Estimation of human intended motion and its phase for human-assist systems

Human-assist robots are designed to reduce the burden on the body when lifting objects and assist the elderly, disabled, and other people with muscle weakness in their daily activities. They also avoid accidents and assist motions by recognizing what kind of motion the user is about to make. Therefore, human-assist systems need to estimate the motion intention of a user in real time. The sooner the assist robot can accurately recognize the user's motion, the sooner the assist robot can plan which motion to assist and how to assist it to guarantee success. If the user's intended motion and its phase are estimated in the early stage of the motion, the assist robot can Figure out how the user is moving by comparing with standard motion models to assist the motion as necessary. This paper proposes a method to estimate the user's intended motion and its phase simultaneously in real time based on integrated information consisting of the user's posture, motion, EMG signals, and the surrounding environment. Two kinds of artificial neural networks are applied in the proposed method. Damping neurons are used in the artificial neural network to estimate the motion phase effectively. The intended lower-limb motions and their phases in daily living motion are estimated in real-time. The effectiveness of the proposed method was evaluated by performing experiments of lower-limb motion.