A Deep Learning-Aided Framework for Joint Angle Estimation of an Upper Limb Rehabilitation Robot

Abstract

The problem of inverse kinematics in serially manipulated upper limb rehabilitation robots involves deducing joint rotation angles from the position of the end-effector. Unlike forward kinematics, inverse kinematics lacks systematic solution approaches, and it is especially challenging for certain robot morphologies. This study proposes a deep learning-based model to estimate joint angles from a specified end-effector position. The model shows considerable effectiveness in calculating joint angles for a variety of target positions. The enhanced position-tracking capability of the proposed algorithm than existing analytical methods will enable the development of efficient controllers in future.