photo: ChatGPT/Illustrative picture; generated by AI
Europe’s rail sector is testing how real-time AI can move from pilot projects to daily operations. The ERCI webinar on real-time AI applications in rail brings together two practical cases: robot-assisted tunnel surveys in Germany and edge-AI wagon identification in Belgium.
10:00 The webinar opens with an introduction to the European Railway Clusters Initiative (ERCI), which brings together 18 innovation-driven clusters across 17 European countries. The network focuses on supporting SMEs and research institutions through matchmaking, R&D collaboration, and international visibility, while also tracking emerging trends such as AI, automation, and sustainable mobility.
The agenda of the day is two real-world case studies followed by discussion moving beyond theory. The focus is not on futuristic concepts, but on what is already being tested, implemented, and scaled in the rail sector today.
10:10 The first presentation begins with LAT Group, a German company working in railway infrastructure, telecommunications systems, and digital network technologies. The speakers say that despite technological progress, railway construction remains largely manual and labour-intensive.
This becomes the starting point for innovation. AI and robotics are not introduced as abstract solutions, but as tools designed to solve very concrete problems: improving safety, reducing manual workload, and increasing precision in environments where human error can have serious consequences.
10:20 The discussion shifts to one of the sector’s most pressing challenges: labour shortages and inefficiencies in technical workflows. LAT says that high-voltage specialists, already in short supply, spend up to 60% of their time on documentation rather than technical work.
At the same time, the issue of cable damage is a critical operational and financial burden. Tens of thousands of incidents are reported annually in Germany alone, often caused by incomplete or outdated cable records. It results in costly repairs, delays, safety risks, and disruptions to rail operations, turning documentation gaps into system-wide inefficiencies.
10:35 This problem becomes the entry point for robotics. LAT explains how the idea initially seemed straightforward: combine cable detection tools with a mobile robot to reduce manual searching and improve accuracy on construction sites.
However, early attempts quickly reveal the gap between concept and reality. Robotics cannot simply be "added" to railway environments. The company turns to universities and research partners, launching a series of studies to answer fundamental questions, whether robots can move safely on tracks, respond to commands, and operate within strict railway safety regulations.
10:45 Testing moves into real railway tunnels, where the project takes a decisive step forward. The robotic platform shows its ability to navigate uneven terrain, avoid obstacles, and operate in complex environments that are often difficult or dangerous for humans. Yet the biggest challenge proves to be human, not technical. Construction teams are doubtful about adopting the system. The robot is seen as complex, inconvenient, and not truly autonomous. This becomes a turning point in the project, showing that technology adoption depends as much on usability as on performance.
In response, LAT shifts its focus towards human-robot interaction, particularly voice control. The goal is to make the robot intuitive and accessible for workers on-site, eliminating the need for constant manual control through devices.
This introduces a new layer of complexity. Railway environments are noisy, especially in tunnels, with echoes and overlapping sounds from machinery. Training the AI to recognise commands reliably becomes a major technical challenge, especially for safety-critical actions such as leaving the track. The solution involves continuous learning loops, ensuring commands are detected even in difficult acoustic conditions.
11:00 The project has now reached a stage where mobility, sensing, and detection systems are successfully integrated. The robot can carry advanced sensor equipment, operate for several hours, and identify structural issues such as tunnel cracks.
However, the focus is shifting again, from data collection to data usability. The challenge is no longer capturing information, but transforming it into clear, actionable insights that can be used directly by workers, ideally on mobile devices rather than complex systems.
11:10 During the Q&A session, LAT makes it clear that the project is still evolving. Rather than presenting a finished product, the company positions itself on a learning curve, balancing technical development with organisational change. The value of the project lies not only in potential efficiency gains, but also in closer collaboration with customers, improved understanding of infrastructure needs, and the ability to attract new talent. Robotics, in this sense, becomes part of a broader strategic transformation rather than a standalone tool.
The webinar then shifts to PHOENIX AI, introducing a completely different perspective on real-time AI in rail. While LAT focuses on infrastructure and construction, PHOENIX AI addresses data collection and operational visibility.
The company presents an edge AI solution that transforms standard CCTV cameras into intelligent sensors capable of reading wagon identifiers, container codes, and other critical information in real time. The key idea is powerful: instead of replacing infrastructure, enhance what already exists.
11:20 PHOENIX AI points out the limitations of current systems, where manual monitoring still dominates and real-time data is often unavailable. This leads to inefficiencies in tracking, scheduling, and billing, particularly when dealing with foreign wagons or containers.
The company positions edge AI as a solution to these challenges. By processing data locally, directly on-site rather than in the cloud, the system avoids latency, reduces bandwidth requirements, and ensures greater control over sensitive operational data. This also aligns with growing concerns around data sovereignty in Europe.
A real-world deployment at the Port of Antwerp-Bruges demonstrates the system in action. Using existing CCTV infrastructure, the solution achieves real-time recognition at speeds of up to 40 km/h, with accuracy levels exceeding 95%.
The system outputs structured data for each wagon and train, enabling integration into digital twins, operational platforms, and billing systems. For operators, this means a shift from fragmented, manual processes to continuous, automated visibility across the network.