GERMANY: Speaking at the Intelligent Rail Summit 2025 in Bled, Hannah Richta, Head of Algorithms for Operations at infrastructure manager DB InfraGO, updated the audience on progress with deployment of ADA-PMB, an AI-driven dispatching assistant designed to optimise train operations.
ADA-PMB combines mathematical optimisation, heuristics and machine learning to provide real-time recommendations for train dispatchers, with the aim of reducing delays and improving overall network efficiency.
Pilot deployments
Initially piloted on the Berlin S-Bahn in 2020, ADA-PMB is now active in four regions: on routes around Frankfurt-am-Main, on the München – Lindau route, on the Stuttgart – Plochingen line and in Nordrhein-Westfalen.
The tool is able to analyse train positions, assess infrastructure availability and consult operating rules, after which it makes tailored suggestions to the dispatcher. Options could include amendments to route settings, changes to the stopping patterns of trains, or looping of services to allow others to pass.
Recommendations are generated within 60 sec and delivered at 1 min intervals, ensuring a rapid response in high-conflict scenarios. Richta told the conference in Slovenia on November 5 that the tool treats all operators equally, and that rigorous checks are undertaken by the Bundesnetzagentur regulatory body to prevent bias toward DB trains.
Safety remains paramount, she said. ADA-PMB only acts as an adviser to staff, who are still able to decide whether to accept its suggestions, while existing railway safety systems carry out final checks. Accurate data input, especially on maintenance and track closures, is critical for its functionality.
Future plans
Plans for the future include integrating ADA-PMB directly into dispatching screens by December 2026, replacing the current standalone display. From 2027, DB InfraGO aims to scale the system across more of the country and implement a digital workflow for recommendation acceptance and communication, reducing reliance on phone calls.
Full functionality, including seamless integration and automated communication with signalling staff, is expected by the end of 2027.
Speaking exclusively to Railway Gazette International, Richta explained that ‘in areas where we have dispatchers — usually areas with dense traffic — the dispatcher accepts or rejects the [AI] recommendation. Then the dispatcher communicates the decision to the signaller who operates the actual switches and signals. Currently this is done by phone, but the target is to have a digital workflow for these messages.’
She added that signalling centres retain hard-wired protocols to ensure that dispatchers and signallers can never authorise an unsafe movement. ‘In some areas we do not have a separate dispatcher’, Richta said. ‘There the signaller is also the dispatcher. In these cases, like on parts of the München – Lindau line, the signaller receives the dispatching recommendations [from the AI tool], makes their decision, conducts the safety check and implements the movement authority.’
ADA-PMB takes its main data feed from the infrastructure manager’s existing dispatching tools, but this can be augmented with information about maintenance possessions from the timetable planning team, while ad-hoc updates can also be provided from regional control centres, Richta reported.
