Segmented Soft Pneumatic Bending Actuator with Artificial Neural Network for Parameter Prediction

Abstract

Most of the models of soft robotic gloves can do flexion and extension. However, there are other hand rehabilitation exercises such as tendon glide which requires variations of finger pose. It cannot be done with just a single motion of flexion or extension. Individual joint control is needed in order to achieve the normal-to-maximum range of motion when doing the hand exercises. In this paper, a design of a segmented soft pneumatic bending actuator (sPBA) with individual joint control using PID with Artificial Neural Network (ANN) for parameter prediction is presented. Each joint has its individual inlets and has pneumatic chambers that bends when supplied with air. A pneumatic control setup is developed to control the three finger joints; the Metacarpophalangeal (MCP), Proximal Interphalangeal (PIP), and the Distal Interphalangeal (DIP). Varying the air pressure supplied to the joints achieves different bending positions of the finger. An experimental setup was developed in order to characterize and gather data that is used to develop ANN for predicting the bending angle-to-pressure parameter. A total of 2197 images are captured from the different combinations of pressure which are equivalent to different bending angles. A simple PID control was used to achieve the desired bending. The setup has a mean-square-error (MSE) of 1.85007 at validation with overall R of 0.9994 and a maximum error of 5.4 kPa pressure in joint 1 at low pressure. This setup will be useful to develop a soft robotic rehabilitation glove.