Real-Time Estimation of Intended Lower-Limb Motion for Assist Robots

Motion assist robots have been studied to assist human motion for elderly, disabled, injured persons. The motion assist robots can be classified into two types: wearable robots which are directly attached to the user's body and external robots which assist the user's motion from the outside of the user. It is important for those robots to estimate motion intention of users in real-time to assist their daily motion effectively. This paper presents a method to estimate human lower-limb motion intention and its motion phase in real-time for the motion assist robots. By detecting motion phase in addition to the intended motion in real-time, the robots can figure out when and how human motion are assisted even in the middle of the motion. In the proposed method, integrated information about human physical and biological data and the surrounding environment are used to estimate the motion intention and its phase. Damping neuron has been applied in the artificial neural networks for the intended motion estimation and also for the motion phase estimation. Furthermore, changing velocity from the output of the artificial neural network is considered to figure out the intended motion in early stage. Experimental results show that lower-limb motion and its phase are effectively estimated with the proposed method. Furthermore, it is shown that the motion phase is correctly estimated even if the user gives up the intended motion during the motion.