上肢康复训练轨迹定制优化及柔顺跟踪控制

Customization and Optimization of the Upper Limb Rehabilitation Trajectory and Compliance Tracking Control

卒中后早期的被动康复训练可以促进患者脑神经重塑甚至重获肢体运动能力。多数上肢康复机器人中,只将参数化规则曲线作为被动训练的轨迹,没能融合康复医师的经验及患者的个体特征,缺乏3维个性轨迹的光滑优化;此外,训练过程中的安全柔顺交互的问题未得到解决。针对以上问题,本文基于新型末端牵引式3自由度上肢康复机器人开展轨迹个性化定制优化以及跟踪控制研究。首先,在笛卡尔空间应用导纳算法及力补偿策略实现理疗师对轨迹的定制。然后,用道格拉斯–普克法压缩原始轨迹数据得到型值点,保留初始轨迹的拓扑形状。接下来,用非均匀有理B样条曲线(non-uniform rational b–spline,NURBS)进行插值,并融合动态切换概率和t变异改进蝴蝶优化算法(butterfly optimization algorithm,BOA)优化拟合的轨迹。最后,设计基于径向基(radial basis function,RBF)神经网络的滑模自适应以及速度前馈加比例–积分–微分(proportional–integral–derivative,PID)的控制策略实现末端轨迹跟踪,并增加导纳–阻抗控制形成基于力阈值的多模式柔顺跟踪控制,保证人机交互力较大时患者的安全。结果表明:定制机器人末端轨迹的拖动力在5 N以内;改进的BOA算法具有更高的收敛速度和精度,优化的轨迹曲率和更小;轨迹跟踪控制策略可以将跟踪误差控制在6 mm内;模式切换在0.3 s内响应,能顺应较大交互力,从而提高训练的安全性。

Early passive rehabilitation training after stroke can promote brain nerve remodeling and even regain motor ability. Most upper limb rehabilitation robots use parameterized regular curves as passive training trajectories, which fail to integrate the experience of rehabilitation doctors and the individual characteristics of patients, or lack smooth optimization of three-dimensional customized trajectories and safe interaction during training. Therefore, this research studied the trajectory customization, optimization and tracking control strategy based on the novel end-effector 3degrees of freedom(DOFs) upper limb rehabilitation robot. Firstly, the admittance algorithm and force compensation strategy were applied in Cartesian space to realize the customization of the trajectory by the physiotherapist. Then, the original trajectory data were compressed by Douglas-Peucher method to obtain a few via points. The non-uniform rational B-spline curve(NURBS) was used for interpolation, and an improved butterfly optimization algorithm(BOA) based on dynamic switching probability and t mutation was used to minimize the trajectory curvature. Finally, the sliding mode adaptive control based on the radial basis function(RBF) network and the speed feedforward plus PID control strategy were designed to control the trajectory tracking. In addition, the admittance-impedance control was added, forming a multi-mode tracking controller, to make the training safe with the large human-robot interaction force. The results showed that the dragging force of trajectory customization is within 5N. The improved BOA algorithm has higher convergence speed and accuracy, and the curvature of the optimized trajectory is smaller. The control strategy can resist the interference within the threshold and keep the tracking error within 6mm. The mode can switch in 0.3s, to compliant to large interaction force and improve the safety of training.