At the EMVA Conference in Rome, amid discussions of vision systems and industrial automation, Ruben Schilling of Deutsche Bahn delivered one of the most compelling glimpses into the future of driverless, automated transportation. His talk wasn’t just about machine vision—it was about rethinking how trains move, who—or what—operates them, and what it will take to automate an entire railway system.
A Crisis and an Opportunity
Schilling began by outlining a growing problem: train drivers are becoming scarce. The profession is struggling to attract new recruits, and demographic shifts aren’t helping. Meanwhile, governments across Europe are pushing to shift more traffic from roads to rail in pursuit of environmental and efficiency goals.
The solution? Full automation. But the implications extend far beyond simply removing the human driver. “With automation,” Schilling explained, “you’re no longer tied to human schedules. You can increase the frequency of trains, improve service flexibility, and even redesign operational workflows from scratch.”
The Tech Behind the Tracks
Transitioning to fully autonomous trains is no small feat. It involves a host of technologies borrowed from robotics and adapted to the uniquely complex world of railways. These include real-time localization, machine vision, and AI-driven operational systems. Traditional rail infrastructure also plays a role—specifically digital interlockings and next-generation control centers.
Schilling shared a breakdown of what makes train automation uniquely challenging: varying weather and light conditions, limited sensor range, complex rail geometries, and the need for absolute safety in life-critical applications.
“You need very high-performance detection to avoid false positives—like unnecessary braking—but also to respond instantly to real threats,” he said.
Machine Vision: More Than Just Driving
Schilling’s team has been researching automated rail technology for nearly a decade. Along the way, they’ve explored a wide array of use cases beyond just driving. From vegetation management and crowd control, to obstacle detection, smoke and fire alerts, and post-collision analysis, machine vision systems are being trained to handle the many scenarios railways encounter.
In one project, for example, Schilling’s group developed a sensor fusion system using infrared, gas, and RGB cameras to detect fire and smoke along rail corridors. “Grass, oil, even discarded wood can ignite near the tracks,” he noted, “and early detection is essential.”
Another experiment involved post-collision detection—recording and analyzing impacts that occur during train operation. Using a mix of accelerometers, strain gauges, and cameras, the team simulated and captured various overrun events. “We wanted to match human driver performance,” Schilling said. “And in some areas, we’re getting very close.”
Safety and the Machine Learning Dilemma
Despite the buzz around AI and machine learning, Schilling cautioned that its role in safety-critical systems like rail remains limited—for now. “In railway environments, you can’t rely on statistical evidence alone,” he explained. “You need causality and traceability.”
That’s why his team takes a hybrid approach: combining traditional physical models with AI when possible, but ensuring every decision made by a system can be explained and verified. “We’re not just engineering machines. We’re engineering trust.”
Building Towards Certification
The team has moved beyond theory and prototypes. From early internal test platforms to public demonstrators like Sensus F, developed with Bosch and Siemens, each iteration has brought new levels of functionality and realism. The current phase is a full-fledged research project aimed at achieving the highest safety integrity levels for obstacle detection and other critical functions.
Using off-the-shelf sensors—LiDARs, cameras, ultrasonic units—the team retrofitted two distinct train platforms to demonstrate compatibility and data-sharing potential across manufacturers. Schilling emphasized this as a key step in building a scalable, fleet-wide ecosystem.
Simulating the Future
To support certification, Deutsche Bahn has also developed a full sensor simulation environment based on EMVA standards. This “virtual rail” allows for rigorous testing of rare or dangerous scenarios without needing thousands of real-world kilometers. The simulation includes everything from physics-based sensor models to scenario scripting tools.
“Testing edge cases—like a deer crossing the tracks—is nearly impossible in real life at scale,” Schilling said. “Simulation makes certification achievable.”
What’s Next?
Currently at Technology Readiness Level 6, Schilling estimates it will take another 3–5 years before such a system could be operational. The project is already engaging with Germany’s National Railway Safety Agency and certified assessors to build its safety case.
But don’t expect to see these trains replacing drivers on high-speed lines just yet. For now, the team is focused on lower-speed applications (up to 40 km/h), such as yard operations and route provisioning. “Human drivers themselves can’t always react fast enough at high speeds,” Schilling remarked, “so paradoxically, automation could eventually become safer even at high speeds—but that’s a topic for the future.”
A Global Effort
While Deutsche Bahn leads this effort in Germany, they’re not alone. Schilling acknowledged interest and collaboration from counterparts in France (SNCF), Denmark, Japan, and beyond. “There’s global momentum,” he said, “and the race is on to define what the world’s first truly open-environment automated train will look like.”
His closing note to the audience was simple, yet reassuring: “Despite everything I’ve said about vision systems, rest easy—today’s trains don’t rely on vision for safety. But the ones of tomorrow might. And they’ll be ready.”
