Cooperation between a human traffic manager and an AI assistant for an improved railway infrastructure resilience

Tom Rousseau (1) , Kahina Amokrane (2) , Maroua Meddeb (2) , Nicolas Renoir (1) , Mathis Brunat (1) , Mathias Fort (1) , Lucas Schott (2) , Sébastien Mahler (1) , Hervé Berthou (1)

1 SNCF
2 IRT SystemX

This article deals with railway traffic management, which includes tasks such as traffic planning, resource allocation, service adaptation and passenger information. Operators monitor the real-time movement of trains, passengers and resources, mitigate unexpected events and ensure safety. Human operators currently perform these complex tasks using their expertise. However, technical aspects of railways and concurrent disturbances increase cognitive load and biases, affecting traffic management and passenger satisfaction. To address these challenges, we propose an AI-based railway traffic manager assistant that combines Machine Learning (ML) and Human-Machine Interaction (HMI) to support operators in their daily tasks and assists them to make decisions when facing critical situations. This article outlines the design approach and introduces the initial assistant version. User-centred evaluation yields preliminary results from limited-scale experiments.

Mots clés: Railway Operations; Resilient Systems; Decision Support System; Reinforcement Learning; Human - Machine Interaction; Cognitive Workload

Domaines: Informatique [cs]

Tom Rousseau, Kahina Amokrane, Maroua Meddeb, Nicolas Renoir, Mathis Brunat, et al.. Cooperation between a human traffic manager and an AI assistant for an improved railway infrastructure resilience. Transport Research Arena (TRA2024, Apr 2024, Dublin, Ireland. ⟨hal-04547672⟩