Two-wheeled vehicles that use conventional stability-control systems need to lean to steer, which makes it challenging for rider-assistance technologies to tell whether a rider is deliberately turning or beginning to lose balance and risk a fall. To overcome this issue, researchers at Shibaura Institute of Technology (SIT) in Japan have created a rider-intent-aware control system that can differentiate between intentional cornering and instability, activating stabilization support only when necessary.
The study was led by Associate Professor Hiroaki Kuwahara from the Department of Machinery and Control Systems at Shibaura Institute of Technology (SIT), Japan, alongside second-year master’s student Shota Tsukase from the Graduate School of Systems Engineering and Science at the same institution. The team aimed to address a major drawback in conventional stability-control systems, which typically react only to vehicle motion and can sometimes interfere with a rider’s intended actions. They published their research on June 19, 2026, in IEEE/ASME Transactions on Mechatronics.
“We thought haptic technology could go beyond simple force feedback and actually help interpret rider intent,” Kuwahara said. “By studying how the rider and vehicle interact, we set out to design a mobility system that intervenes only when necessary.”
A Bicycle Designed to Interpret Rider Intention
To accomplish this, the researchers built a steer-by-wire bicycle in which no mechanical connection links the handlebars to the front wheel, unlike a traditional bike. Instead, the connection is electronic, enabling the system to monitor steering inputs and rider–vehicle interactions while still preserving natural steering feel through haptic feedback, which provides force-based sensations that reflect how the bicycle is responding.
The researchers combined the platform with a machine-learning-based system that classifies rider intent. This system relies on a long short-term memory (LSTM) neural network, a machine-learning model designed to detect patterns in time-series data. Prior to training, the researchers applied K-means clustering, an unsupervised learning method, to group riding data into three categories: straight-line riding, cornering, and instability.
Using data gathered from riding experiments, the LSTM model evaluated variables including steering angle, vehicle speed, roll angle, lateral acceleration, and reaction torque. These inputs allowed the system to understand both the bicycle’s dynamic state and the interaction between rider and machine. By integrating these datasets, the model was able to identify riding conditions in real time.
Distinguishing Intentional Turns from Instability Issues
The results showed that the system was able to reliably classify different riding situations and, most importantly, separate deliberate cornering from unstable motion—even though both involve similar leaning behavior. This ability is essential, since intervening during a normal turn can interfere with the rider’s control, while intervening at the right moment during instability can help prevent a crash.
“Two-wheeled vehicles naturally lean when turning, so it is important to tell the difference between intentional maneuvers and conditions that could lead to a fall,” Kuwahara explains. Our system analyzes both vehicle dynamics and rider interaction to distinguish between them and delivers stabilization support only when it is truly needed.
Once it identified the riding state, the control system reacted based on the situation. In cases of deliberate steering or cornering, the stabilization function stayed off, allowing the rider to fully maintain control. However, when it detected signs of instability, the system automatically engaged stabilization assistance to help the rider regain balance. Experimental results indicated that the method could accurately detect riding conditions and deliver support at the right time without affecting natural bike handling.
Assistance Without Overriding Control
The researchers suggest that developers could ultimately adapt the technology for use in electric bicycles, electric motorcycles, shared bike systems, and delivery vehicles. It may also offer particular benefits for older adults and novice riders by providing added stability while preserving a natural riding feel.
“Our aim is to go beyond traditional automation and develop human-cooperative control,” Kuwahara says. “Instead of replacing the rider, the system interprets their intentions and intervenes only when instability is detected. We believe this approach can help create safer and more user-friendly next-generation mobility.”
In future work, the team aims to extend the system so it can identify a broader variety of riding scenarios and environmental factors, such as different road conditions. Ultimately, they hope to create intelligent rider-assistance systems that cooperate with the rider, improving safety while maintaining full control and maneuverability.
Read the original article on: Tech Xpore
Read more: The world’s first fully robot-staffed hotel is set to open in 2027