Motion Generation of a Wearable Hip Exoskeleton Robot Using Machine Learning-Based Estimation of Ground Reaction Forces and Moments

In this paper, motion generation of a wearable hip exoskeleton robot based on ground reaction forces and moments (GRF/M) estimated by machine learning algorithms is presented. Neural network (NN), random forest and regression support vector machine (SVM) methods were employed for training models based on lower limb motion of the exoskeleton user. In order to train the models, ten human subjects participated in generating the training and test datasets. All participants were asked to walk on an instrumented treadmill at different walking speeds to obtain models with a high variation in the walking speed. Knee, ankle and toe joint angles, angular velocities and angular accelerations were considered as input features to the models. Moreover, users' walking speed, weight, and height were used as constant input features in each model. The GRF/M estimation results are provided and the accuracy of the three methods are compared to each other. In addition, the results are compared with previous works from literature which show better accuracy in almost all estimated GRF/M components. As an application of this work, assistive flexion/extension motion generation of the hip exoskeleton based on the proposed NN model-based step detection and control is presented.