Co-adaptive Kalman filtering in a naive rat cortical control task

Control of prosthetic devices is possible via extra-cellular recordings from cortical neurons. Many of the current cortical control paradigms consist of analyzing the relationship between cortical activity and measured arm movements, and then using this known relationship to map cortical activity to similar prosthetic arm movements. However, measured arm movements are not feasible for amputees or patients with mobility limitations hindering their ability to perform such movements. Here we explore an alternative approach using a rat model in which subjects learn prosthesis control via an adaptive decoding filter that adjusts to the modulation patterns recorded from neurons in the motor cortex. Our methodology takes into account the ability of a subject to learn an effective response strategy in conjunction with online filter adaptation. A modified Kalman filter is demonstrated to "co-adapt" by training on past periods of significant modulation during expected prosthetic device movement. Feedback pertinent to completing the cortical task is given to aid the animal in adopting a response strategy maximizing reward. One subject was able to perform the task consistently above chance after 2 days (4 sessions) of training.