Human-robot interaction control of an intrinsically compliant parallel wrist rehabilitation robot

  • Tanishka Goyal

    Student thesis: Doctoral Thesis

    Abstract

    The field of rehabilitation has undergone tremendous transformation in recent years. From conventional forms of rehabilitative therapy that included a monotonous, repetitive exercise to the inclusion of rehabilitation robots to make the therapeutical treatment less daunting. Rehabilitation robots provide an objective, engaging, and inexpensive alternative to traditional practices while reducing the burden on the healthcare system as well as the patient. However, in the last few decades, the development of such devices was more focused on the lower limb. Due to the complexity of movements, the devices available for rehabilitating the wrist are limited in the literature. Therefore, this research aims to develop a compliant parallel robot for wrist rehabilitation in three degrees of rotational freedom: Pronation/Supination (PS), Flexion/Extension (FE), and Adduction/Abduction (AA). Novel control strategies have been developed to guide the robot in assisting the patient in achieving the rehabilitative goal. The developed prototype follows an end-effector design with a parallel mechanism. Intrinsically compliant Biomimetic Muscle Actuators (BMA) power the prototype and provide the necessary movement. Since these actuators have inherent hysteresis and transient characteristics, a heuristic model has been developed to provide an accurate and time-efficient relationship. The rehabilitation robots, by definition, work in proximity to the human subject and work in partnership; hence, physical interaction is certain. The physical human-robot interaction has highly nonlinear and uncertain dynamics. Therefore, the Koopman Operator theory has been employed to develop a system identification model. The Koopman Operator is a mathematical tool that linearizes highly nonlinear dynamical systems by lifting the state space into an infinite dimensional space. This data-driven approach helps identify the nonlinear system dynamics and develop a trajectory-tracking controller for wrist rehabilitation. The effectiveness of the Koopman Operator is also tested in designing an adaptive controller to predict anatomical stiffness. In a healthy person, the anatomical stiffness is accommodated by the neuromuscular system, which is affected by stroke. Hence, a successful controller should adapt to the anatomical stiffness and the altered physiological capabilities. The Koopman Operator was used to develop the model for predicting anatomical stiffness, which depends on the axis of rotation and the geometric orientation. Rehabilitation therapy can be considered a joint task undertaken by the human subject and the robot with physical interaction. In other words, it can be deemed a coordination game with the human and the robot as the two players. The human’s strategy is unknown to the robot, but the controller should interpret the human subject’s intention. Therefore, an adaptive estimation method was developed to estimate the human’s intention and then assist them in achieving the goal while fulfilling the common objective of wrist rehabilitation. The concept of modeling the human subject and the robot as two agents with a common goal are then extended to exploring them as two independent energy sources. As active human participation is crucial for prompt recovery, the robot is expected to decrease its energy dissipation to increase the level of involvement from the patient. An autodidactic algorithm was developed to estimate the transactive energy between humans and robots during physical interaction. The energy dissipation of the human and the robot was mapped for each orientation attained during the rehabilitation session. The physiological capabilities and the effects of stroke vary from patient to patient. Therefore, it is crucial that the controller can adapt to diverse needs. Accordingly, smart avatars were programmed to learn from the human subject in real-time and provide an energy-efficient rehabilitation trajectory. The smart avatars included a controller with energy optimization to modify the trajectory to minimize the robot’s energy dissipation and an Inverse Dynamics model to simulate the subject and estimate the subject’s involvement. The avatar was then appended with an Assist-as-Needed controller that calculates the robot’s participation in achieving the goal successfully. The essential contributions of this research are the development of an intrinsically compliant parallel robot for wrist rehabilitation with energy-efficient control algorithms. The algorithms developed in this research were successfully tested with healthy human subjects; however, extensive clinical trials with neurologically impaired subjects are required to establish the efficiency of the proposed prototype.
    Date of Award2023
    Original languageEnglish
    SupervisorShahid Hussain (Supervisor), Elisa Martinez-Marroquin (Supervisor) & Nick Brown (Supervisor)

    Cite this

    '