,” in2018 21st Interna- tional Conference on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1592–1597

    work page 2018

  13. [13]

    Decentralized, autonomous train dispatching using swarm intelligence in railway operations and control,

    C. Yong, M. Ullrich, and L. Jiajian, “Decentralized, autonomous train dispatching using swarm intelligence in railway operations and control,” in7th International Conference on Railway Operations Modelling and Analysis - RailLille2017. IAROR, 2017, pp. 521–540

    work page 2017

  14. [14]

    Towards a conflict prevention strategy applicable for real-time railway traffic management,

    S. Van Thielen, F. Corman, and P. Vansteenwegen, “Towards a conflict prevention strategy applicable for real-time railway traffic management,” Journal of Rail Transport Planning & Management, vol. 11, p. 100139, 2019

    work page 2019

  15. [15]

    Self-organized rail traffic for the evolution of decentralized mobility,

    SORTEDMOBILITY , “Self-organized rail traffic for the evolution of decentralized mobility,” 2021, accessed October 20th, 2022. [Online]. Available: https://www.sortedmobility.eu/

    work page 2021

  16. [16]

    To- wards self-organizing railway traffic management: concept and frame- work,

    L. D’Amato, F. Naldini, V . Tibaldo, V . Trianni, and P. Pellegrini, “To- wards self-organizing railway traffic management: concept and frame- work,”Journal of Rail Transport Planning and Management, vol. 29, p. 100427, 2020

    work page 2020

  17. [17]

    RECIFE- MILP: an effective MILP-based heuristic for the real-time railway traffic management problem,

    P. Pellegrini, G. Marli `ere, R. Pesenti, and J. Rodriguez, “RECIFE- MILP: an effective MILP-based heuristic for the real-time railway traffic management problem,”Intelligent Transportation Systems, IEEE Transactions on, vol. 16, no. 5, pp. 2609–2619, 2015

    work page 2015

  18. [18]

    Self-organized structures in a super- organism: do ants “behave

    C. Detrain and J.-L. Deneubourg, “Self-organized structures in a super- organism: do ants “behave” like molecules?”Physics of Life Reviews, vol. 3, no. 3, pp. 162 – 187, 2006

    work page 2006

  19. [19]

    Stop Signals Provide Cross Inhibition in Collective Decision-Making by Honeybee Swarms,

    T. D. Seeley, P. K. Visscher, T. Schlegel, P. M. Hogan, N. R. Franks, and J. A. R. Marshall, “Stop Signals Provide Cross Inhibition in Collective Decision-Making by Honeybee Swarms,”Science, vol. 335, no. 6064, pp. 108 – 111, 2012

    work page 2012

  20. [20]

    Swarm Robotics: Past, Present, and Future,

    M. Dorigo, G. Theraulaz, and V . Trianni, “Swarm Robotics: Past, Present, and Future,”Proceedings of the IEEE, vol. 109, no. 7, pp. 1152–1165, 2021

    work page 2021

  21. [21]

    Cognitive network architecture systems to provide intelligent services: An intelligent self-organization approach with a game-based incentive mechanism,

    Y . Liu, J. Gui, and N. Xiong, “Cognitive network architecture systems to provide intelligent services: An intelligent self-organization approach with a game-based incentive mechanism,”IEEE Systems, Man, and Cybernetics Magazine, vol. 9, no. 1, pp. 25–36, 2023

    work page 2023

  22. [22]

    Self-organizing software models for the internet of things: Complex software structures that emerge without a central controller,

    D. Arellanes, “Self-organizing software models for the internet of things: Complex software structures that emerge without a central controller,” IEEE Systems, Man, and Cybernetics Magazine, vol. 7, no. 3, pp. 4–9, 2021

    work page 2021

  23. [23]

    Individual rules for trail pattern formation in argentine ants (linepithema humile),

    A. Perna, B. Granovskiy, S. Garnier, S. C. Nicolis, M. Lab ´edan, G. Theraulaz, V . Fourcassi´e, and D. J. T. Sumpter, “Individual rules for trail pattern formation in argentine ants (linepithema humile),”PLOS Computational Biology, vol. 8, no. 7, pp. 1–12, 2012

    work page 2012

  24. [24]

    Self-organized traffic via priority rules in leaf-cutting ants,

    D. Str ¨ombom and A. Dussutour, “Self-organized traffic via priority rules in leaf-cutting ants,”PLOS Computational Biology, vol. 14, no. 10, pp. 1–13, 2018

    work page 2018

  25. [25]

    Traffic instabilities in self-organized pedestrian crowds,

    M. Moussa ¨ıd, E. G. Guillot, M. Moreau, J. Fehrenbach, O. Chabiron, S. Lemercier, J. Pettr ´e, C. Appert-Rolland, P. Degond, and G. Ther- aulaz, “Traffic instabilities in self-organized pedestrian crowds,”PLOS Computational Biology, vol. 8, no. 3, pp. 1–10, 2012

    work page 2012

  26. [26]

    Mutual an- ticipation can contribute to self-organization in human crowds,

    H. Murakami, C. Feliciani, Y . Nishiyama, and K. Nishinari, “Mutual an- ticipation can contribute to self-organization in human crowds,”Science Advances, vol. 7, no. 12, p. eabe7758, 2021

    work page 2021

  27. [27]

