The conditioning of neural firing for a reliable brain machine interface

Since 1973, biomedical engineers have targeted brain-machine interfaces (BMI) as a means of overcoming severe paralysis in patients by substituting machine function for disabled physical faculties. However, the 60 to 95% accuracy of these BMIs limits their practical application. To fully rehabilitate patients, a BMI must enable unsupervised interaction in the physical world with 100% accuracy. We have developed and demonstrated a high-accuracy BMI by using a novel, unconstrained behavior task. This task calls for alternating periods of time during which it is either appropriate or inappropriate to respond. We first demonstrate that rats can be conditioned to consistently perform this task using levers, making zero errors over as many as seven sequential sessions (over 1700 trials). We then replace responding on one lever with the activation of a BMI neural switch based on intracortical recordings from single neurons. Rats have controlled this BMI interface without making an error for over 32 minutes, approximately 200 trials. This new protocol serves as a test-bed for the development of reliable BMIs that enable real-world interaction.