Interactive Hierarchical Brain-Computer Interfacing: Uncertainty-Based Interaction between Humans and Robots

Current non-invasive brain-computer interfaces such as those based on electroencephalog-raphy (EEG) [1] suffer from the problem of low signal-to-noise ratio, making fine-grained moment-by-moment control tedious and exhausting for users. To address this problem, we have previously proposed an adaptive hierarchical approach to brain-computer interfacing: users teach the BCI system new skills on-the-fly and these skills are later invoked directly as high-level commands, relieving the user of tedious lower-level control. However, the high-level commands learned from user demonstrations are often not reliable due to incomplete or insuf-ficient data. In this paper, we address the unreliability of such learned high-level commands by proposing an interactive hierarchical BCI. The proposed approach utilizes an uncertainty metric in the learning algorithm to determine whether the learned high-level command is reliable enough to be performed in the present context. The BCI system interacts with the user to make the best decision at each stage. We illustrate the approach using an interactive hierarchical BCI for controlling a simulated wheeled robot. In a study involving two human subjects controlling the robot in a simulated home environment, each subject successfully used the system to complete a sequence of five different navigational tasks. Our results sug-gest that interactive hierarchical BCIs can provide a scalable and robust way of controlling complex robotic devices in real-world environments.