Concept of a Co-Adaptive Training Environment for Human-Machine Interfaces Based on EMG-Control

Abstract

Learning how to reliably operate HMIs is challenging and time-consuming especially if the underlying control signal is based on biosignals like myoelectric activities. This work describes a concept of a co-adaptive training environment. The human user performs a given task and various features are extracted for performance measurement. Based on single-objective optimization our concept looks for optimal parameters helping the user to improve the performance in the next tasks. Parts of the proposed concept were already implemented.

Model and experiments to optimize co-adaptation in a simplified myoelectric control system

“…To overcome difficulties associated with establishing and maintaining an efficient decoder despite limited neural recordings and changes in signal quality, closed-loop adaptation of the decoder are increasingly studied and used to complement subjects' adaptation in brain machine interface (Iturrate et al, 2015;Dangi et al, 2013;Orsborn et al, 2012;Shenoy and Carmena, 2014). As similar issues occur for myoelectric controls, co-adaptation strategies are starting to be explored there as well (Hahne et al, 2015;Rezazadeh et al, 2012;Tuga et al, 2013). In both situations, however, the dynamics of the concurrent adaptation of the human and machine needs to be carefully considered since it is likely to be complex and to greatly impact the effectiveness of particular co-adaptation settings.…”

“…To overcome these limitations in terms of adaptation capabilities, a strategy is to complement human adaptation by an adaptation of the machine, or decoder. This strategy is currently being explored with increasing success in the context of brain machine interface (Vidaurre et al 2011, Gilja et al 2012, Orsborn et al 2012, Dangi et al 2013, Shenoy and Carmena 2014, Iturrate et al 2015, Shanechi et al 2016, and starts also being developed in the context of myoelectric controls (Rezazadeh et al 2012, Tuga et al 2013, Hahne et al 2015. However, many questions remain as to the particular form and setting of the machine co-adaptation that would best complement the human adaptation.…”