While humans often use touch to understand the physical world, many robots lack this ability, relying instead on cameras and computer-vision approaches.
Recently, tactile sensors with high resolution and high dynamic range but low cost have been developed, for example, magnet-elastomer-based sensors. The physical properties of the skin or biomimetic features such as fingerprints can also help with perception.
A recent study published on arXiv.org expands the sensing capabilities of magnet-elastomer-based tactile sensors by adding biomimetic features, including fingerprint ridges. Researchers have proposed a low-cost magneto-elastomer structure with double layers based on human subcutaneous anatomy.
The results show that fingerprint ridges significantly improve the sensor’s ability to classify materials with distinct surface properties across a range of velocities.
Tactile sensing typically involves the active exploration of unknown surfaces and objects, which makes it particularly effective in processing the characteristics of materials and textures. A key property extracted by human tactile perception is surface roughness, which relies on measuring vibrational signals using a multi-layered fingertip structure. Existing robotic systems lack tactile sensors that are able to provide a high dynamic sensing range, sense physical properties, and maintain low hardware costs. In this work, we introduce the reference design and fabrication process of a miniature and low-cost tactile sensor consisting of a biomimetic dermal structure, including an artificial fingerprint, dermis, epidermis, and an embedded magneto-sensor structure, which acts as a mechanical receptor. works in. Converting mechanical information into digital signals. The presented sensor is capable of detecting high-resolution magnetic field data via Hall effect and generating high-dimensional time-frequency domain features for material texture classification. Additionally, we investigate the effects of different surface sensor fingerprint patterns to classify materials through both simulation and physical experimentation. After extracting the time series and frequency domain features, we calculate the k-nearest neighbor classifier to differentiate between different materials. The results of our experiments show that our biomimetic tactile sensor with fingerprint ridges can classify materials with more than 8% higher accuracy and less variability than ridge-less sensors. These results, together with the low cost and adaptability of our sensors, demonstrate high potential for reducing the barrier of entry for a wide range of robotic applications, including model-less applications for texture classification, material inspection and object recognition. Touch sensing is included.
Research Article: Dai, K., “Design of Biomimetic Tactile Sensors for Material Classification”, 2022. Link: https://arxiv.org/abs/2203.15941