    Helbing, S

    D. Helbing, S. L ¨ammer, and J.-P. Lebacque,Self-Organized Control of Irregular or Perturbed Network Traffic. Boston, MA: Springer US, 2005, pp. 239–274

    work page 2005

  28. [28]

    Gershenson,Self-Organizing Urban Transportation Systems

    C. Gershenson,Self-Organizing Urban Transportation Systems. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012, pp. 269–279

    work page 2012

  29. [29]

    Agent-based frame- work for self-organization of collective and autonomous shuttle fleets,

    A. Bucchiarone, M. De Sanctis, and N. Bencomo, “Agent-based frame- work for self-organization of collective and autonomous shuttle fleets,” IEEE Transactions on Intelligent Transportation Systems, vol. 22, no. 6, pp. 3631–3643, 2021

    work page 2021

  30. [30]

    Towards self-organizing lo- gistics in transportation: a literature review and typology,

    B. Gerrits, W. van Heeswijk, and M. Mes, “Towards self-organizing lo- gistics in transportation: a literature review and typology,”International PREPRINT 11 Transactions in Operational Research, vol. 31, no. 3, pp. 1309–1374, 2024

    work page 2024

  31. [31]

    Efficient self-organization of informal public transport networks,

    K. M. Mittal, M. Timme, and M. Schr ¨oder, “Efficient self-organization of informal public transport networks,”Nature Communications, vol. 15, no. 1, p. 4910, 2024

    work page 2024

  32. [32]

    When slower is faster,

    C. Gershenson and D. Helbing, “When slower is faster,”Complexity, vol. 21, no. 2, pp. 9–15, 2015

    work page 2015

  33. [33]

    A review of real-time railway and metro rescheduling models using learning algorithms,

    M. Jusup, A. Trivella, and F. Corman, “A review of real-time railway and metro rescheduling models using learning algorithms,” in21st Swiss Transport Research Conference (STRC 2021), 2021, p. 27 p

    work page 2021

  34. [34]

    Flatland-RL: Multi- agent reinforcement learning on trains,

    S. Mohanty, E. Nygren, F. Laurent, M. Schneider, C. Scheller, N. Bhattacharya, J. Watson, A. Egli, C. Eichenberger, C. Baumberger, G. Vienken, I. Sturm, G. Sartoretti, and G. Spigler, “Flatland-RL: Multi- agent reinforcement learning on trains,” 2020

    work page 2020

  35. [35]

    Pachl,Railway Operation and Control

    J. Pachl,Railway Operation and Control. VTD Rail Publishing, 2002

    work page 2002

  36. [36]

    The ON-TIME real-time railway traffic management framework: A proof-of-concept using a scalable standard- ised data communication architecture,

    E. Quaglietta, P. Pellegrini, R. Goverde, T. Albrecht, B. Jaekel, G. Marli `ere, J. Rodriguez, T. Dollevoet, B. Ambrogio, D. Carcasole, M. Giaroli, and G. Nicholson, “The ON-TIME real-time railway traffic management framework: A proof-of-concept using a scalable standard- ised data communication architecture,”Transportation Research Part C: Emerging Techn...

    work page 2016

  37. [37]

    Closing the loop in real-time railway control: Framework design and impacts on operations,

    F. Corman and E. Quaglietta, “Closing the loop in real-time railway control: Framework design and impacts on operations,”Transportation Research Part C: Emerging Technologies, vol. 54, pp. 15–39, 2015

    work page 2015

  38. [38]

    Decentralised multi-agent coordination for real-time railway traffic management,

    L. D’Amato, P. Pellegrini, and V . Trianni, “Decentralised multi-agent coordination for real-time railway traffic management,”arXiv preprint arXiv:2502.08324, 2025

    work page arXiv 2025

  39. [39]

    Railroad simulation using OpenTrack,

    A. Nash and D. Huerlimann, “Railroad simulation using OpenTrack,” inComputers in Railways IX, C. Brebbia, J. Allan, G. Sciutto, and S. Scone, Eds. WIT Press, Southampton, United Kingdom, 2004, pp. 45–54

    work page 2004

  40. [40]

    Simulation- optimization interface format,

    COSYS-ESTAS Universit ´e Gustave Eiffel, “Simulation- optimization interface format,” 2024, accessed November 5th,

    work page 2024

  41. [41]

    Available: http://recife.univ-eiffel.fr/sharedData/ SimulationOptimization interface format/

    [Online]. Available: http://recife.univ-eiffel.fr/sharedData/ SimulationOptimization interface format/

    work page

  42. [42]

    Assessment of self-organizing railway operations and recommendations,

    SORTEDMOBILITY consortium, “Assessment of self-organizing railway operations and recommendations,” deliverable 5.2 of the SORTEDMOBILITY project, grant agreement N 875022. Accessed 11/05/2024. [Online]. Available: https://www.sortedmobility. eu/fileadmin/contributeurs/SORTEDMOBILITY/Files/WP5/D52 AssessmentOfSelf-OrganizingRailwayOperationsAndRecommendati...

    work page 2024