A Temporal Planning Framework for Disruption Aware Dynamic Route Optimization in Heterogeneous Railway Systems
Source: arXiv:2606.14582 · Published 2026-06-12 · By Pollob Chandra Ray, Sabah Binte Noor, Fazlul Hasan Siddiqui
TL;DR
This paper addresses the complex problem of dynamic route optimization and disruption-aware scheduling in heterogeneous multi-gauge railway systems. Such systems feature trains with varying speeds, stopping patterns, and track gauge compatibility constraints that markedly increase operational complexity, especially on single-track networks requiring frequent track switching. The existing literature largely focuses on high-level timetable generation without fully modeling operational actions like turnout switching, leading to reliance on human operators and associated safety risks. To overcome this gap, the authors propose a temporal planning framework, named DART, that encodes railway operations including gauge constraints and disruption scenarios into PDDL 2.1. This model enables generation of detailed, conflict-free, timestamped operational plans specifying train routes, turnout operations, and recovery actions in presence of stochastic disruptions such as blocked tracks, blocked trains, engine failure, and slowdowns.
The framework is evaluated on a novel benchmark set of 200 instances reflecting varied network sizes—up to 1,000 track points and 120 trains—with both nominal and disrupted scenarios. Using two state-of-the-art temporal planners, POPF and OPTIC, coupled with plan validation via the VAL tool, experimental results demonstrate scalable and efficient generation of executable plans that respect gauge compatibility and ensure safe disruption management. This work substantially extends prior literature by integrating low-level operational details with disruption recovery in heterogeneous railway contexts, reducing dependency on human intervention and enhancing safety and punctuality.
Key findings
- DART framework encodes gauge compatibility constraints explicitly, allowing only feasible trains-to-track assignments within multi-gauge networks.
- Four disruption types are modeled: blocked tracks, blocked trains, engine failures, and speed slowdowns with respective recovery strategies.
- Benchmark dataset contains 200 temporal planning instances with up to 1,000 track points and 120 trains, across nominal and disrupted scenarios.
- POPF and OPTIC planners generate conflict-free temporal plans in all benchmark instances, with makespan scaling predictably as network size and disruption severity increase (Fig 5).
- Temporal plans include detailed low-level actions with precise timestamps, e.g., turnout operations requiring 1-minute durations and boarding times modeled explicitly.
- Plan validation via VAL confirms all generated plans are executable and respect temporal and resource constraints.
- The approach reduces reliance on manual decision making in disrupted railway operations, addressing a gap noted in human error incident statistics (50.77% in Bangladesh Railways).
- The temporal planning formulation models heterogeneous speeds and stopping requirements per train, improving realism over homogeneous assumptions common in prior optimization work.
Threat model
The adversary is modeled as stochastic disruptions—blocked tracks, blocked trains, engine failures, and speed slowdowns—that randomly occur during operations. These events impede normal schedule adherence by blocking resources or reducing speeds. The model assumes disruptions have known durations and are observable by the system. There is no consideration of malicious attackers or partial observability; disruptions are environmental and nondeliberate.
Methodology — deep read
The threat model considers disruptions in heterogeneous multi-gauge railway networks, specifically blocked tracks, blocked trains, engine failures, and speed slowdowns. The system is assumed fully observable and deterministic, with known durations of disruptions and trains following constant speed kinematics for tractability. The adversary is environmental stochastic disruptions, with no deliberate sabotage assumed.
Data comprises a new benchmark problem set created by the authors with 200 temporal planning instances: 100 nominal and 100 with disruptions. Instances scale from small to very large networks (up to 1,000 track points, 120 trains), incorporating gauge diversity, track connectivity, and disruption types. Each instance defines initial positions, destinations, gauge assignments, and disruption statuses.
The core algorithmic framework models the railway system in PDDL 2.1, capturing dynamic components through durative actions with temporal constraints. The domain includes typed objects (train, engine, gauge-type, track-point, station), predicates encoding spatial and infrastructure state (e.g., train-at, track-gauge, connected), and gauge compatibility constraints enforce correct matching of trains to compatible tracks. Disruption predicates represent blocked tracks/trains, slowdowns, and engine failure states with associated durative recovery or rerouting actions.
Durative actions with temporal preconditions and effects model movement, turnout switching (requiring 1 minute), boarding, and auxiliary engine coupling. Actions have start, over-all, and end conditions ensuring safety and consistency. The framework optimizes makespan while ensuring a conflict-free plan.
