Development of Robotic Brace for AIS Treatment: Interaction Dynamics and Control

Farhad Farhadiyadkuri

Publikation: Bog/antologi/afhandling/rapportPh.d.-afhandling


Scoliosis is an abnormal curvature of the human spine in the coronal plane combined with rotated vertebrae in severe cases. Adolescent Idiopathic Scoliosis (AIS), with an overall prevalence of 2% to 3%, is by far the most common form of scoliosis that occurs in adolescents without any known reason. Bracing is the most common conservative treatment of AIS, although surgery is also prescribed in cases with higher Cob angles. The bracing treatment is targeted at preventing the curvature from progressing and correcting the abnormal curvature to avoid surgery. Full-time rigid braces, e.g. Milwaukee, Wilmington, Boston, and Chêneau braces were developed, while the patients are recommended to wear the brace more than 18 hours a day for a long time. The night-time rigid braces, e.g. Charleston and Providence, were proposed to reduce the wear time. The rigid braces are good for curvature correction, but they may cause pain and skin breakdown because of their rigidity. Therefore, soft brace, e.g., TriaC and SpineCor were developed.
One of the challenging issues with the braces currently used in clinics is that the braces are not equipped with advanced technology, e.g. sensors, active actuators, and active control systems. Active braces can make a closed-loop treatment by taking sensory data from patients and applying controlled force on the torso using active control systems and active actuators. While only a few active braces have been recently developed, they have not been commercialized to be implemented in clinical treatments. Therefore, developing active bracing treatment has still too many issues to be solved.
The other disadvantage of the current braces is that adjusting the brace on the patient’s torso to provide the required in-brace correction pressure is currently performed offline. The patients wear the brace for a few weeks, X-ray images are taken regularly, and the doctors adjust the brace based on the X-ray images and their own experience. It is a long time process and concerns about the harmful effects of repetitive radiation exposure have been reported. Computational modeling and numerical simulation have been used to predict the biomechanical behavior of the AIS bracing treatment. However, the accuracy of the model needs to be improved and the in-brace correction pressure is still unclear.
In addition, the in-brace correction pressure is currently controlled passively by regulating the tightness of the brace’s strap. But this method cannot guarantee that the predefined pressure is provided. Because it is an offline and passive method, and there is no online monitoring of the pressure exerted on the torso. Although position and force control strategies are used in a few research works to control the active braces, the position and force cannot be controlled independently because they are dynamically dependent. Impedance control has been widely implemented in robotic applications, but there are no attempts in which impedance control is implemented in the AIS bracing treatment.
Safety, comfort, and compliance are three important factors that need to be guaranteed before applying any robotic treatment to the human subject. Because the robot has direct contact with the patient’s torso. On the other hand, Digital Twin (DT) and Machine Learning (ML) are advanced technologies that have entered many application domains to guarantee the accuracy of computational models and predict the performance of the physical system in advance. However, no DT and ML have been implemented in the AIS bracing treatment.
The last challenging issue considered in this thesis is how to make the active control strategies adaptable to the changes that occurred in the torso during the treatment. The human torso is a complicated environment with variable mechanical properties. It means that the impedance of the patient’s torso is variable during different activities. Therefore, the active control strategies also should have variable impedance gains to be adapted to the unexpected changes in the torso. Learning-based variable impedance control is widely used in robotics, but it has not been applied to the AIS bracing treatment. With leaning-based variable impedance control, the robot can learn the desired variable impedance by taking actions on the environment, receiving observations, and updating the controller gains.
This dissertation aims to develop a robotic brace for AIS treatment, create computational models, propose novel impedance control strategies, integrate DT and ML with the AIS bracing treatment, and propose a novel reinforcement learning variable impedance control. The first goal is to develop a robotic brace equipped with sensors, active actuators, and control strategies. Second, computational models, including analytical models, Multi Body-Finite Element (MB-FE) Simscape model, and MB ADMAS model are created to predict the biomechanics of the AIS bracing treatment. The computational models are validated using in-vivo data and numerical simulations. Besides, Force-based Impedance Control (FIC) and Position-based IC are implemented. Model Reference Adaptive IC and Novel PIC are proposed to overcome the limitations of the typical IC. The numerical simulations and experiments are performed to verify and validate the proposed controllers. Furthermore, a DT of the AIS bracing treatment is created in Simscape and a Neural network (NN)-based regression model is proposed to identify the unknown parameters of the DT and validate the DT using real-time data from the corresponding physical system. Finally, a reinforcement learning-based variable impedance control strategy is proposed. The numerical simulations and experiments to verify and validate the proposed reinforcement learning-based variable impedance control will be carried out as an ongoing work in February 2023.
ForlagÅrhus Universitet
Antal sider98
StatusUdgivet - 2022


Dyk ned i forskningsemnerne om 'Development of Robotic Brace for AIS Treatment: Interaction Dynamics and Control'. Sammen danner de et unikt fingeraftryk.