Model Optimisation and Predictive Control of a Pneumatic Driven Ankle Rehabilitation Robot

Rehabilitation robotics is an emergent field of research aimed at assisting healthcare services and impaired patients in reducing the workload of physical therapists and making therapeutic exercises more accessible. However, with any system designed to be used in direct contact with a human and especially with those in a vulnerable or injured state, rehabilitation robots must be held to a high standard of performance and reliability. Improved patient comfort and safety can be achieved by using compliant actuators like pneumatic muscles; however, any inherently compliant structural element of a system introduces time-variance and nonlinearity to its dynamics. Intelligent controller design is essential to this end. Therefore, this thesis presents the development, application, and verification of a predictive intelligent control scheme for use with an ankle rehabilitation robot.
The Compliant Ankle Rehabilitation Robot (CARR) is used as the platform for developing intelligent control strategies during this work. Actuated by Pneumatic Artificial Muscles (PAMs), the system is inherently compliant but also time-varying and dynamically complex. Initially, a feasibility study is conducted to develop modelling and control methods for a single PAM, in which the PSO modelling method is created, approximating a PAM to a mass-spring-damper system and using particle swarm optimisation to calculate optimal parameters, effectively simplifying the system. A piecewise method is also proposed, accounting for nonlinearity by using multiple models for different operating modes and positions. This modelling method is then used to develop an MPC strategy for the PAM, alongside other intelligent controllers, ILC and optimised PID. In comparative experiments, the developed MPC is confirmed to perform well, reducing the motion lag associated with other control methods and providing accurate and smooth motion.
The developed MPC is then implemented to the CARR. Kinematics equations are used to combine the four actuators in parallel to effectively control multiple degrees of freedom. Based on the complex arrangement of the CARR mechanisms, two controller setups are proposed: single-model in which one dynamic model is used to approximate all four actuators, and dual-model, in which the upper- and lower-pairs are modelled separately. Experiments validate the functionality of the developed controllers in comparison with PID and ILC, and results show that each MPC implantation has strengths and weaknesses in different operating scenarios.
Finally, experiments with 15 participants are performed to validate the controller’s performance in a situation approximating the real function of the CARR. Measurements show good tracking accuracy for the single-model setup and good consistency between participants for the dual-model setup, while participant responses suggested several improvements to the CARR design.
The research problems addressed in this thesis are the difficulty in accurate motion tracking control of inherently compliant systems, which are necessary in rehabilitation robotics to provide improved safety and comfort to patients, and the necessity of rehabilitation robots to be validated with participants. The contributions of the work presented here are development of a PSO-based modelling method for dynamic modelling of PAMs, MPC scheme for motion tracking of PAMs, experimental comparison data validating the developed MPC alongside other intelligent controllers, application of the developed controller to the CARR, and participant involved validation experiments of the intelligently controlled CARR. Using the developed strategies, results, and suggested further work in this thesis, a compliant ankle rehabilitation robot suitable for use in clinical environments can be realised.