A musculoskeletal model-based Assistance-As-Needed paradigm for assistive robotics

University of Technology, Sydney. Faculty of Engineering and Information Technology.Robotic systems which operate collaboratively with their human operators to provide assistance are becoming reality, and many different paradigms for administering this assistance have been developed. A promising paradigm is Assistance-As-Needed, which aims to provide physical assistance specific to the individual requirements of the operator. This requires that the needs of the operator be determined, which is challenging as they depend on both the task being performed, and the capability of the operator to perform it. Current solutions use performance-based methods which critique the operator from observations obtained during tasks, and then adapt assistance based on how they performed. This approach has shown success in applications such as robotic rehabilitation. However, empirical performance-based methods have inherent limitations, primarily due to the numerous observations required before the operator’s assistance needs can be determined. The ideal Assistance-As-Needed paradigm should be able to determine the operator’s assistance requirements without prior observations, and with respect to arbitrary tasks. This thesis presents a novel Assistance-As-Needed paradigm using models to estimate the assistance needs of the human operator. An optimisation model is developed which utilises a publicly available musculoskeletal model representing the human upper limb to estimate their strength, which is compared to the strength required by the task being performed to gauge their assistance requirements. An advantage of this model-based approach is it allows effects on the operator’s assistance requirements due to task and physiological factors to be predicted. Furthermore, it avoids many of the limitations faced by empirical performance-based approaches since it does not require empirical observations. The model-based paradigm is demonstrated and evaluated in a number of simulated tasks involving the upper limb. Calculated upper limb strength is analysed with respect to factors such as the limb position, the direction of force at the hand, and muscular impairment. The calculated strength is shown to predict behaviours similar to those described in the literature. Experimental evaluation is performed by implementing the paradigm on a specially developed robotic exoskeleton to govern the assistance it provides a subject in a number of experimental tasks. The model-based Assistance-As-Needed paradigm is shown to successfully govern assistance towards specific muscles when needed in the tasks performed. Means of improving the paradigm, including methods for fitting the model to the subject, and the inclusion of additional physiological factors in the calculation of their assistance requirements is discussed