AVT: Adhesive visual tactile sensor for anthropomorphic and precise manipulation of tiny mechanical components

Grasping and manipulation of tiny objects have always been a significant challenge in manufacturing and industrial automation, especially in tasks such as small-part sorting, kitting, and assembly preparation. The tiny size makes it hard for vision systems to detect and locate small pieces with high accuracy. And they offer only limited contact regions for robotic hands to pick. Thus, this study proposes a non-prehensile manipulation method for tiny components based on a developed adhesive visual tactile sensor. This sensor incorporates a transparent adhesive layer with a visual tactile sensor, enabling the picking of multiple pieces using adhesive force while detecting the object types and locating the positions. We developed a two-fingered dexterous robotic hand that mimics intuitive human behavior, utilizing one finger to adhere a tiny component and another finger nail to detach it. The proposed sensor is equipped on the sensor finger, while the other fingertip features a nail structure. Inspired by human behavior, the hand can stick and detect objects using the sensor finger, and detach the picked objects from the adhesive surface to the target positions using the nail finger. We also propose the associated planning methods for finger and robot motions. Experimental results show that the method can effectively classify and organize various scattered tiny components with a classification accuracy of 96.4% and an average positioning error of 0.51 mm, validating the proposed method for manipulating tiny mechanical components in manufacturing applications.