A Tactile Sensor for an Anthropomorphic Robotic Fingertip Based on Pressure Sensing and Machine Learning

Abstract

Feedback in robotic systems is imperative to accomplishing the required task. Speed controllers, visual feedback and other sensors that offer a closed-loop control system dramatically increases the effectiveness of the system. Having a sense of touch greatly increases the ability to pick-up and manipulate objects in robotic manipulation systems. Currently, precise manipulation tasks in various industries still rely on human operators, but a sense of touch on robot hands can improve the systems capabilities to perform like a human. Measuring force, contact location and detecting slip are all abilities to be desired from a tactile sensor. However, current sensors are complex to manufacture as they require different technologies to accomplish different readings. This research presents a pneumatic tactile sensor that is easy to produce and offers multiple abilities such as force feedback and slip detection using machine learning processes. The sensor showed consistent relations between its pressure readings and applied force. Slip detection was also found to be 80% confident using a decision tree classifier. The implementation of machine learning provides the improvement on other sensors to categorize data from more realistic scenarios. The results show the potential to have a one-dimensional sensor perform multiple tasks by using advanced analysis techniques.