上海科技大学知识管理系统

Many object recognition techniques rely on visual-based detection, requiring high-quality cameras and substantial computing power for high recognition accuracy. When visual detection severely deteriorates under external interferences such as buried scenarios under poor light conditions, adding complementary sensing modalities is essential for maintaining recognition accuracy. Here, we report a tactile recognition strategy, by using a tactile-based sensing array attached on a mechanical hand, which permits real-time acquisition of an object's local topography and stiffness. The obtained information is then processed by a tactile associated machine-learning strategy for object recognition and ten different types of objects can be identified and classified with a recognition rate of 82.1%. In burial scenarios, human and debris can be robustly identified on the tactile system with the recognition accuracy above 80%. © 2021 IEEE.