For evaluation, the authors use two state-of-the-art temporal planners, POPF and OPTIC, to generate plans, applying the VAL plan validator to ensure soundness. Experiments vary network size and disruption severity, analyzing scalability and robustness to disruptions.
Reproducibility is facilitated by the detailed benchmark dataset, domain and problem PDDL files, and plan validation tools. However, the code and planners themselves are existing tools without custom implementations disclosed. The seed strategies and hyperparameter tuning for planners are not explicitly detailed, but standard usage is implied.
A concrete example is given using a six-point network with dual-gauge tracks and two trains of different gauges needing to cross in opposite directions with boarding stops and turnout switching. A temporal plan assigns start times for each action, respecting gauge constraints and turnout operations. This illustrates how the framework generates detailed executable plans beyond high-level timetabling.
In summary, the methodology systematically integrates heterogeneous infrastructure modeling, stochastic disruption representation, and temporal planning to produce low-level actionable operational plans enabling safer, automated railway route optimization and recovery.
Technical innovations
- Encoding multi-gauge railway infrastructure with explicit gauge compatibility predicates in a temporal planning domain.
- Integration of stochastic disruption types (blocked tracks/trains, engine failures, speed slowdowns) within the temporal planning framework for dynamic rescheduling.
- Generating low-level executable operational plans including turnout switching, passenger boarding, and auxiliary engine attachment as timed durative actions.
- Development of a large-scale, disruption-aware railway planning benchmark with up to 1,000 track points and 120 trains for evaluating temporal planners in realistic heterogeneous settings.
Datasets
- DART Railway Temporal Planning Benchmark — 200 instances — proprietary dataset created by authors, includes small to very large network scenarios with disruptions
Baselines vs proposed
- POPF planner: success rate = 100% on benchmark vs OPTIC planner: success rate = 100%
- Makespan scales predictably as network size increases; no clear winner between POPF and OPTIC (Fig 5)
- Plans validated with VAL: 100% plans executable and conflict-free vs no baseline—baseline methods do not generate executable low-level plans
Figures from the paper
Figures are reproduced from the source paper for academic discussion. Original copyright: the paper authors. See arXiv:2606.14582.
Fig 1: Railway network with six track points (A–F). Circles denote stations and diamonds denote
Fig 2 (page 4).
Limitations
- Assumes full observability and deterministic disruption durations, limiting realism in unpredictable or partially known environments.
- Trains modeled with fixed speeds and without acceleration/deceleration dynamics for computational simplicity.
- Disruption types limited to blocked tracks/trains, engine failure, and slowdowns; does not consider other operational issues like signaling failures or crew availability.
- No adversarial or worst-case disruption evaluation reported, only stochastic scenarios synthetically generated.
- The framework relies on existing temporal planners without custom optimizations for very large-scale or highly congested networks.
- Code and dataset not publicly released at time of writing, limiting immediate reproducibility.
Open questions / follow-ons
- How would the framework perform under partial observability or uncertain disruption durations?
- Can acceleration/deceleration dynamics be incorporated without excessive computational overhead?
- How to extend disruption modeling to incorporate broader operational disturbances such as signaling faults or crew availability?
- What is the impact of adversarial or worst-case disruption sequences on plan robustness and recovery?
Why it matters for bot defense
For bot-defense and CAPTCHA practitioners, this paper illustrates how temporal planning can be applied to complex resource-constrained scheduling problems subject to real-time disruptions. Although the domain is railway systems, similar complexity arises in orchestrating multi-step, time-sensitive defense workflows involving heterogeneous resources and unpredictable adversarial events. The approach of encoding detailed operational constraints and recovery actions as temporally timestamped plans may inspire analogous frameworks in bot-defense for dynamic challenge sequencing or attack mitigation pathways. The disruption taxonomy and encoding of recovery strategies also provide insights into designing adaptive defense mechanisms that maintain safety and continuity under adverse conditions. However, integration with machine learning approaches common in bot detection remains an open direction. Overall, the paper demonstrates the value of combining domain-specific constraints with automated temporal planning for robust, low-level operational control.
Cite
@article{arxiv2606_14582,
title={ A Temporal Planning Framework for Disruption Aware Dynamic Route Optimization in Heterogeneous Railway Systems },
author={ Pollob Chandra Ray and Sabah Binte Noor and Fazlul Hasan Siddiqui },
journal={arXiv preprint arXiv:2606.14582},
year={ 2026 },
url={https://arxiv.org/abs/2606.14582}
}