CASIA OpenIR  > 复杂系统管理与控制国家重点实验室  > 先进机器人
外格式上肢康复机器人系统设计与控制方法研究
罗林聪
Subtype博士
Thesis Advisor侯增广
2019-05-31
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Name工学博士
Degree Discipline控制理论与控制工程
Keyword上肢康复机器人 外骨骼机器人 被动训练 强化学习 控制策略
Abstract

由脑卒中、脊髓损伤和脑损伤导致的中枢神经损伤,造成患者肢体运动功能障碍,严重影响其日常生活自理能力。临床研究表明,运动康复治疗可以促进患者肢体运动功能的恢复。传统康复治疗是治疗师以人工操作的方式辅助患者完成训练任务。随着患者人数的增加,康复训练任务给治疗师带来沉重的体力负担,导致许多患者无法获得所需的康复训练,限制了康复治疗效果。外骨骼康复机器人关节分布与人体对应,能模拟人体日常生活动作,在提供复杂三维空间运动训练方面具有显著优势。相比于传统人工康复训练方式,康复机器人能提供时间长、精度高、可重复性强的运动训练,减轻治疗师体力负担的同时降低了康复训练成本。本文在国家自然科学基金重点国际合作研究项目“康复机器人主动自适应控制策略与在线评价方法研究与应用”(61720106012)、北京市科技计划项目“主动自适应上肢康复机器人研发”(Z161100001516004)的支持下,针对上肢康复训练,围绕机器人的系统设计和相关康复训练控制方法展开研究。本文的主要工作和创新点如下:

1. 针对神经损伤患者不同关节的康复训练需求,设计了两款上肢康复机器人:外骨骼式上肢康复机器人和三自由度腕关节康复机器人。外骨骼式康复机器人具有五个主动关节,可以为上肢提供单关节、多关节及三维空间运动训练;机器人采用弹簧减重系统实现机器人关节重力平衡,并利用四个被动滑动副解决关节轴心对齐问题。腕关节康复机器人利用圆弧导轨与滑块结构实现机器人关节与人体前臂的同轴旋转,并采用钢丝绳与绞盘传动方式提升机器人关节的反向驱动能力。在机械结构的基础上,本文完成了机器人控制系统和上位机软件系统的设计。

2. 针对外骨骼式康复机器人冗余自由度逆运动学求解问题,提出了肘关节抬升角约束项,以确保解的唯一性。基于人体上肢运动学约束,建立以抬升角为自变量的优化目标函数。由于目标函数解析解困难,也难以利用梯度下降法求得近似解,本文利用遗传算法求解最优抬升角,并建立从空间位置坐标到抬升角的映射关系,从而提高机器人实时控制中最优抬升角与逆运动学的计算效率。

3. 为实现外骨骼式康复机器人的被动训练模式,本文针对训练任务的周期性特点,设计了基于迭代学习控制的轨迹跟踪控制方法,并提出了一种增量式高阶迭代学习控制算法,以减小控制器输出的振荡,提高控制器的收敛速度。为了减少机器人对患者的过度干预,在被动训练模式的基础上,设计了一种基于虚拟“管道”模型的训练控制方法,以允许患者的部分运动自由。

4. 为了在外骨骼式康复机器人平台上实现主动训练模式,提出了一种基于强化学习的辅助控制方法。该方法包含两层控制环路:内环利用传统的闭环控制算法实现轨迹跟踪;外环利用强化学习辨识操作者的运动意图,并为内环提供参考轨迹。该方法的特点在于:运动意图的估计不依赖于人机动力学模型,规避了复杂的动力学辨识过程;具有动态自适应能力,对于不同的操作者和任务,可在不修改控制器的情况下学习得到正确的辅助策略。

5. 提出了两种自适应“按需辅助”康复训练控制策略,使机器人在辅助患者完成训练任务的同时,能够提高患者的主动参与度。在基于患者刚度系数辨识的控制策略中,建立目标回报函数用于权衡轨迹跟踪误差和患者输出功;然后,根据辨识得到的患者刚度系数求解最优阻抗控制参数,使回报函数最大化。针对工作空间中患者肢体运动能力的不一致性问题,提出了基于患者运动功能障碍模型的控制策略:康复训练中,利用神经网络建立患者的运动功能模型,并采用贪婪式算法更新网络权值,以产生任务挑战激励患者的主动参与;同时,利用自适应算法调节任务难度系数以适应不同功能损伤程度的患者。上述两种训练控制策略,分别通过仿真和临床对比实验进行了验证。

