Brain–Machine Interfaces

Abstract

Motor brain–machine interfaces (BMIs) allow subjects to control external devices by modulating their neural activity. BMIs record motor cortical activities, use a decoding algorithm to infer the subject’s intended movement and control a prosthetic device, and provide visual feedback to the subject. Thus BMIs can be viewed as closed-loop control systems. In this chapter, we review the computational components of a BMI and the common decoders used in the field. We then discuss in detail a recent control-theoretic high-rate BMI decoder, termed adaptive optimal feedback-controlled point process filter (OFC-PPF), which has significantly improved performance and robustness. This decoder characterizes the spikes directly using a point process model and learns the model parameters using closed-loop decoder adaptation. The decoder also models the BMI as an optimal feedback-control system to better infer the brain’s intention during adaptation. This decoder significantly improves the speed and accuracy of model adaptation. Moreover, at steady state, the learned point process filter improves performance over the state-of-the-art Kalman filters due to the fast control and feedback rates and the point process encoding model.