A Two-Layer Adaptive Assist-As-Needed Control Scheme for Rehabilitation Robotics

Abstract

Assist-as-needed (AAN) strategies aim to enhance active participation, thereby promoting neuroplasticity and facilitating motor recovery. However, due to variations in injury severity and motor abilities, non-adaptive AAN control strategies may provide suboptimal assistance. This letter proposes a two-layer adaptive AAN control scheme that integrates a task-adaptive control strategy with a subject-adaptive algorithm. The scheme utilizes surface electromyography signals, which directly reflect the neuromuscular state, to dynamically assess the subject's motion performance and engagement. The task-adaptive control strategy employs a Kalman filter to fuse torque data from both motion performance and physiological state, enabling within-task adjustments that foster active participation. Furthermore, the subject-adaptive algorithm modulates the assistance level inter-task based on motion performance and physiological state, ensuring that the challenge level aligns with the subject's functional capabilities. Experimental validation was conducted with 10 healthy subjects, confirming the effectiveness and feasibility of the proposed scheme.