Instrumented finger platform for a systematic investigation on the control of hand exoskeletons with compliant materials

A high number of traumas and neurophysiological disorders each year, along with the aging population, has increased the number of people with compromised hand functionality. The need of providing rehabilitative and assistive devices for patients has encouraged the study in hand exoskeletons. Since rigid structures are bulky and may cause fatigue, soft and compliant components have gained attention from the researchers as they can provide better wearability, improve the portability and help in performing more natural movements. The aim of the project is to develop an instrumented testing platform to allow a systematic investigation into the control of a hand exoskeleton made with compliant materials. The platform is comprised of a 3D-printed finger with angular position sensors, a load cell to measure the endpoint force and a DC motor which controls the exoskeleton through a cable-driven system. The device allows to perturb the finger with a cyclic motion, both in the air and touching the load cell. Furthermore, a Lagrangian model was proposed to map the motor torque to the fingertip force. In order to validate the testing platform, two stages of identification process have been performed to evaluate the missing parameters of the model and then to predict the fingertip force. The first phase of the experiment provides for a repeated sine wave disturbance which moves the finger in the air by means of a single extension wire. Using the acquired data, a simple linear regression is used to estimate the missing parameters of the Lagrangian model: tension moment arm and friction loss. The algorithm is applied by dividing the joints motion in four different regions. During the second phase, the instrumented finger is placed in contact with a compression load cell and a cyclic motion is applied to the flexion cable to allow finger flexion. Implementing the estimated parameters acquired in the first phase, it is possible to evaluate the exerted endpoint force by knowing the cable tension, which is indirectly measured from the motor torque. The preliminary results indicated that our approach to simplify the dynamic equations with the Lagrangian model was not satisfying, with low accuracy in terms of shape (R^2 = 0.36±0.13) and large prediction jumps which are due to state transition. Moreover, a time delay between predicted and measured force has been detected. To further improve the model, parameters estimation has been updated and simplified by using one single couple of parameters for the entire acquisition. To evaluate the time delay, two alternative solutions based on a similarity measure have been proposed. The most promising technique, which minimises the squared Euclidean distance, compensates the delay and allows to have a good fit in terms of shape (R^2 = 0.97±0.01). However, the peak amplitude of the model is around four times lower than the measured one. Considering the complexity of the research field, the limited time available allowed to face the first issues of the system, but it has not been possible to further improve the research and incorporate the actual model for developing an open-loop controller. However, as results are highly reproducible, a better force prediction can be achieved by upgrading the platform with a pretension mechanism for the exoskeleton cable and by improving the parameters estimation.