Chapter Three – EMG-Controlled Human-Robot Interfaces: A Hybrid Motion and Task Modeling Approach

This chapter describes a hybrid motion and task modeling framework for human-robot interfaces based on electromyogram (EMG) signals. The framework incorporates contextual information of an ongoing task, in the form of a task model, into a traditional EMG pattern recognition method. In this chapter, the basic concepts of EMG motion classification and their application in human-robot interfaces are presented first with a brief summary of the techniques that are currently available. Robustness and reliability of motion classification are key features for prosthetic devices and human-assisting manipulators controlled using EMG signals, especially when practical applications are considered. Task modeling is then introduced to improve the motion classification performance since it can provide a prediction or a reference of the user's future behavior. Many studies have been conducted to demonstrate how task models can support classification and prediction in the field of human interfaces. Two methods are briefly presented in this chapter. A case study of the EMG-controlled human-robot interface using task modeling is introduced in the rest of the chapter. A prosthetic hand is controlled with motion commands classified from forearm EMG signals. Motion classification is achieved based on both a probabilistic neural network and a task model using a Bayesian network for motion prediction. Experimental results of robot manipulation indicate that the hybrid EMG motion classification framework outperforms traditional methods used for comparison.