🤖 AI

A Temporal Planning Framework for Disruption Aware Dynamic Route Optimization in Heterogeneous Railway Systems

This paper proposes a PDDL 2.1-based temporal planning framework that dynamically optimizes routes and manages disruptions in heterogeneous, multi-gauge railway systems by generating conflict-free, executable operational plans to reduce reliance on manual decision-making and enhance safety.

Original authors: Pollob Chandra Ray, Sabah Binte Noor, Fazlul Hasan Siddiqui

Published 2026-06-15

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Pollob Chandra Ray, Sabah Binte Noor, Fazlul Hasan Siddiqui

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a railway system not as a simple set of tracks, but as a massive, complex puzzle where every piece is a different shape and size. Some pieces are wide, some are narrow, and some are double-wide. Now, imagine you have to move dozens of different trains across this puzzle, but they can only fit on tracks that match their specific "shoe size" (their gauge). To make things harder, the puzzle pieces sometimes get stuck, the trains break down, or the weather forces them to crawl instead of run.

This paper presents a new "smart brain" (a framework) designed to solve this puzzle automatically, even when chaos strikes. Here is the breakdown in simple terms:

1. The Problem: A Chaotic Puzzle

Railway systems are often a mix of different track widths (like in Spain or Bangladesh). A train built for a wide track cannot run on a narrow one, and vice versa.

The Old Way: Humans try to figure out the best route and schedule. When a track gets blocked or a train breaks down, humans have to panic and make quick decisions. The paper notes that in places like Bangladesh, human error causes over 50% of operational accidents.
The Gap: Previous computer programs were good at making a "timetable" (a list of when trains should leave), but they didn't tell the workers exactly what to do step-by-step (like flipping a switch to change a track). They left the dangerous, detailed work to humans.

2. The Solution: The "Time-Traveling Chef"

The authors created a system called DART (Disruption Aware Railway Temporal Planning). Think of this system as a highly organized chef who doesn't just write a menu (the schedule) but also gives a second-by-second recipe for cooking the meal.

Temporal Planning: Instead of just saying "Train A goes to Station B," the system says, "At 10:00 AM, Train A leaves. At 10:01 AM, the switch at the junction flips. At 10:02 AM, Train B can start moving." It manages time and space simultaneously.
The "Shoe Size" Rule: The system is programmed to know that a "wide shoe" train can only walk on "wide floor" tracks. It automatically checks this rule before planning any move.
Handling Disasters: The system is trained to expect four types of "kitchen fires":
1. Blocked Track: A road is closed. The system instantly finds a detour or tells the train to wait.
2. Blocked Train: A train is stuck. The system tells other trains to go around it.
3. Slowdown: Bad weather makes trains slow. The system recalculates the time it takes to get there.
4. Engine Failure: A train breaks. The system automatically schedules a "tow truck" (an auxiliary engine) to hook up and pull it.

3. The Test: The "Stress Test" Gym

To prove this "smart brain" works, the researchers didn't just test it on a small toy set. They built a massive digital gym with 200 different scenarios:

Small to Huge: They started with tiny networks and scaled up to massive ones with 1,000 track points and 120 trains.
Nominal vs. Chaos: They tested it when everything was perfect (Nominal) and when everything was going wrong (Disrupted).
The Judges: They used two top-tier AI planners (named POPF and OPTIC) to act as the judges. They also used a "validator" (like a referee) to check if the AI's plan was legal and safe.

4. The Results: A Flawless Performance

The results were impressive:

Perfect Safety: The AI generated plans that were 100% conflict-free. No two trains tried to be in the same place at the same time.
Handling the Chaos: Even when the network was huge and full of breakdowns, the system kept working. It successfully solved 99 out of 100 of the hardest, most chaotic scenarios.
The Main Culprit: The data showed that the biggest cause of delays wasn't blocked tracks or broken trains, but slowdowns (like bad weather). The system handled these by adjusting the timing, proving it could adapt to real-world unpredictability.
Speed: The system could generate these complex, second-by-second plans in seconds or minutes, even for the largest networks.

The Bottom Line

This paper introduces a way to turn railway management from a "guessing game" into a precise, automated science. By treating train routes like a complex, time-sensitive puzzle, the system creates a detailed, step-by-step instruction manual that ensures trains arrive on time, stay safe, and know exactly what to do when things go wrong. It removes the guesswork from the hands of human operators and puts it into a reliable, mathematical framework.

Technical Summary: A Temporal Planning Framework for Disruption-Aware Dynamic Route Optimization in Heterogeneous Railway Systems

1. Problem Statement

Railway operations in heterogeneous, multi-gauge networks face significant challenges in ensuring safety and punctuality. Unlike uniform metro systems, these networks involve trains with varying speeds, stopping patterns, and critical infrastructure constraints, specifically gauge compatibility (e.g., meter gauge vs. broad gauge trains cannot run on incompatible tracks). These constraints severely limit route options and increase coordination complexity.

