Implementation of Neural Network Control for Foot Prosthesis as Foot Function Reconstruction in Post-Amputation Patients

Abstract

—The purpose of this paper is to show the robot’s bionic legs can be moved automatically by applying a deep learning neural network as a control system to improve the function of the bionic legs. The deep learning neural network control system resembles the nerve network in the legs, so it requires a dataset of thigh muscle strength variations, and knee joint angles during the process of walking, going up and down stairs. This dataset is used in the design using the Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM) model through a training process so that an optimal model is obtained using the Tensor flow API to be implemented into the prosthetic leg system. Deep learning control systems require a lot of data for model training, so this study uses a combination of sensors, namely the FSR402 sensor and the MPU sensor. By using a control system based on RNN-LSTM the performance of the robot’s leg movements is better and has a very small error.