Developing a glove type sensing system with accurate estimation of the thumb posture

Abstract

As the aging of the society progresses and medical treatment improves, the population of post-stroke survivors and spinal cord injury patients increases. In order to enhance the life quality of these patients, rehabilitation of their impaired limbs is essential. Especially the rehabilitation of the distal hand is important, as the recovery of the distal hand is known to be the most slow and limited. Recently, many compact wearable hand robotic devices have been developed for rehabilitation and assistance of the hand. Developing a compact sensing system that could correctly measure the hand configuration is important in these devices to control the joint coordinates with natural grasp trajectories, and evaluate the functionality of the hand quantitatively without losing the reachability and range of motion of the hand. Bending sensors enable compact design of glove-type hand robotic device, but the measurement of the thumb configuration is challenging due to the multi-degree of freedom movement, especially for the carpometacarpal(CMC) joint of the thumb. Also there are no internationally accepted single definition of the CMC joint coordinate system in hand rehabilitation and biomechanics. A bending sensor attached on any hand location tends to be deformed by movement of thumb in all DOFs which means that it always involves coupling effect. Considering that most locations at hand involve coupling effect, it is important to find sensor locations that have minimal coupling effect to make the thumb joint angle measurement more accurate. Utilizing the appropriate definition for thumb joint angle estimation is also im-portant to correctly reconstruct the thumb joint motion. In this dissertation, the thumb joint kinematics are investigated and the definition that could correctly estimate the thumb joint motion is suggested. Also a methodology to find optimal positions of the thumb CMC joint abduction sensor and flexion sensor is introduced, based on analysis of the deformation of the hand with 3D scanned images of the hand surface. The optimal positions of the sensors, derived by the suggested methodology, are evaluated by experimenting the bending sensors on those positions and comparing the estimated joint angles with respect to the reference CMC joint angles measured by 3D motion capture system. Brief calibration of the sensor signal into joint angle was done with the evaluation data, and the results were compared with the actual CMC joint angle. For 6 tested subjects the root-mean-square error of the calibrated sensors were 2.7 deg for the abduction sensor and 2.4 deg for the flexion sensor.