A bio-inspired filtering framework for the EMG-based control of robots

There is a great effort during the last decade towards building control interfaces for robots that are based on signals measured directly from the human body. In particular electromyographic (EMG) signals from skeletal muscles have proved to be very informative regarding human motion. However, this kind of interface demands an accurate decoding technique for the translation of EMG signals to human motion. This paper presents a methodology for estimating human arm motion using EMG signals from muscles of the upper limb, using a decoding method and an additional filtering technique based on a probabilistic model for arm motion. The decoding method can estimate, in real-time, arm motion in 3-dimensional (3D) space using only EMG recordings from 11 muscles of the upper limb. Then, the probabilistic model realized through a Bayesian Network, filters the decoder's result in order to tackle the problem of the uncertainty in the motion estimates. The proposed methodology is assessed through real-time experiments in controlling a remote robot arm in random 3D movements using only EMG signals recorded from ablebodied subjects.