Digital human model and training task planning-based adaptive assist-as-needed control for upper limb exoskeleton

Abstract

To address the challenges of diminished motivation and increased fatigue observed in participants during active rehabilitation training, this study proposes a digital human model-based adaptive assist-as-needed (DHM-AAAN) control for an upper limb exoskeleton. This control framework consists of two main sub-controller loops: an outer sub-controller loop that determines the necessary assistive force, and an inner sub-controller loop which enables the exoskeleton to accurately replicate target movements while applying the assistive force derived from the outer sub-controller loop. Within the outer sub-controller loop, a strategy known as the digital human model and task performance evaluation (DHM-TPE) is employed to evaluate participants’ mobility capabilities and overall condition. Based on the assessment results, parameters such as radius, frequency, and assistive force are dynamically adjusted for multi-period trajectory tracking tasks through the implementation of an adaptive frequency oscillator (AFO) algorithm integrated with a digital human model. In the inner sub-controller loop, a barrier Lyapunov function-based hybrid force/position control with shifting error constraints (BLF-HCS) controller is introduced. This controller utilizes radial basis function neural networks (RBFNN) and error offset functions initialized with random values. The BLF constrains the exoskeleton’s tracking error, considering potential deviations from the desired initial position during the early phases of movement. To validate the effectiveness of the proposed controller, this study presents joint simulation results of the rehabilitation training cycle for circular task trajectories, experimental results from individual participants, and the average results from 6 participants.