Improving Effectiveness of Robot-Assisted Ankle Rehabilitation via Biomechanical Assessment and Interaction Control

Abstract

The human ankle joint plays a significant role in maintaining body balance during ambulation, but it is particularly susceptible to musculoskeletal and neurological disorders. A general rehabilitation program for ankle injuries requires intensive efforts from therapists and patients over prolonged sessions. Robot-assisted rehabilitation solutions, as therapeutic adjuncts to facilitate clinical practice, have been actively researched in the past few decades and provide an overdue transformation of rehabilitation from labour-intensive operations to technology-assisted operations. Various rehabilitation devices have therefore been developed for the treatment of ankle injuries to reduce the physical workload of therapists and supplement the resources required to facilitate a comprehensive rehabilitation regime so that adequate therapy can be delivered to patients. However, the effectiveness of existing ankle rehabilitation robots is limited by a variety of shortcomings including kinematic incompatibility (misaligned centre of rotation of the robot and the ankle joint), non-compliant actuation, less than three rotational degrees of freedom (DOFs), and the lack of real-time ankle assessment and adaptive interaction training schemes. This research aims to improve the effectiveness of robot-assisted therapy for the treatment of musculoskeletal and neurological ankle injuries. The fundamental technology is the development of an intrinsically-compliant ankle robot with real-time biomechanical assessment and interaction control. The current device is named the Ankle Assessment and Rehabilitation Robot (AARR). The AARR has a bio-inspired design, with the functions of both assessment and rehabilitation, devised after a systematic review of a variety of ankle rehabilitation devices. Mechanically, this novel robot is designed with flexibility in generating varying training ranges of motion (ROMs), and employs four intrinsically-compliant Festo fluidic muscles (FFMs) that mimic skeletal muscles to actuate three rotational DOFs. Functionally, the AARR incorporates sensor-based and model-based ankle assessment techniques to facilitate the robotic control for enhanced safety and rehabilitation efficacy. The ankle assessment protocol for this robot aims to extract biomechanical information of the ankle joint from sensors and computational models. Ankle biomechanical assessment via sensors is implemented by three magnetic rotary encoders for measuring ankle position and a six-axis load cell for measuring patient-robot interaction. Two computational models are developed for estimating ankle ligament kinematics and passive joint torque. They are distinguished by different definitions of their rotation axes, where the torque model moves about three perpendicular rotation axes (named the PRA-Model) while the rotation axes of the ligament kinematics model are not perpendicular (named the Non-PRA-Model). The proposed ankle assessment techniques are demonstrated to be valid and reliable in extracting ankle biomechanical information during the robotic training through comparisons with published data and experimental validation. The trajectory tracking of the AARR is implemented by controlling the individual FFM length in joint space, or a cascade controller with position feedback in task space (the outer loop), and force feedback in joint space (the inner loop). With position controllers in the low level, two adaptive interaction training schemes are proposed to enhance patient engagement and rehabilitation efficacy. One scheme is implemented through the predefined trajectory that is adaptive to the movement intention of the patient. The other scheme employs a high-level admittance controller whose performance is adaptively tuned according to real-time patient-robot interaction. Experiments were conducted on a sprained ankle to evaluate the proposed control strategies when implemented on the AARR, with all normalised root mean square deviation (NRMSD) values of the trajectory tracking at less than 5.4%. To conclude, the AARR has the potential for clinical applications of ankle assessment and rehabilitation, and both interaction training schemes are safe and effective for patients by considering the movement intention and real-time patient-robot interaction.