Other Abstract

Stroke, spinal cord injury and cerebral injury usually lead to the motor dysfunction, which has a serious impact on the ability of patients to conduct activities of daily living. Clinical researches in rehabilitation therapy have shown that the exercise training can promote the recovery of patients' motor functions. The conventional rehabilitation therapy is that therapists assist patients to conduct training tasks in a manual way. With the increase in the number of patients, the labour intensive and time-consuming rehabilitation training becomes a heavy burden for therapists, and patients can not receive the required amount of rehabilitation training. Exoskeletons have the advantages in providing complex training exercises in workspace because the joints of exoskeletons correspond to human joints. Robot-assisted therapy can provide continuous, intensive and precise rehabilitation training, release therapists from the heavy labor, and reduce the cost of therapy. Supported by the International Cooperation Program (Grant 61720106012) and the Beijing Science and Technology Project (Z161100001516004), this paper focuses on the design and control of the rehabilitation robot systems for the upper limb rehabilitation therapy. The main contributions and innovations of this research are as follows:

1. An upper limb exoskeleton and a 3-DOF wrist robot are developed for rehabilitation therapy of neurologically impaired subjects. The exoskeleton including five active joints can provide single-joint, multi-joint and 3-D space training tasks. For the mechanics of the exoskeleton, a spring mechanism is used for the gravity compensation, and four passive joints are designed for the alignment between the robot and human joints. The wrist robot uses an arc guide rail and slide mechanism for aligning the robot joint with human forearm, and adopts a cable transmission to decrease joint friction. The control architecture and software for the both robot systems are designed.

2. In order to resolve the inverse kinematics of the exoskeleton with redundancy, a swivel angle about the elbow joint is proposed to make the uniqueness of the solution. A optimal objective function with the swivel angle as independent variable is established based on the kinematic constraint of upper limb. It is difficult to obtain analytical solutions of the optimal objective function, and get approximate solutions using gradient descent method. In order to improve the computation efficiency in real-time robot control, a genetic algorithm is used to find the optimal swivel angle, and then the relation mapping from wrist position to the optimal swivel angle is built.

3. Considering the periodicity of passive training tasks, an tracking control scheme based on iterative learning control (ILC) is proposed for the passive training on the exoskeleton, where a incremental  high-order ILC is designed to reduce the oscillation of the controller output, and promote the convergence of the controller. In addition, a virtual tunnel-based control is developed to improve the voluntary participation of patients.

4. Aiming to realize the active training, an assistive control scheme based on reinforcement learning is proposed for the upper limb exoskeleton. The proposed control scheme has a hierarchical architecture: the inner control loop is responsible for trajectory tracking; the outer loop consists of reinforcement learning algorithms which are used to learn the motion intention of patients and provide the reference trajectory for the inner loop. The method is characterized in that the implementation does not depend on the complex dynamics model of the human-robot system, and it has the dynamic adaptability.

5. Two adaptive assist-as-needed control strategies are designed to promote the active engagement of patients. In the first control scheme, the optimal impedance control is determined based on the estimation of patients' stiffness to maximize the reward function which is defined as the weighted sum of the trajectory errors and the patients' effort. Considering the fact that patients have heterogeneous motor capability in workspace, the motor capability model-based control strategy is designed, where a neural network is utilized to model the functional capability of patients. In order to encourage patients' engagement, the weight vectors of the neural network are updated using a greedy algorithm to make the task challenge for patients. Simultaneously, a challenge level modification algorithm is employed to adapt the task challenge to patients with various levels of impairment according to their task performance. The proposed control strategies are validated by simulation and clinical experiments.

Pages144
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23833
Collection复杂系统管理与控制国家重点实验室_先进机器人
Recommended Citation
GB/T 7714
罗林聪. 外格式上肢康复机器人系统设计与控制方法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
ThesisLLC.pdf(11908KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[罗林聪]'s Articles
Baidu academic
Similar articles in Baidu academic
[罗林聪]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[罗林聪]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.