Estimating The Direction Of Force Applied To The Grasped Object Using The Surface Emg

In the passive haptic system and other VR fields, there is a great demand for measuring the direction and magnitude of various forces applied to a grasped object. To acquire this force, there are mainly two kinds of methods; one is directly measuring it by the sensor attached or embedded to the object surface or the glove type device, and the other is estimating it by measuring the change of the surface electromyography (sEMG) of the forearm. The former is used a lot but has problems that it is difficult to stick the sensor due to the object's surface, and sensors prevent fingers and objects from touching directly. The latter approach has a potential to release us from above problems. However, most of their works focused on estimating a certain direction of force, like a grip force, thus, there is no research to estimate multidirectional loading forces.To solve these problems, we propose a solution for estimating direction and magnitude of the force applied to the grasped object using sEMG of the forearm with the convolutional neural network (CNN), as the gesture recognition field uses. By constructing training system that can measure sEMG signals and the force applied to the grasped object, we trained a model, and this model realized over 95% accurate directional estimation and the estimation of force magnitude with less than 7% NRMSE.