Estimating grf using simple biomechanics implemented neural network model

In this study, we proposed implement method between SLIP model that explains the walking dynamics well and a neural network (NN). The proposed method has two major steps. First is biomechanical model step, and second is NN process step. STEP 1. From CLM, estimate two direction of GRFs (F_v,F_ap) from sacral trajectory data. STEP2. The simplest form of the NN estimates the actual ground reaction force from the first estimate. Among 32 steps of treadmill walking data, we used 24 steps for training and 8 steps for validation. As a result, both R^2 value of both GRF(F_v,F_ap) were increased 0.25,0.69, respectly. In conclusion, the biomechanical domain knowledge of human gait was applied to increase the accuracy of estimation without increasing the number of neural networks and learning samples.