Modelling and adaptive interaction control of a parallel robot for ankle rehabilitation

Abstract

Ankle sprain is one of the most common forms of musculoskeletal injury and an extensive rehabilitation program is typically required to avoid recurrent injuries. Various robotic platforms had therefore been developed for the rehabilitation 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 the patients. However, as significant variability can be present in terms of joint characteristics and severity of impairment between different patients, further development is still required to enhance the adaptability of such robotic systems and to improve the suitability of existing devices as clinical measurement tools. This research addresses the above issues through the development of an ankle rehabilitation robot that can be adapted to different users with minimal hardware or software modifications. In this research, a novel ankle rehabilitation robot was developed to improve on existing platform based solutions. The proposed approach utilises the user's ankle as part of the robot kinematic constraint to allow more accurate measurement and control of foot-shank displacement and moments. To improve the adaptability of the robot, a new online parameter estimation algorithm for biaxial ankle kinematics and a generic computational ankle model were developed to facilitate the adjustment of robot stiffness in real time so that actuating effort of the robot is reduced in stiff foot configurations to prevent application of excessive forces/moments. Such a scheme had been tested through simulation and an experiment involving passive range of motion exercises and it was found that the proposed method does indeed reduce the weighted sum of position errors and actuator effort. A multi-input multi-output actuator force controller was also proposed in this work so that the coupling between actuator currents and actuator forces is taken into consideration by the proposed control scheme to improve force tracking performance. The stability of the force controller was investigated through inclusion of higher order dynamics due to presence of compliance in the actuator and force sensor and different gains were then determined for different “decoupled” directions to allow larger gains be applied in more stable directions when compared with uniform gain approaches where each actuator is controlled independently using its own force feedback loop. Both simulation and experimental results involving force control of actuators had shown the proposed approach to be capable of improving the force control performance. Additionally, an assistance adaptation scheme was proposed to allow automatic adjustment of the controller inputs and parameters to realise rehabilitation exercises in an assist-as-needed manner. The novelty of this control scheme is that it can limit assistive moments applied in kinematically constrained or stiff environments. This is accomplished through the use of alternative error dependency functions in the feed forward moment adaptation law which limits the growth of the feed forward moment once certain error threshold is exceeded. Furthermore, the assistive moments are also reduced through the implementation of a reference trajectory modification scheme which acts to alter the reference trajectory to reduce the position errors, again when certain error threshold is exceeded. As certain rehabilitation exercises require that a minimal interference be provided by the robot when the user is capable of moving ahead of the existing rehabilitation trajectories, a work based robot stiffness adjustment module was also included in the adaptation scheme to reduce the resistance applied to the user when the movement made is ahead of the reference trajectory. This scheme monitors the negative work done by the robot and acts to reduce the robot stiffness in the direction of reference motion when negative work is detected, and thus has the effect of reducing any resistive forces applied along this direction. A series of simulation case studies and experiments involving both constrained and unconstrained environments had been carried out to test the efficacy of each of the proposed modules and the results have shown that these schemes can deliver their intended effects. The algorithms and models developed in this work were then combined to form an adaptive interaction control framework. Although this framework was implemented on the proposed ankle rehabilitation robot, it is equally applicable to rehabilitation robots targeted at other joints/limbs.