Motion generation of a wearable hip exoskeleton robot using machine learning-based estimation of ground reaction forces and moments

Statistical data acquired from US citizens in 2013 showed that the overall percentage of all disabilities for all ages in this country was around 12.6%, in which the "ambulatory disabilities" had the highest prevalence rate (7.1 %). This amount is estimated around 7.2% for all Canadian adults, which corresponds to more than 2.5 million people. In order to improve the quality of life of those with ambulatory disabilities (e.g., paraplegic people), wearable robotic exoskeleton is being developed in our lab. In this project, Ground Reaction Forces and Moments (GRF/M), which are important data for closed-loop control of an exoskeleton, is estimated based on lower limb motion of a wearable hip exoskeleton user. This method can reduce manufacturing cost and design complications of these types of robots. In order to model GRF/M, Neural Network, Random Forest and Support Vector Machine algorithms are utilized. Afterward, the achieved results from the three algorithms are compared with each other and some of the most recent similar studies. In the next step, the trained models are employed in an online control loop for assisting a healthy exoskeleton user to walk easier. The device applies forces on the user's upper thigh, which reduces the required torque of the hip flexion-extension joint for the user. Finally, the exoskeleton's performance is compared experimentally with the cases when the device is not powered or it is simply following the user's motion based on the inverse kinematics. The results demonstrate that the presented algorithm can help the exoskeleton user to walk easier.