Design of varying control based on human’s motion proficiency for human - machine cooperative system under physical interaction

Abstract

Physical training enables us to recover from disability in rehabilitation or to acquire motion skill. So far, many devices that assist human's motion such as power assist suit have been developed. However, these devices do not consider human's proficiency level or recovery degree, therefore some adjustments of the device i.e. how much the device generates its force, will be required according to the human's proficiency level. In this paper, focusing on stabilization of an inverted pendulum, we propose a varying controller design method that changes its contribution for stabilization based on human's proficiency level. The controller is designed based on Model Predictive Control (MPC) for fast on-line calculation. Human's proficiency level is calculated from convergence level of the inverted pendulum system. The convergence level is obtained by a norm of state variables which is coordinate-transformed so that its norm decreases monotonically according to convergence. Based on human's proficiency level, the controller is designed to change its contribution. Experimental results show the degree of controller contribution is decreased for high skilled subjects and increased for low skilled subjects.

Contact force regulation in physical human-machine interaction based on model predictive control

“…Simpler design choices could be discarded over more complex control strategies involving adaptiveness [27,28], robustness [29][30][31], and predictiveness. Among the different approaches available, MPC can be selected due to its internal robustness, dynamic performance, and capabilities to avoid constraint violations dictated by the plant and the actuation system [32][33][34][35]. In the MPC strategy, the system to be controlled is totally or partially known a priori, not only through a state space quadruplet representation but also in possible constraint equations involving both the plant and the actuation parts.…”

“…For example, while in dos Santos et al [32] an MPC loop is used to find optimal stiffness parameters for a lower limb-rehabilitation exoskeleton impedance controller, Erickson et al [33] couples the MPC with a learning-based model for pHMI-assisted dressing. Moreover, Teramae et al [34] and Okada et al [35] exploit MPC online calculations to adapt the control algorithm, respectively, to assist a patient rehabilitative movement only when needed or to comply with various pHMI proficiency levels during limb motion training and rehabilitation.…”