Abstract:  In order to assist the patients with hemiplegia to carry out multi-mode elbow rehabilitation training， a soft elbow rehabilitation exoskeleton robot was developed.  In addition， a coordination control strategy was proposed based on human muscle torque estimation and adaptive neural network compensation. The surface electromyogram signals were used to identify the motion intention of human and adjust the trajectory of rehabilitation training. The closed-loop control stability of the control algorithm was proved by Lyapunov method. A real-time control experimental platform was established to conduct the trajectory tracking experiments and free active training experiments based on human motion intention. The experimental results show that the proposed control strategy may guarantee the trajectory tracking accuracy in passive training processes. Moreover， the movement trajectory in active training processes may be adjusted based on the motion intention of patients and realize active rehabilitation training with different intensities. 

Key words: soft elbow exoskeleton, muscle torque estimation, adaptive neural network, motion intention, coordination control

摘要： 为辅助偏瘫患者进行多模式肘关节康复训练，研制了一种软质肘关节康复外骨骼机器人，并提出了一种基于人体肌力矩估计与自适应神经网络补偿的协调控制策略。利用表面肌电信号来识别人体的运动意图并调整康复训练轨迹，采用Lyapunov方法证明了控制算法的闭环控制稳定性。搭建了实时控制实验平台，并开展了基于运动意图的轨迹跟踪实验与自由主动训练实验。实验结果表明，所提控制策略能保证被动训练过程的轨迹跟踪精度，并且可以根据患者的运动意图调整主动训练过程的运动轨迹，实现不同强度的主动康复训练。

关键词: 软质肘关节外骨骼, 肌力矩估计, 自适应神经网络, 运动意图, 协调控制