The problem is further exacerbated by stochastic disruptions, including blocked tracks, immobilized trains, engine failures, and speed slowdowns. Existing research predominantly focuses on high-level timetabling, often omitting low-level operational details such as track switching (turnout) coordination. Consequently, critical decisions are left to human operators, increasing the risk of safety incidents. Furthermore, many existing optimization models (e.g., MaxSAT, MILP, metaheuristics) assume homogeneous train characteristics or fail to generate executable, timestamped action sequences, limiting their utility in real-time disruption management.

2. Methodology

The authors propose the Disruption Aware Railway Temporal Planning (DART) framework, which models railway operations as a temporal planning problem using PDDL 2.1 (Planning Domain Definition Language).

2.1 Domain Modeling

The framework encodes the railway environment using:

Object Types: track-point, station, train, engine, and gauge-type.
Predicates: These capture spatial configurations (e.g., train-at), infrastructure connectivity (connected, track-accessible), gauge compatibility (train-gauge, track-gauge), and disruption states (track-blocked, engine-damaged).
Numeric Functions: Used to model dynamic parameters such as distance, train-speed, slowdown factors, and boarding-time.
Durative Actions: The core of the model consists of ten durative actions representing:
- Movement: drive-train and drive-engine (handling gauge compatibility and track clearance).
- Service: board-passengers.
- Disruption Recovery: attach-engine (coupling auxiliary engines to damaged trains), drive-assisted-train, resolve-train-blockage, and clear-blocked-track.
- Infrastructure: turnout (switching track alignment).

2.2 Disruption Taxonomy

The framework explicitly models four disruption types with specific recovery strategies:

Blocked Track: Resolved by waiting or rerouting via gauge-compatible paths.
Blocked Train: The train remains immobilized until resolved; other trains are rerouted.
Slowdown: Travel time is recalculated based on a slowdown factor ( $\alpha$ ), or the train is rerouted.
Engine Failure: An auxiliary engine is dispatched, attached, and used to resume movement.

2.3 Benchmark Generation

To evaluate the framework, the authors constructed a 200-instance benchmark divided into:

100 Nominal Instances: Standard operations without disruptions.
100 Disrupted Instances: Introducing blocked tracks, blocked trains, slowdowns, and engine failures.
The instances scale monotonically across four categories (Small, Medium, Large, Very Large), ranging from 3 to 120 trains and up to 1,000 track points.

3. Key Contributions

Temporal Planning Domain for Heterogeneous Railways: A novel PDDL 2.1 domain that explicitly encodes gauge compatibility constraints, diverse train profiles, and low-level operational actions (turnouts, boarding).
Integrated Disruption Management: The framework handles four distinct disruption types within the planning process, generating executable recovery plans rather than just adjusted timetables.
Scalable Benchmark Dataset: A comprehensive set of 200 problem instances designed to test scalability and robustness under increasing network density and disruption severity.

4. Experimental Results

The framework was evaluated using two state-of-the-art temporal planners, POPF and OPTIC, and validated using the VAL plan validator.

Plan Validity: All 199 successfully generated plans (out of 200) were verified by VAL as semantically correct, satisfying all temporal constraints, preconditions, and goal conditions.
Planner Performance: Both planners produced identical makespan, plan length, and delay values across all solved instances. This suggests the solution space is tightly constrained by gauge and safety rules, leading both planners to converge on the same optimal protocols.
Scalability:
- Nominal Operations: Makespan remained stable (~34 minutes) across scales, while plan length grew proportionally with network size.
- Disrupted Operations: Metrics increased monotonically with problem scale. The single failure occurred in the largest configuration (120 trains, 1,000 points, 108 concurrent disruptions) which hit the 30-minute time limit.
Delay Analysis:
- Slowdowns were the primary contributor to total delay (60.1%), followed by engine failures (30.4%) and blockages (9.5%).
- Spearman correlation analysis confirmed perfect positive correlations ( $\rho = 1.0$ ) between system size metrics and computation time, and strong correlations ( $\rho \ge 0.976$ ) between disruption intensity and total delay.

5. Significance and Claims

The paper claims that the DART framework offers a scalable, verifiable solution for enhancing the safety and efficiency of heterogeneous railway systems. By shifting from high-level timetabling to low-level, timestamped operational planning, the framework reduces dependence on manual decision-making for critical tasks like track switching.

The authors position this work as a practical foundation for intelligent railway management, demonstrating that temporal planning can generate feasible, safety-compliant strategies even in large-scale, disruption-prone environments. The study concludes that while the current framework assumes deterministic disruption durations, future work aims to extend it to support plan repair under uncertainty and integration with real-time data streams.

Note: The paper does not claim to solve the NP-hard nature of the general problem but demonstrates that temporal planners can effectively handle specific, constrained instances of heterogeneous railway routing within practical time limits for the tested scales.