Multilevel Control Strategy of Human-Exoskeleton Cooperative Motion via Gait Optimization and Fixed-Time Adaptive Techniques

A multilevel control strategy is proposed in a lower-limb exoskeleton to reduce interaction torques and enhance compliance during human-exoskeleton cooperative motion. First, a gait dataset encompassing various movement patterns is constructed through gait acquisition experiments conducted on our lightweight device, with dynamic time warping (DTW) employed for data alignment. At the high level, dynamic movement primitives (DMPs) combined with Gaussian mixture models (GMM) and Gaussian mixture regression (GMR) are utilized to learn multiple demonstration trajectories and generate reference trajectories. At the middle level, an admittance controller is designed to derive the desired exoskeleton trajectory from human-exoskeleton interaction torques. At the low level, an adaptive fixed-time controller, incorporating barrier Lyapunov functions (BLF) and fuzzy logic systems (FLS), is developed to address model uncertainties and output constraints. Finally, the effectiveness of the proposed strategy is validated through both simulations and experiments in active and passive training modes, demonstrating robust tracking, bounded error dynamics, and reduced interaction torques. Note to Practitioners - In wearable motion assistance, an exoskeleton should effectively follow the user's movements. However, individual gait patterns significantly vary due to many differences in individual height, weight, age, and other factors. To address these challenges, this study presents a multilevel control strategy that tackles three key issues. Firstly, at the high level, individual gait is learned and optimized using several artificial intelligence algorithms, which ensures that the exoskeleton adapts to diverse motion characteristics and styles, together with stochastic movements. Secondly, at the middle level, an admittance controller is designed to realize physical human-robot interaction (pHRI). In this framework, both active and passive modes of the exoskeleton can be switched to each other by adjusting an admittance parameter. Finally, an adaptive fixed-time controller is developed to enhance joint position tracking performance at the low level. The proposed three-layer control strategy can be efficiently implemented on real-time embedded controllers, which reduces human-exoskeleton impedance and improves individual wearable comfort performance.