Abstract:
In this study, we propose an adaptive hybrid neuromechanical control to control an exoskeleton (ADAHYN). It fuses two bio-inspired approaches: 1) model-free adaptive neural locomotion control (ANLC) and 2) biomechanical template model-based control (BTBC). The adaptive neural control is composed of decoupled neural central pattern generators (CPGs) utilizing Hebbian-based synaptic plasticity and adaptive neural rhythmic post-processing networks utilizing radial basis function (RBF) with error-based learning. A learning from demonstration technique (LfD) is also deployed here for gait personalization. It facilitates a user or an expert to teach the network in a short learning time (less than two minutes) for individual intrinsic walking style in terms of frequency and pattern where the profiles are applied during the swing phase. While the adaptive neural control can provide gait personalization, the template control simultaneously offers leg compliance and posture balance for gait adaptation to environmental changes. It is based on force modulated compliant hip (FMCH) control and used for the hip compliance modulation during the stance phase. The hybrid control framework inherits robustness against external perturbation by design and can implicitly handle sensory feedback malfunction. Furthermore, it enables the exoskeleton’s adaptability to new environmental conditions, as demonstrated in walking experiments with treadmill speed changes and slope climbing.
Link: