Efficiently operate trains in urban areas with a railway control system that utilizes AI

As cities continue to expand, rail is expected to become a key component of urban mobility systems. Compared to automobiles and air transport, rail is highly energy efficient and has a relatively low environmental impact, making it an important element of sustainable transport systems. Therefore, researchers have investigated advanced train control technologies to further improve operational efficiency, passenger comfort, and punctuality while reducing energy consumption.

In this context, Professor Masashi Miyatake from the Faculty of Engineering and Applied Sciences collaborated with Mingyu Liu, a doctoral student from the Sophia University Graduate School of Science and Engineering, to research a new AI-based framework for speed control of urban railways. The results of this study were published in Volume 14. IEEE access March 25, 2026.

Reinforcement learning is a field of artificial intelligence (AI) that has emerged as a promising approach for autonomous train operations, as it allows AI agents to learn optimal control strategies by repeatedly interacting with the environment and improving their behavior through trial and error. However, implementing this into rail operations poses several practical challenges, including incomplete information about operating conditions, train inertia, braking delays, and stringent safety requirements.

To overcome this, researchers developed a recurrent reinforcement learning framework based on the Recurrent Soft Actor-Critic (RSAC) algorithm. RSAC is an AI technique that allows train control systems to learn from past driving patterns and adapt to changing railway conditions over time.

“Our reinforcement learning-based algorithms were able to adapt train operation control to a wide range of track and vehicle conditions. ” Professor Miyatake explains:

Unlike traditional reinforcement learning methods, the proposed approach used a recurrent neural network that can retain information from previous train states. This allows the system to understand time-based relationships related to train operations, such as traction response delays, braking history, and inertia effects, improving decision-making in partially observable conditions. The team also used a training approach that allows the AI ​​system to first learn from examples of expert driving behavior before learning on its own. By learning these optimized driving patterns early in training, the AI ​​is now able to learn faster, make better decisions, and develop more stable and efficient train control behaviors.

Apart from this, the team has also integrated safety filters into the framework. Safety filters override potentially unsafe control commands generated by AI policies and ensure compliance with operational constraints such as speed limits and braking feasibility. This mechanism helped ensure safe operation even when the learned policy encountered unfamiliar situations.

The proposed framework was evaluated through various simulations of urban rail operations over a section of about 2 kilometers between stations. The simulation environment is designed to closely resemble real-world railway conditions, including uphill and downhill tracks, varying speed limits, and train arrival times.

Furthermore, the proposed AI framework was also compared with several other widely used reinforcement learning algorithms. These include Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO), Twin Delayed Deep Deterministic Policy Gradient (TD3), and Soft Actor-Critic (SAC), which are commonly used to train autonomous decision-making systems.

“Among all the methods tested, the framework we developed learned faster and performed better than all the methods tested.” explains Professor Miyatake.

The study also demonstrated superior energy efficiency. Although dynamic programming achieved the theoretical minimum energy consumption under ideal conditions, the developed system achieved comparable performance while remaining suitable for real-time operation. On the other hand, traditional reinforcement learning approaches result in unstable behavior and high energy consumption.

The findings suggest that this framework could help introduce cleaner and more convenient public transport by promoting accurate and energy-efficient train operations at relatively low cost. This research highlights the benefits of rail as an energy-efficient and environmentally friendly transport system that contributes to a sustainable future. In the future, this research could contribute to strengthening the role of railways as a sustainable transport system, supporting climate change mitigation and sustainable urban development.

The findings also highlight the potential of AI-powered rail systems to support safer, greener, and more reliable urban transportation in future smart cities.

About Sophia University

Founded in 1913 as a private Jesuit-affiliated university, Sophia University is one of the most prestigious universities located in the heart of Tokyo. Sophia University offers education through 29 departments in 9 undergraduate schools, 25 majors in 10 graduate schools, and enrolls more than 13,000 students from around the world.

Based on the spirit of “For Others, With Others,” Sophia University truly values ​​internationalism and love for one’s neighbor, and conducts education and research that transcends national, linguistic, and academic boundaries. Sophia emphasizes the need for interdisciplinary and interdisciplinary research to find solutions to the most pressing global problems such as climate change, poverty, conflict and violence. Over the past century, Sofia has made a dedicated effort to develop forward-looking graduates who contribute their talents and learning for the benefit of others, “bringing the world together” and paving the way for a sustainable future.

Website: https://www.sophia.ac.jp/

About Sophia University Professor Masashi Miyatake

Dr. Masafumi Miyatake is a professor and dean of the Faculty of Applied Science, Faculty of Engineering, Sophia University. He received his bachelor’s, master’s and doctoral degrees in electrical engineering from the University of Tokyo. His research focuses on energy management, renewable energy, power conversion, and transportation electrification, especially for rail and electric vehicles. He has published more than 300 research papers to date and is known for his work on smart and energy-efficient transportation systems. IEEE member and IEEJ senior member.