Hidden Markov Model of anticipatory signals for grasp and release from surface electromyography
Hosea Siu, MVL Graduate Student
Surface electromyography (sEMG) is a technique for recording natural muscle activation signals, which may be used as control inputs for exoskeletons and prosthetic devices. Previous experiments have used these signals with both classical and pattern-recognition control methods to actuate such devices. The results of a grasp and release experiment were used to develop an intent-recognition system based on a Hidden Markov Model (HMM) of sEMG data. The system was tested on data from 16 individuals and a mix of shifted sensor positions to check for robustness to sensor misalignments. This study is a novel application of upper-limb movement intent recognition in movements involving object contact – a scenario that is closer to actual use situations for exoskeleton and prosthetic devices than similar previous experiments.
A mean labeling accuracy of 79.22% for cup grasping was achieved through the HMM method. The HMM was also able to detect the presence of pre-shaping motions made by participants, which were unseen in the kinematic labeling of the data. Intra-participant training for feature space dimensionality reduction gives better performance than inter-participant training by nearly a factor of two. HMM labeling of user intent shows potential as a control mechanism for a dynamic grasping task involving user contact with external objects and noise from variations in sensor placement. However, further work is required to improve the accuracy of this system, and to test its performance as part of an upper-limb device controller.