人机交互当中提倡以人为本，并且需要关注复杂的环境情况、控制系统以及与人交互的安全性，必须能够密切的配合协调。协作式机器人具备与人安全交互的能力，并且于近年逐渐普及，未来或可成为大众家中常见的配备。根据世界卫生组织研究指出，每年有数千万人受到脑卒中的影响，而要恢复正常生活的主要方法就是透过康复运动训练。由于医疗资源与地域分配不均等问题，若是患者可在家中自行展开，卒中患者无需频密到康复中心，则可降低治疗成本和提高效率。利用机械辅助康复运动是未来的一项趋势，增加了远程控制的可行性，康复师可同时协助多位患者，减轻人力不足的问题。本研究目的在于探讨分析协作式机械臂进行远程复健训练的可行与有效性。搭建以 Robotic Operating System (ROS) 为基础的居家远程训练系统。采用计算机视觉的方案获取辅助器材定位，通过高精度传感器采集患者相关运动信息作为系统反馈。利用阶梯式触发自适应性调整方案，实时调整训练路径，速度。系统中集成表面肌电采集装置，六轴力矩传感器，协作式机械臂，深度相机以达到监测，各硬件之间流畅的信息交流及支持远程控制的目标及透过图形界面促成信息的简单读取和操控。该系统可提供多种康复训练模式。本研究采用与其余康复机械不同的夹爪做为末端执行器，配合简易的辅助器材，同时简化了准备过程，使其不止可协助大范围训练，也可支持精细的手部运动，这些都是配备其他末端执行器无法轻易完成的。使得患者可以在初期得到全面的肌肉训练，维持肌肉力量，避免挛缩，脱臼等问题为后续训练产生影响。此外，本研究中也通过利用模仿学习的控制方案，为患者找出适合自身身体素质的最佳训练路径，除了可针对患者个性化之外，也可将训练效果进一步提升，同时搭配着混合的控制策略，兼顾安全性与个性化。系统方案经过物理仿真系统 Gazebo 的验证其安全性和可行性，训练效果透过人体骨骼肌肉仿真平台 OpenSim 验证，并且配合专业医疗人员建议调整优化。本研究中的远程训练系统经自愿者验证后，训练效果均比有经验的人士指导的训练来的更加全面和更高的肌肉激活度。各硬件采集到的信息可互相作为反馈，实时调整训练方案。系统中所设计的各种安全措施也确保了训练的安全进行与舒适度，在志愿者中获得较好的反馈，充分地展示了其未来可作为远程家用康复训练系统的潜力及进一步拓展了协作机械臂的应用场景。
Human-robot interaction advocates human-oriented and needs to pay close attentionto complexity of the environment, control systems, the safety measures taken and mustbe able to work and coordinate closely. Collaborative robots possess the abilities to worksafely with human and gradually became more popular in recent years, it may become acommon device in our homes in the future. According to WHO research statistics, tensof millions of people suffered from stroke each year.Rehabilitation training is essential torecover most of their lost motor functions. Due to the uneven distribution of medical andmanpower resources, patients could not be attended effectively. In home training wouldsignificantly boost the accessibility of treatment and reduce medical cost.Utilizing roboticdevices on assisting rehabilitation training is trend in the future, increasing the feasibilityof remote control, and rehabilitation practitioners can assist multiple patients at the sametime to alleviate the problem of insuffcient manpower.This research focus on investigating the feasibility and potential of collaborativerobots in performing rehabilitation trainings remotely. Computer vision methods are utilized to locate and attach the assistive equipment,collects patient-related movement information through high-precision sensors as a system feedback. Adjust the training path inreal time, speed through step-wise triggered safety system. The system was built based onRobotic Operating System (ROS) platform to integrate surface EMG, six-axis force/torquesensors, RGB and depth cameras, and robot controller to enable a smooth sensing, communication, and control of the system.A graphical user interface is also designed to facilitateeasy data access and remote controls. The system simplified preparation process and isable to perform gross-level as well as fine level motor training with the our new approachutilizing a gripper-based end-effector with simple assist equipment.It enables patients toget comprehensive muscle training in the early stage, maintain muscle strength, and avoidproblems such as contractures and dislocation to impose an impact on subsequent trainingprocesses.. Imitation learning methods are integrated into the control algorithm to personalize training path for each patient, hybrid control methods are utilized, ensuring thesafety and personalization during training, meanwhile maximizing the training effects andeffciency. A mixed control strategy, with security and personalization is also utilized atthe same time.The framework built was tested by simulating the control of the system in Gazeboand the training effects (muscle activation level) in OpenSim. The performance was alsovalidated with human subjects and generally achieved a better training effect as comparedto a trainer and received positive feedback. The system clearly demonstrated its safetyduring training process and potential of acting as a remote rehabilitation system and thepromising usage of collaborative robots in rehabilitation training, successfully extendedthe application scenarios.
