Abstract
Recently, BCI(Brain-computer interface) and BMI(Brain-machine interface) come into the research limelight. However, nonsynchronous spontaneous action potentials and evoked action potentials exist contain brain signal, and we need an interface model between brain and machine for control and stability. We have already proposed collaborative learning system consisting of reinforcement learning and brain signal. Brain signal is interpreted as a deliberate assignment of the subject, and we utilize reinforcement learning in control and stability for BCI. In this paper, we discuss the usefulness of collaborative learning for BCI using reinforcement learning. We first design the collaborative learning system with near-infrared spectroscopy (NIRS), and apply it to maze problem. In addition, we discuss the comprehensive evaluation of collaborative learning system in terms of difficulty of the problem, precision of the problem and the mental load to the subject, and show the usefulness of the proposed system.
