Transformation Of Joint Space Trajectories To Muscle Activations Using An Artificial Neural Network

Abstract

An artificial neural network has been utilised to model the transformation of limb joint angle and load information into the appropriate muscle activations. In this transformation scheme, an adaptation of the equilibrium point hypotheses is utilised for the control of a limb with both single joint and double joint muscles. A set of activations were input to the muscles' torque-angle characteristic equations and the corresponding joint angles at equilibrium for specific loads were determined. This data was utilised to develop the required inverse model. This work shows that the artificial neural network has functional similarities to its biological counterpart.