Passive Brain-Computer Interfaces for Robot-Assisted Rehabilitation

Abstract

Stroke patients must exercise intensely with rehabilitation robots to achieve satisfactory rehabilitation outcome, but ensuring appropriate exercise difficulty is a challenging task. Brain-computer interfaces would be suitable for such difficulty adaptations since they capture both conscious and subconscious aspects of workload, but have seen little use in rehabilitation. This chapter reviews previous work on passive brain–computer interfaces and highlights the practical challenges of applying the technology to motor rehabilitation. Preliminary results of a study on workload estimation in a rehabilitation robot with healthy subjects are then presented. Adaptive stepwise regression is used to estimate different types of workload from electroencephalography signals recorded at different sites. Results show that electroencephalography can achieve more accurate workload estimation than autonomic nervous system responses and that adaptive estimation methods can further improve accuracy. However, the number of electrode sites needs to be reduced and issues such as motion artefacts must be resolved before passive brain-computer interfaces can be used in motor rehabilitation.