A groundbreaking robotic system has successfully overcome a major technical challenge in railway operations by achieving intelligent decoupling of moving trains. Developed by Sichuan Guoruan Technology Group Co., Ltd., the "decoupling robot" can follow alongside a train, extend its mechanical arm precisely between carriages, and perform decoupling actions instantly with remarkable stability and efficiency.
At the Xi'an Xinfengzhen Station, China's busiest freight train marshalling yard, a test was conducted where a freight train moved steadily at 7 kilometers per hour. A decoupling robot running in the same direction swiftly extended its mechanical arm and completed the decoupling process in under five seconds.
This highly capable robotic system, nicknamed the "Dragon-Catching Hand," was entirely developed and manufactured by Sichuan Guoruan Technology Group. The company is the first in China to break through the technical barriers of embodied intelligent technology for railway decoupling, achieving fully unmanned decoupling operations. Decoupling is considered one of the most hazardous tasks in railway marshalling yards. Couplers must work within the narrow gaps between moving carriages, manually performing 60 to 70 decoupling actions per person daily. For a long time, railway marshalling has relied on manual decoupling, a global challenge in the railway sector that has become a significant bottleneck in the development of digital and intelligent marshalling yards.
On a test track within the China Railway Rail Transit High-Tech Industrial Park in Chengdu's Jinniu District, freight carriages stand ready, connected by heavy couplers. A robotic arm precisely inserts itself into the gap between carriages, grasps, lifts, and with a "click," the coupler disengages. This area serves as the company's product training ground and pilot testing platform, where all products undergo rigorous testing before deployment.
The decoupling robot achieves a success rate of 99.8% for static decoupling, where carriages are stationary. In dynamic decoupling scenarios, the robot must run synchronously with the moving train and perform decoupling simultaneously. Critical breakthroughs have been made for the most challenging application: the hump yard embodied intelligent decoupling robot. This robotic application fundamentally eliminates the safety risks associated with manual operations.
The development of the decoupling robot began in 2016 and immediately faced significant obstacles. The controller, often called the robot's "brain," was initially entirely imported and proved poorly adaptable to the complex environments of railway marshalling yards. Decoupling tasks are varied, the recognition environment is complex, and precision requirements are extremely high. The system must accommodate different train models and adapt to various weather conditions like rain, snow, and fog, as well as challenging scenarios such as obstruction by coal dust.
The company embarked on a path of independent innovation, starting with the localization of core components. It collaborated with domestic partners to independently develop recognition systems, achieving domestic substitution for imported radar technology. By integrating AI, machine vision, large models, and other advanced technologies, the company built a multi-modal fusion perception system, enabling the robot to make autonomous decisions and perform precise operations.
The most difficult phase from R&D to practical application was real-world validation. In 2019, the company invested 30 million yuan to establish a shared public pilot platform for rail transit, becoming one of the few testing bases in China focused on "active safety + AI recognition" scenarios. This platform can simulate extreme environments like falling rocks, mudslides, heavy fog, and torrential rain, allowing hardware and algorithms to iteratively improve through rigorous testing. In 2025, the company, in collaboration with other institutions, successfully applied for a key national railway administration research project on static decoupling.
In 2024, the company's decoupling robot was recognized as a Sichuan Provincial Major Technical Equipment First-of-its-Kind Product. Plans are underway to collaborate with the China Railway Xi'an Group Co., Ltd. to achieve the practical implementation of the hump yard embodied intelligent decoupling robot at Xinfengzhen Station within the year.
The decoupling robot is not only agile but also possesses sharp "vision." The key to decoupling lies in a small hole, less than one centimeter in diameter, hidden beneath the coupler between carriages. Although difficult for the naked eye to discern, the robot can accurately locate this target even under complex conditions like coal dust coverage, rain, snow, or darkness.
This visual capability was developed by training on massive datasets. Following the 2008 Wenchuan earthquake, the company began monitoring slopes along railway lines. Nearly two decades of on-site railway project experience have allowed the company to accumulate a valuable data repository—35 million meticulously annotated rail transit images, which serve as a crucial resource for solving technical challenges.
Through a closed-loop process involving data annotation, model training, virtual simulation, and real-world verification, the team continuously refines its AI algorithms. This enables the robot to identify hazards like foreign object intrusions and falling rocks with a false alarm rate of less than 1%.
Leveraging this data accumulation, the company has developed another specialized skill: an AI disaster prevention robot. Acting as an intelligent "sentry" along railway lines, it can automatically identify disasters such as mountain collapses, landslides, falling rocks, and mudslides and issue timely alerts, with a single device covering up to 4 kilometers.
Currently, the decoupling robot has been deployed in applications with the Shaanxi Coal Group, as well as power plants in Inner Mongolia and Guangdong, generating over 30 million yuan in orders last year. Business growth has exceeded 20% annually for the past two years. The AI disaster prevention products are already in use across 12 of China's 18 railway bureaus, demonstrating significant application results.
The intelligent transportation sector has become highly competitive, with traditional security firms, industrial robot companies, and new tech startups all entering the field. However, few companies possess a deep understanding of specific industry scenarios like railways, can address the demands of extreme working conditions, and have high reliability and practical engineering capabilities. The company's core strategy focuses on "digging deep in one area while broadening slightly," advancing R&D iteration and cost reduction to lower the price of a single decoupling robot system from tens of millions to between 3 and 5 million yuan. It is also expanding the application scenarios for embodied intelligent technology from rail transit to energy, ports, logistics, chemicals, and safety and disaster prevention, while accelerating international patent布局 and exploring markets in Europe and Southeast Asia.
The market potential for decoupling robots alone is estimated to reach hundreds of billions of yuan if comprehensive intelligent application is achieved.
The development team leader, a graduate of the University of Electronic Science and Technology of China, led the breakthrough in hump yard decoupling technology despite having no personal experience in manual decoupling. The key was leaving the laboratory to learn directly from experienced veteran workers. While static decoupling is performed on stationary trains, the most difficult challenge is hump yard decoupling, where freight trains move slowly at 3 to 12 kilometers per hour. Workers must jog alongside and apply force instantly while moving. Veteran workers emphasize the need for a skillful touch and a specific "jolt" or feeling at the precise moment of successful decoupling.
To capture this intangible "feel," the team spent months observing veterans at Xi'an Xinfengzhen Station, studying their movements and techniques. They equipped the robot with a six-dimensional force sensor, effectively giving it a "sixth sense" to perceive minute changes in force and torque across six dimensions in real-time. Through repeated real-world testing, they collected data from each successful decoupling, using协同 training of large and small models to translate the critical "jolt" into a precise, actionable threshold the robot can understand and execute.
True research and development requires understanding real-world application scenarios. By immersing themselves in the actual working environment, the team could grasp genuine customer needs, learn from the wisdom and experience of veteran workers, and perfectly integrate AI large model technology with practical applications.
Comments