RESUMO
In myo-controlled prosthetic hands, surface electromyographic signals are used to operate the hand actuators. A pre-requisite for effective control is that the intended movement is decoded from muscle activity. Simpler approaches use pattern recognition techniques, which assume a finite set of possible actions. However, this leads to unnatural, discontinuous control. Proportional controllers do not require a finite set of actions to be specified in advance but are difficult to use, particularly with dexterous multi-fingered hands. Here we discuss a control module which continuously predicts the intended movements from recorded multi-channel electromyographic activity. The module can be seen as a (simplified) forward model of the dynamics of the intact hand. We describe a procedure for estimating model parameters from hand movement and muscle activity data, and discuss its application to the proportional myoelectric control of a prosthetic hand.
