Estimation of Arm Movement from the Neural Activities of the Primary Motor Cortex

We succeeded in reconstructing muscle activities from neural activities and in reconstructing joint angles during static condition. However, the reconstructed joint angles during movement did not have a good correlation with the real joint angles. Therefore, we adopted a servo mechanism to compensate for the error in the joint angles. This error mainly arises due to muscle properties such as the velocity-tension relationship or the length-tension relationship. In order to overcome the limitations due to these muscle properties, we divided an artificial neural network into two networks-one for movement control and one for posture control. We also trained the gating network to switch between the two neural networks. As a result, the gating network switched the modules properly, and the accuracy of the estimated angles improved.