A review of robot grasping tactile perception based on deep learning

For robots, grasping is a necessary skill, and agile grasping by robotic hands is currently a cutting-edge research hotspot. Tactile perception serves as the primary sensory channel for both humans and robots to perceive the surface properties of objects, playing a fundamental role in enabling robots to perform dexterous operations. Tactile perception allows a robotic arm to sense the geometric contour, roughness, hardness, and other material properties of an object during the grasping process. It provides the robotic arm with predictive information on force and angles, thereby enhancing the efficiency of grasping. In recent years, deep learning has achieved remarkable advancements in various industrial domains, including intelligent sorting, defect detection, textile manufacturing, and autonomous driving. These achievements have spurred researchers to shift their focus from machine learning to deep learning in the study of tactile perception for agile robotic manipulation. Despite the unprecedented progress made in deep learning-assisted robot tactile perception, there are still some unresolved challenges in this field. This article begins by discussing the implementation methods of robot tactile perception and then provides a comprehensive overview of the current research and application status of deep learning-based robot tactile perception. Firstly, it highlights the latest advancements in tactile perception for dexterous robot operations. Secondly, it presents an overview of the sources and data acquisition methods employed in existing research. Additionally, the article summarizes the applications of deep learning in robot tactile perception. Finally, it explores current trends and potential future research directions in the field of tactile perception during robot grasping.