White paper: AI-supported dispatching

• Rail traffic is growing rapidly, making disruptions harder to contain at major hubs
• To address this, T-Systems tested AI-supported dispatching through Flatland simulations of Frankfurt and Cologne rail networks, computed with and without disruptive scenarios
• The findings indicate AI-supported assistance functions can help anticipate conflicts and optimize sequencing and overtaking decisions while human overseeing remains central to rail operations

• Explore the bottlenecks impacting seamless rail dispatching
• Learn how rules, search methods, and AI can optimize dispatching decisions
• Understand why suitable evaluation standards are as important as algorithms for optimal results
• Explore near-term assistance functions for sequencing, overtaking and early conflict warnings
• Discover what it takes to move from simulation to real-world operations, with dispatchers remaining in control

The white paper leverages three approaches from the T-Brain Mobility Challenge in the Flatland simulation environment for Frankfurt and Cologne rail networks: two-stage planning with neighbourhood search, rule-based control, and coordinated multi-agent planning. It examines their recurring building blocks, performance under disruptions, evaluation metrics and the prerequisites for moving from simulation to human-in-the-loop rail operations. The findings demonstrate how AI-supported dispatching can prevent network blockages, with all simulated trains reaching their destinations in the scenarios tested, and also provide a foundation for further validation using real-world rail data, operational constraints, and robust evaluation criteria.