Neural network based control method implemented on ambidextrous robot hand

Abstract

Human hands have natural ability to perform number of tasks precisely without exact knowledge. This paper investigates the key differences in performance when conventional controllers are combined with Neural Networks (NN). All the tests are performed on our uniquely designed 3d printed multi-finger ambidextrous robot hand. The ambidextrous hand is actuated by pneumatic artificial muscles (PAMs) and able to bend its fingers in both ways left side and right side offering full ambidextrous functionality. The approach followed here is to use force sensors intelligently by implementing them on fingertips of the hand. In our control method, grasping trajectory of each finger combines its data with the neighboring fingers to get an accurate result. Results gathered from the tests are summarized in the table 5.