面向下肢康复机器人临床应用的双向感知增强人机交互方法

During the rehabilitation training assisted by robots, the active participation of the patient's intention and the acquisition of effective sensory and perceptional feedback, is helpful for accelerating the reconstruction of the injured neural system and improvement of the rehabilitation results. The proprioception is one of the most important sources for patients to obtain the feedback information. However, the present lower limb rehabilitation robots have deficiencies in the following two aspects: accurately recognizing patients’ intention and strengthening patients’ proprioception. Therefore, a type of bidirectional perception enhanced human-robot interaction method for the clinical application of lower limb rehabilitation robots, is proposed in this project. On one hand, the recognition of the lower limb movements and the estimation of continuous variables using multiple neural signals, the unified characterization and fusion algorithm for the neural signals and force-position information, and online learning and self-adaptation technologies for the intention recognition model, are to be researched. As a result, a method for accurate and online recognition of human motion intention, based on the fusion of the biomechatronic information, can be designed, to reinforce the robot’s perception of patients. On the other hand, the associated issues about improving patients' proprioception based on the synergic control of the functional electrical stimulation (FES) and robot, are to be investigated. These issues include that, the FES close-loop control methods based on the feedback of sEMG and force-position information, the cooperative control strategies of the FES and robot, and the collaborative optimization method for the myodynamia and fatigue of the key skeletal muscles, etc. Based on the studies mentioned above, the bidirectional perception enhanced and personalized rehabilitation strategies are to be designed and the clinical trials are to be implemented.

机器人辅助的康复训练中，患者意图的主动参与并获得有效的感知觉反馈，能促进神经通路重建、改善康复效果，其中，本体感觉是人体获得反馈信息的重要来源。然而，现有下肢康复机器人系统在精确感知患者意图与强化本体感觉反馈两个方面还存在较大不足。为此，本项目提出面向下肢康复机器人临床应用的双向感知增强人机交互方法。一方面，研究基于多模态神经信号的人体下肢动作识别与关节连续量的估计方法、神经信号与力位信息的统一表征与融合算法、意图识别模型的在线学习与自适应技术等，建立基于多模态信息融合的运动意图精确识别方法，增强机器人对患者的感知。另一方面，研究基于功能性电刺激(FES)与机器人的协同控制技术增强患者本体感觉的相关问题，包括：基于表面肌电和力位信息反馈的FES闭环控制、FES与机器人的协同控制、及下肢关键骨骼肌的肌力与肌疲劳的协同优化等。在上述研究基础上，建立双向感知增强的个性化康复策略，并完成临床实验。

针对现有康复机器人技术的不足，本项目对基于肌电和脑电信号的人体运动意图精准识别、新型运动想象脑机接口（BCI）范式以及基于BCI的患者参与度增强策略进行了研究和创新，构建了强化患者主动参与的机器人辅助康复训练方法并创新研制了全周期下肢康复机器人样机。.首先，针对已有方法在应用肌电信号识别人体运动意图存在的识别率低和鲁棒性不足等问题，项目组提出了基于神经肌骨协同建模与自适应学习的sEMG-关节扭矩建模方法，实现了基于人体下肢关键骨骼肌对下肢关节扭矩的可靠预测。其次，提出了基于FES和VR反馈的新型运动想象BCI范式，有效提高了被试执行下肢运动想象的能力和效率。进而，提出了区域注意力卷积神经网络、改进的共空间模式等算法，有效提高了脑电信号的解码准确率。最后，研究并提出了基于BCI和训练任务在线优化技术提高康复训练中患者参与水平的一系列方法，并通过实验证明了方法的有效性。.基于上述研究，项目组自主研制了国际首款可在完整康复周期为下肢障碍患者提供训练的机器人，并完成了20例临床试验，患者反馈良好。基于相关成果发表SCI和EI检索论文13篇，其中第一标注期刊论文5篇，获得授权国际发明专利1项、国家发明专利5项，并获得第48届日内瓦国际发明展金奖、2021年中国自动化学会技术发明奖一等奖等奖项。