Recently, Fudan University announced thatThe high-precision, low-cost "adaptive visual and tactile AI sensor" independently developed by the team of Fudan University's Trusted Embodied Intelligence Research Institute made its public debut at the 2025 World Artificial Intelligence Conference (WAIC).
Fudan University stated that with the emergence of visual and tactile technology, it is becoming possible for robots to have a sense of touch, solving the "last mile" problem of operational accuracy.
It is understood that domestic research and development of embodied robots has relatively complete visual channels, but the tactile technology for embodied intelligence is still lacking.
In the past, the sensing technology based on pressure signals was a single sensory channel. However, in order to break through the bottleneck of operating capabilities, the next generation of embodied robots must have multi-dimensional tactile sensing capabilities similar to human skin.
Multidimensional force decoupling is a classic mechanics problem, and it is also a difficulty in the embodied intelligence industry.
Force touch is the most important sensory channel in the interaction between machines and the environment. However, embodied robots currently lack non-visual force touch channels at the sensory level.
In the face of complex environmental interactions, decomposing vectors is a challenging technical issue.
To solve this problem, the team's tiny camera built into the sensing layer plays a key role.
When the surface is exposed to force, the particles in the sensing layer will be displaced.Then the force deformation information is captured by the camera, and with the help of AI computing power, complex tactile signals can be converted into high-dimensional visual data, and these visual signals can be accurately translated into force distribution information, achieving multi-dimensional accurate decoupling of force.
The sensor is extremely sensitive, reaching 40,000 sensing points per square centimeter, and has ultra-high spatial resolution.
According to reports, the minimum perception threshold of human fingertip skin is approximately between 0.1 N and 0.2 N.
In contrast,The "adaptive visual and tactile AI sensor" developed by a team from Fudan University has a sensitivity limit that is 10 times that of the human body.
As a result, this technology has application potential in industries such as precision processing, electronic assembly, and high-precision sorting.
According to the demonstration, the adaptive visual and tactile AI sensor developed by Fudan University can pick up tofu, pick up potato chips, and grab jelly.
