—ShanghaiTech researchers develop a virtual training ground for tactile intelligence
How can robots learn to touch the world?
When people pick up an egg, twist off a bottle cap, or tie their shoelaces, their fingers constantly send tactile information to the brain, helping it judge an object’s shape, position, and the amount of force being applied. For robots, touch is equally important. It is a fundamental capability for performing delicate manipulation tasks and interacting effectively with the physical world.
Recently, a research team led by Assistant Professor Xiao Chenxi from the School of Information Science and Technology (SIST) at ShanghaiTech University received the Best Student Paper Award at the 2026 IEEE International Conference on Robotics and Automation (ICRA 2026), held in Vienna, Austria. ICRA is widely recognized as one of the premier international conferences in robotics and automation. The award honors student-led research that demonstrates outstanding innovation and impact in the field.
As embodied intelligence continues to advance, robots are expected not only to see their surroundings but also to understand them through touch. However, teaching robots to use tactile information remains a significant challenge.
Training directly on actual robots is often expensive and time-consuming. As a result, researchers typically train robots in virtual environments through computer simulation before transferring the learned skills to real-world systems. Unlike vision, however, tactile sensing involves complex physical processes such as deformation, contact forces, and friction. High-fidelity simulations can accurately model these interactions but often require substantial computational resources, while faster simulation methods usually sacrifice realism.
To address this challenge, Xiao’s team developed ETac, a lightweight and efficient tactile simulation framework that can be viewed as a virtual training ground for robotic touch. Within this environment, researchers can efficiently simulate the pressure, deformation, and tactile feedback generated when robots’ fingers interact with objects. This allows robots to practice grasping and manipulation tasks repeatedly in simulation.
Experimental results show that ETac achieves a strong balance between accuracy and efficiency. The framework is able to reproduce the responses of real tactile sensors while significantly reducing computational cost, providing high-quality training data for tactile learning and dexterous manipulation.
The researchers believe ETac could become a valuable tool for tactile intelligence research, enabling robots to acquire complex manipulation skills more efficiently in virtual environments and accelerating progress in embodied intelligence, tactile perception, and dexterous robotics.
The award represents another international recognition for ShanghaiTech’s research in robotics and artificial intelligence. In recent years, faculty and students from the SIST have received numerous honors at leading international conferences in robotics, computer vision, artificial intelligence, and human-computer interaction.
Xiao’s group has long been engaged in conducting research in robotic perception, dexterous manipulation, tactile intelligence, and human-robot interaction. The team will continue exploring how robots can better perceive and understand the physical world, advancing key technologies for embodied intelligence and real-world applications.
The award-winning paper was authored by Xu Zhe (third-year master’s student), Zhao Feiyu (second-year master’s student), and Huang Xiyan (third-year master’s student), with Prof. Xiao Chenxi serving as the corresponding author.