Brain-Machine Interface Control via Reinforcement Learning

We investigate the capabilities of reinforcement learning (RL) to create a brain-machine interface (BMI) that uses Q(lambda) learning to find the functional mapping between neural activity and intended behavior. This paradigm shift is intended to address the issue of paralyzed and amputee patients whom are physically unable to move, which is necessary to train traditional supervised learning BMIs. We created a RLBMI architecture incorporating a rat behavioral paradigm for prosthetic arm control. The performance results show 'proof of concept' that RLBMI can learn the temporal structure of neural signals to control a prosthetic arm