Robotic Assistive Optimization and Control Using Neural Dynamics and Adaptive Neural Network
提出一种神经动力学阻抗优化与轨迹自适应方法，用于软体外骨骼的踝关节辅助，通过自适应控制实现轨迹跟踪，实验表明最大跟踪误差降至0.016弧度，阻抗参数在3个步态周期内收敛。
Humans can naturally learn and adapt to walking patterns in a variety of terrains. To simulate this learning characteristic, this article introduces a neural dynamics-based impedance optimization and trajectory adaptation approach for our designed soft exosuit, with a dual-driven configuration to assist both ankles of individuals. This method adaptively learns the impedance of the human ankle joint using measured interaction forces and dynamically adjusts trajectories to align with real-time human-robot interaction. Additionally, an adaptive control framework integrating neural dynamics-based optimization with several adaptive laws is developed to achieve stable tracking of updated reference trajectories, with Lyapunov stability analysis confirming uniform ultimate boundedness (UUB) of the closed-loop system. The designed controller offers the benefit of concurrently addressing trajectory adaptation, force control, and impedance tuning for soft exosuits. Experimental validation on human subjects across various terrains demonstrates that the proposed method reduces maximum trajectory tracking error to 0.016 rad (lower than PID and ADRC controllers) and enables impedance parameters to converge within 3 gait cycles. The controller concurrently addresses trajectory adaptation, force control, and impedance tuning, offering a lightweight (8 kg) and wearability-optimized solution for walking assistance.