In this talk I will present results obtained in the ESPRIT project Adaptive Brain Interfaces. It concerns the development of a wearable brain-computer interface; that is, the EEG-based recognition of mental states that are associated with simple commands. Our approach seeks to develop individual interfaces since no two people are the same either physiologically or psychologically. Thus the interface adapts to its owner as its neural classifier learns user-specific filters.

We have developed individual interfaces for 8 healthy persons. Our current prototype robustly recognizes three mental states from on-line spontaneous EEG signals. The percentage of correct recognition--true positives--is not very high (around 70%), but:

* our neural classifier is embedded in a system that makes decisions every 1/2 second;

* the percentage of wrong responses--false positive--is below 5%; this is extremely important from a practical point of view.

The fact that the subject and his personal interface learn simultaneously from each other allows subjects to master the system in a very short time: one of the subjects who did not have any previous experience with brain interfaces achieved excellent control in just 5 days of training. I will show a video reporting this case.

Finally, I will also discuss how such an interface can be used by motor-disabled persons to control a wheelchair.