Active lower limb exoskeletons are increasingly used in rehabilitation therapy. An important aspect and ongoing research topic in developing such assistive devices is safe human–machine interaction. Another major challenge in the control of exoskeletons is determining the subject's intention to move for providing cooperative assistance. To address both challenges, we first present the design of a new lower limb exoskeleton with variable stiffness actuators for compliant coupling between human and robot. The exoskeleton's actuated hip, and knee joints are controlled via a stiffness-dependent torque control. In a second step, we derive a detailed model for the swing phase using the Lagrange formalism. We utilize the model and an interaction torque measurement based on the deflection of the serial elasticity to estimate the subject's motion intention. This estimation is realized by nonlinear state estimation using an unscented Kalman filter (UKF). The UKF is able to track the hip and knee torque contributed by the subject during normal gait in real time and provides more robust results than the inverse dynamics approach.
Read full abstract