RNN Based Knee Joint Muscular Torque Estimation of a Knee Exoskeleton for Stair Climbing

Abstract

This study presents the use of a recurrent neural network to estimate knee joint muscular torques for the development of assistive control strategies of a knee exoskeleton in stair climbing applications. To identify the correct timing of giving assistive torques during the stair climbing process, integrating with a lower limb dynamic model with the foot-force measured data is a common way to derive the knee joint torque profile for gait analysis. However, this estimation method which requires the installation of pressure sensors on the sole of the feet has drawbacks including the inconvenience of exoskeleton wearing and increased moving difficulty. The fact that stair climbing is a sequential movement thus allows us to apply a recurrent neural network to obtain the relationship between the knee joint muscular torque and lower limb gait. Stair climbing experiments on a knee exoskeleton wearer reveal that the trained neural network is able to perform the desired knee joint torque estimation whose results can be applied to derive proper assistive torques in the presence of human-robot interaction.